A few weeks ago I decided to start a company. I expected the long hours but I didn’t expect this.

I’ll say more about what we’re building soon. For now, know that it lives at the intersection of two things I’ve spent my life circling: AI, and making good decisions and predictions.

I want to be honest about why I am doing this. Not ambition, or not only that. It was the plain arithmetic of one life: a finite number of years where I’d still have the energy and the freedom to build something of my own. I found so much curiosity and joy in what I’d discovered that not pursuing it felt like the real risk. The cost of trying was failure. The cost of not trying was a regret I believe I would carry for a long time, or not, we’ll see.

So why the knot?

Subscribe now

The model in your head

As it turns out, the brain is a prediction machine. It builds a model of the world and checks it against what actually happens, thousands of times a day. When the model matches reality, you feel calm and capable, even when you’re busy. That calm doesn’t come from having nothing to do. It comes from knowing what to expect.

For eighteen years in Big Tech I could read my world. I knew what a good week felt like, who to call when something broke, how a launch was supposed to go. I knew my way around. In the past year, I had cleared most of my schedule; most days were consistently around a few themes: family, health, tinkering, and rest.

A startup wrecks that model.

Most days, what happens doesn’t match what I expected, because I don’t have good expectations yet. What’s the right price? Who’s the right first customer? What does an investor’s silence mean? Was today a good day or a bad one? I can’t reliably call any of it. Every surprise trips a small alarm: something’s wrong, fix it now. One alarm is fine. It’s a small amount of stress and is part of learning. But stack these alarms day after day and the alarm never resets. You start expecting threats everywhere, and anything unclear starts to feel like danger.

That was the whole of it. The stress didn’t mean I was bad at this. It meant I was somewhere new, where I hadn’t learned what to expect yet. Of course nothing matched. I was still learning the place.

Two traps for founders

The first is comparison. Open any feed and you’ll find a wall of other people’s funding announcements, traction charts, and accelerator wins. Your brain was built to compare you against a village, a few dozen people you actually knew. It was never built to measure you against a global highlight reel that refreshes every time you scroll. So you lose, over and over, against a sample designed to make everyone feel behind. I went deep on this in You’re comparing yourself to people who don’t exist.

The second is identity. At my old job I was “the director who worked on X.” I gave that up on purpose, and now I’m starting over. On day one of a company, you’re no one in particular. The label that used to answer who am I is gone, and losing it stings more than it should, because we’re wired to feel a threat to who we are almost like a threat to the body. There’s nothing weak about that. And the irony isn’t lost on me. I spent a year telling people that life after a career is the place to be, and here I am, reaching for a new title to stand on. Even if it’s one I don’t particularly crave.

What’s helped

None of this is a reason to go back. It’s a reason to work on myself while I work on the company. A few things have helped.

Treat stress as information, not a verdict. When what happens doesn’t match what I expected, I can dig in, get annoyed, and insist the world is wrong. Or I can change what I expect. The first feels better for a minute and costs me later, because I can’t learn anything while I’m busy defending myself. So now I try to ask a better question. Not what’s wrong with me but what is this teaching me that I didn’t know? That turns the stress into information instead of a judgment.

Don’t let the company become your whole identity. When one role carries all of who you are, every bad week in that role feels like the end of the world. So I keep more of myself in view, not less. I’m a son. I’m a father. I’m a runner chasing a sub-20 5K with the stubbornness of someone who’s finished a 50K. I’ve had a meditation practice longer than I’ve had a career. When the company has a rough week, and it will, those parts of me don’t move.

Don’t tie your worth to the scoreboard. Caring about the work is good, and I want to care a lot. But when my worth rises and falls with the numbers, every dip feels like proof I’m not enough. So I try to care about the work and hold the outcome loosely at the same time. Part of that is learning to lose well. You can’t win every week. When something fails, I give it room, get curious about what went wrong, and try to learn from it without turning it into a story about my value as a person.

When you’re really spiraling, stop trying to think your way out. Some days the rut is too deep, and no amount of clever reasoning gets me out, because the problem isn’t in my head, it’s in my body. On those days I move. A hard run calms me down faster than any pep talk. A long stretch outside, away from screens, does in two hours what a week of willpower can’t. And a quiet sit with my breath puts me back in my body when my mind won’t let go.

But knowing isn’t doing

The four points leave something out. I fail at them, often.

I’ll tell you to treat stress as information, then spend an evening revisiting an investor reply for information which might not even be there. I’ll tell you not to watch the scoreboard, then open the dashboard before I’ve said good morning to my kids. I’ll tell you to move instead of spiral, then scroll for half an hour looking through the code to make sure the idea is still sound.

Knowing the mechanism doesn’t switch it off. The wiring doesn’t go quiet just because you’ve named it.

So this isn’t a thirty-day fix. It’s a practice I keep coming back to. The goal was never to feel no stress. It’s to catch it sooner, hold it more lightly, and be a little kinder to myself while I’m in it.

Still learning the place

I still get stressed. I expect I will for a while, because I’m early in learning a world I picked precisely because I didn’t know it yet.

But I don’t read the stress as proof I made a mistake. It’s what it feels like to be somewhere new, before you’ve found your footing, which is exactly where I wanted to be when I left. The work is hard, and I want it to be. The stress was never the price of the work. It was the price of learning to navigate a life I hadn’t lived yet.

I’m still learning. That’s exactly what I left to do.

What follows is that letter, lightly edited. None of it is clever. Life and financial best practices like: don’t carry a credit card balance, freeze your credit, sleep, call me when things go wrong. The boring stuff is the stuff that works, and most of us learn it too late.

I’m publishing it in case you’re sending someone off too, or in case you’re the one leaving. Take what’s useful. Skip the rest.

Subscribe now

Adulting 101

Before any of the rest of it: I am proud of you, and you are more ready for this than you feel. You will get things wrong. That is the point of going. Call me when you want to, not because you have to.

A few principles I hope you carry with you. You do not need to master all of this at once. The goal is to build a good base, avoid big mistakes, and give yourself more freedom over time.

1. Mindset for Life

Your mindset will matter more than almost anything else as life unfolds.

Do not let one mistake become your identity.
You will mess up. Learn the lesson, pay the tuition, move forward. A mistake once is experience. The same mistake again and again is a problem to solve.

Take calculated risks.
A safe life is not the same as a good life. Do not avoid change just because it is uncomfortable.

Do not expect instant success
Nobody owes you anything. Earn trust, build skill, do the work. Compounding works in money, skills, health, and relationships. None of it shows up fast. Most good things take several attempts. Early failure is not proof you cannot do it.

The rest, in no particular order:

• Do not dwell on the past. Feel it, learn from it, then move. What is the point of carrying it?

• Avoid self-pity. Bad things happen to everyone. Feel them, then move.

• Do not let jealousy run you. Someone will always be richer, smarter, luckier, or further ahead. Their success does not shrink your path.

• Do not obsess over what you cannot control. Put your energy where your actions matter.

• Do not try to please everyone. It is impossible and it will make you doubt yourself. Be kind, but have a spine.

• Do not give others power over you. Your worth does not depend on anyone’s approval.

• Do not avoid solitude and silence. You need quiet to hear yourself think.

2. Take Care of Yourself First

You are about to be fully in charge of your own body and mind for the first time.

Sleep is not optional.
Pulling all-nighters feels productive and is not. Lack of sleep wrecks your mood, your memory, and your judgment.

Move and eat like someone who plans to be around a while.
You do not need to be perfect. Walk, get some sun, do not live on energy drinks and vending machines.

Asking for help is a skill, not a weakness.
Office hours, your RA, your advisor, daddy and mommy. You don’t have to white-knuckle a problem for months rather than ask a question. Be the one who asks.

Drink slowly or not at all, and never lose track of your own count.
Don’t drink would be my advice but if you must, be aware. Do not take a drink you did not watch get poured, and do not leave one unattended.

Never get in a car with a drunk driver, and never be one.
Walk, call a ride, sleep on the floor. There is no version of this where saving some money is worth the risk.

Important Reminder
If you are ever in a place where you feel unsafe with your own thoughts, you can go to the campus counseling center or a doctor. You can always call me or your mother, any hour.

3. Protect Yourself From Scams and Theft

Before trying to get rich, avoid getting wiped out.

If it sounds too good to be true, stop.
Pause. Do not rush. Scammers create urgency because clear thinking is their enemy.

Credit freezes
Freeze your credit at the major credit bureaus. It helps prevent identity theft and fraudulent accounts.

Virtual credit cards
Use virtual cards when possible, especially for online subscriptions or merchants you do not fully trust.

Be careful with money and friends
Trust people, but be safe. Money can damage relationships. Be especially careful with loans, investments, shared expenses, and vague promises.

4. Money, Now

Debt is not always evil, but it is dangerous when used casually.

No credit card balances
Never carry credit card debt. Pay the full balance every month. Credit cards are tools, not loans.

Avoid debt unless it has a clear purpose
Debt reduces freedom. The fewer fixed obligations you have, the more choices you have.

Spend less than you earn, even when “earn” is small.
The habit matters more than the amount right now. How much you save will matter more than how cleverly you invest for a long time.

Keep a cash buffer.
Enough to cover a surprise so you are never forced into a bad decision. Build toward a few months of expenses over time. Any amount beats zero.

When you do invest, keep it brutally simple.
Broad, low-cost index funds (VTI or VOO for now). Boring is good. You own the whole market instead of trying to outsmart it. Use a low-cost platform like Vanguard, not a bank. Banks are good for banking and will happily sell you high-fee products that are not in your interest.

5. Money Later (Read This When You Have a Real Salary)

None of this applies until you have a steady income, so do not worry about it yet. When you land a real job, send me a message and we will set it up together in about an hour. Rough order:

1. 401(k) up to the employer match. Free money, take it first.

2. HSA, if you have a high-deductible health plan.

3. Backdoor Roth IRA, if your income is too high to contribute directly.

4. Max the 401(k).

5. Mega-backdoor Roth, if your employer plan supports it.

6. Taxable brokerage for anything beyond that.

And when life gets bigger (a partner, kids, a house, real assets), term life insurance if people depend on you (simple term, never whole life sold as an “investment”), disability and umbrella insurance, a will, and basic estate planning. Not now. Just know it exists and we can talk through them later.

6. The Big Picture

Build a life with options.

• Choose good people. Be a good person.

• Stay humble. Keep learning.

• Do not let one mistake become your identity.

• Spend less than you earn. Avoid stupid debt.

• Invest simply and consistently.

• Protect against the few catastrophic risks.

The goal is not to get obsessed with money. The goal is to make money boring enough that you can focus on the people, the work, and the life that actually matter.

Five weeks ago I ran my second parkrun in 23:47 with a 2-second positive split, and wrote a whole post about pacing finally clicking. Today I ran my third parkrun in 23:20 with a 32-second positive split.

I am excited about the PB, but I am disappointed about the fade and massive positive split.

Subscribe now

The race, in two halves

The first two miles were as clean a race execution as I’ve ever put together. Mile 1 at 4:30/km, mile 2 at 4:31/km. Cadence 180+, stride length 1200mm, vertical ratio 7%. These are essentially my best-ever interval biomechanics, sustained for over 3.2km of continuous racing. If I had been able to keep the pace through 5km I would have ran 5K at 22:30. Nice.

Then mile 3 happened, and every biomechanical marker collapsed in lockstep. Cadence fell 6 spm. Stride length dropped 20mm. Ground contact time stretched 12ms. Vertical ratio jumped from 7% to 7.44%. All of this is to say that my running form just went south and I slowed down a lot.

It turns out that the cause was at km 2. 4:27/km in the second kilometer of a 5K is above my lactate threshold and my average HR jumped 13 bpm between mile 1 and mile 2. I didn’t realize that I was burning candles and by mile 3 I was anaerobic, and there’s no recovery from that inside a 5K. I simply couldn’t slow down enough to clear lactate without slowing down too much to finish.

I knew it was happening at the time. I felt the legs flatten at 2.5km and watched the pace drift up despite holding effort. I still tried valiantly and my max HR in the race was 186, the highest I’ve ever recorded.

What got me here: the May 13 breakthrough

The training between parkruns was four sessions of 6 × 1000+m repeats on the track (in the outer lanes).

Apr 29: 4:27/km average, 24-second spread. A front-loaded blow-up. I opened rep 1 at 4:13/km, the fastest single rep of the entire arc, then faded to 4:36 by rep 6. A new ceiling but no sustainability.

May 6: 4:28/km average, 9-second spread. Controlled negative split. Rep 6 was the fastest rep of the day.

May 13: 4:22/km average, 2-second spread. Six reps between 4:21 and 4:23. The cleanest, fastest workout of the whole arc. The session that suggested 22:30 was on the table.

May 21: 4:26/km average, 6-second spread. An off day with heavy legs, ahead of taper. The body asking for a recovery week, which it got.

The middle two weeks, May 13 in particular, are why this race had 22:30 legs.

The tune-up that predicted more

Three days before the race, I ran a 2.4km tune-up. Same IPPT distance I run before every race now, partly because it gives me a clean year-over-year comparison.

May 27 tune-up: 10:38 at 4:25/km average. 26 seconds faster over the same distance than the tune-up before parkrun #2. Twelve seconds per km faster. Stride length up 58mm from April. Vertical economy holding in the same band. Average HR 164, max 172, well under race-day cardiac ceiling.

The 4-sec/km gap from tune-up to race held last time. April 22 tune-up at 4:37/km predicted a 4:41/km race, which is exactly what happened. If the same conversion had held today, I would have raced at 4:29/km. That’s 22:25.

The race the tune-up predicted didn’t show up. The race I ran is the race I paced.

Same fade, fitter legs

Three races on the same course. Three PBs. The fitness curve is monotonic. I’m 84 seconds faster over the same trail than I was on March 21, and roughly a percentage point higher on age grade.

The pacing curve is not monotonic. Parkrun #2 was the cleanest race of the three in this aspect. Today I opened more aggressively than parkrun #2 (4:32 vs 4:44) and produced essentially parkrun #1’s positive split (32 seconds vs 36). The shape of today’s race is the shape of my first ever parkrun, not my second. The discipline I taught myself in March didn’t carry through to today, mostly because the fitness underneath it changed and I didn’t recalibrate.

Pacing isn’t permanent

Pacing is a skill you re-execute every race, not a virtue you accumulate. The right opening pace at 23:47 fitness (4:44/km) is different from the right opening pace at 22:30 fitness (somewhere around 4:34/km). The shape of the discipline is the same. The numbers shift. And the calibration has to happen fresh every time you toe a line, because you only roughly know what your current fitness is.

The 4:30/km mile 1 today felt conservative. It felt the way 4:44/km would have felt in April. Against the 22:30 fitness I’d built, it was neutral, and 4:27 in km 2 was over the line. I was driving a faster car at the same throttle settings I’d learned on the slower one, and I overshot the corner.

The fix is small in mathematical terms and meaningful in mental terms. Open two to three seconds per km slower. Trust the fitness is there. Let the kick at the end produce the negative split that the discipline is supposed to enable.

I will write this down again in the next race plan and try harder to mean it.

Sub-20, still eventually

Sub-22:30 is the next gate. The fitness for it is already locked in. Today’s first half was 11:14, sub-22:30 pace held for 2.5km. The job is repeating it for another 2.5km, which means opening at 4:33-4:35/km and trusting there’s more in the second half than I expect.

Sub-22 is probably one more training block beyond that. Sub-21 is the gate that asks for a different kind of work. More weekly mileage, real strides, proper plyometrics, possibly some hill repeats. Sub-20 is several gates beyond, and still a long way off.

I wrote in the last post that the climb is patient, whether I am or not. That hasn’t changed. What’s changed is that today gave me a sharper picture of which gates need fitness and which gates need discipline. Sub-22:30 doesn’t need more training. It needs me to execute a plan I already know how to write down. Sub-21 will need both, and a lot more of one of them.

The honest version of where I am: I have 22:30 legs and 23:20 pacing. The fitness curve is doing its job. The pacing curve is mine to close, and it has to be closed in real time, while the race is happening, against a body that always wants to go faster than it should in the second kilometer.

You train for what you want to do. Today I trained for sub-22:30 and ran a 23:20 PB. Next time, I want to be patient enough to let the fitness do its job.

Sub-20 stays the north star. The slope today was steep in the wrong direction for one mile, and it cost me 30-45 seconds. The direction over the longer arc is still right. The process is still working. The pace will keep moving as long as the work keeps showing up, and as long as I keep relearning, race by race, what controlled means.

When I was younger, most of what I pictured about the future came from someone else. Back to the Future shaped how I saw it. The Matrix and Minority Report shaped it differently.

These days, influencers on social media bombard us with predictions of the future. But you didn’t reason your way to those possibilities. You absorbed them, and now you quietly repeat them as if they were your own conclusions.

That mattered less when change was slow. It matters more now. Whatever you believe about AI timelines, the rate of change has gone up, and borrowed predictions age fast. I’m not in the singularity-is-imminent camp. But I’ve stopped trying to find the one person who called it right and reading them like scripture. The more useful move is to get better at the thinking itself, so you can run your own scenarios instead of relying on other people’s.

Here’s the framework I am using.

Four lenses: what’s possible, what should be, what’s likely, and what to avoid

Almost every claim about the future is one of four things. Naming which one you’re looking at is useful, because each one works and fails in predictable ways.

What’s possible: What the technology might someday enable if today’s limits disappeared. This is the type of prediction that runs science fiction, product launches, and most of what gets called “the future” in public. It’s exciting and it spreads, which is exactly the problem. These futures show you the extreme version of everything with no path to get there, and they collapse the moment you check the details. There is usually another issue: the primary beneficiary is an advanced user, such as a surgeon doing something miraculous with the headset, never the ordinary person who buys it to play games. Many of these predictions should be treated as entertainment until proven otherwise.

What should be: Anyone who tells you how the future “ought to” look, backed by numbers and moral weight, is doing this. It’s a strong motivator. A clear, desirable vision pulls people toward it and sometimes bends reality into making it true; we describe paying with credit before the system exists, then go build it. The weakness is that it’s self-centered by design. Everyone’s preferred future is incompatible with someone else’s, and people present their own as inevitable. The skill is spotting when a should be is dressed up as a neutral forecast. “This is where things are heading” often means “this is where I want things to head, and I’d like you to help.”

What’s likely: These predictions are the most honest. They admit the future is uncertain and hold several possibilities at once instead of betting everything on one. They give you tools: scenario sets, base rates, the futures cone with probable outcomes near the center and the tail at the edge. The catch is that a model is only as good as its inputs and its hidden assumptions. A tidy 2x2 of four scenarios feels rigorous and can quietly delete the ninety other futures that didn’t fit the grid. Use the tools, but keep asking what the tool is leaving out.

What to avoid: Cautionary futures, unintended consequences, the failure mode of a well-meaning plan. This one is underrated. Fear is a strong teacher, and a good dystopia changes behavior in a way a roadmap never will. Its failure is the reverse of the what’s possible problem: spend long enough hunting for what could go wrong and you see only that, which makes you as blind as the optimist. A warning with no alternative attached is just despair with footnotes. Pair every what to avoid with a then what.

The practical use isn’t to pick a favorite. It’s to label the next future-claim you encounter, including your own, and ask which lens it’s stuck in. Most predictions are a what’s possible dressed in flowery language, or a what should be pretending to be a what’s likely.

Subscribe now

It’s almost never the extremes

The most useful lesson I’ve taken from all this: stop spending attention on the tails.

Both the jetpack future and the collapse future are seductive and roughly useless for planning, because most of life isn’t either. Most of life is unremarkable. People go to work, eat, sleep, deal with the thing that broke, and enjoy the occasional good day. A realistic picture of the future has to survive contact with that ordinariness. The sharpest exercise I know is to pick something mundane and play it forward. What does a grocery store look like in 20 years? What’s on the shelves, what does it cost, what’s the floor plan, what rules does it follow? You’ll learn more from that than from another story about colonizing Mars.

Three errors keep dragging predictions toward the extremes, and they’re worth naming so you can catch yourself:

• Treating the future as a replacement instead of a layer. New things rarely wipe the slate. They get bolted onto the existing tangle of laws, infrastructure, and habits, and that tangle shapes them as much as they shape it. Anything you can’t get rid of in a decade is a thing you have to plan around, not plan past.

• Assuming things won’t break. The honest future has glitches in it. Products get misused, systems fail, people do the unintended thing. A future with no friction in it is a marketing asset, not a forecast.

• Mistaking a story for data. Hard numbers tell you what happened. The moment you point them forward they turn into a narrative wearing the numbers as a disguise, and that narrative gets less reliable the further out you push it. Confidence in a forecast is not evidence for it.

The impact on society runs through the average person’s average day, not through the exceptional case. The change that matters is usually boring. It’s a tweak to a tax rule, a default setting, a price that drops below a threshold, a form that gets shorter. None of that makes it a blockbuster movie or bestseller book. All of it moves the median Tuesday, and the median Tuesday, multiplied by a few hundred million people, is what “society changed” actually means. When you evaluate a future, ask what it does to the unremarkable middle, not what it does to the hero in the demo.

What the AI debate looks like through these lenses

The loudest future-argument right now is what AI does to us in the next decade, and it’s mostly two camps shouting past each other.

On one side, the doomers. Eliezer Yudkowsky’s 2023 Time piece argues we need to “shut it all down”. The Center for AI Safety’s one-sentence statement, signed by Hinton, Bengio, Altman, Amodei, and Hassabis, places extinction risk from AI alongside pandemics and nuclear war. Daniel Kokotajlo’s AI 2027 scenario lays out a step-by-step path to superintelligence and possible human takeover by the end of the decade. In September 2025, Yudkowsky and Nate Soares published If Anyone Builds It, Everyone Dies, now a New York Times bestseller, whose thesis is exactly what the title says.

On the other side, the accelerationists. Marc Andreessen’s 5,200-word Techno-Optimist Manifesto declares stagnation the real enemy and treats technological acceleration as a moral duty. The e/acc movement organized around it. Leopold Aschenbrenner’s Situational Awareness essay forecasts AGI by 2027 and a US-China arms race over trillion-dollar compute clusters. Anthropic CEO Dario Amodei, who also signed the CAIS statement above, published Machines of Loving Grace in October 2024, a 15,000-word essay arguing the upside of AI is radically underestimated and could compress a century of biological progress into five to ten years.

Run all of this through the lenses and the same trick shows up.

The doomer case is fundamentally a what to avoid. Extinction is the worst outcome anyone can name, and the work of naming failure modes and funding safety research is exactly what what to avoid is supposed to do. The strongest doomer pieces don’t present the bad outcome as one possibility among many. They present it as a what’s likely. AI 2027 reads as a forecast, not a warning. If Anyone Builds It, Everyone Dies puts the inevitability in the title. The value-claim, that extinction would be catastrophic, is sincere and probably right. The probability-claim is painting this as urgent when it’s still one possible future among many. If you remember Y2K, plenty of people thought planes were going to fall out of the sky.

The accelerationist case runs the same pattern in reverse. At its core it’s a what should be: a conviction that progress is good, abundance is the goal, and the ideologies that brake technology are the real enemy. The Techno-Optimist Manifesto is openly a values document; it ends with an enumerated list of enemies. Machines of Loving Grace is upfront about its frame from the first page: a sketch of what a world with powerful AI might look like if everything goes right. But these visions get repeated and recast as what’s likely, with intelligence takeoff and material abundance presented as the path we are already on. Aschenbrenner’s 165 pages make a vivid story of inevitability, and the essay was successful enough to be spun into a $1.5 billion hedge fund. A should be turned into an investment thesis is about as on-the-nose as the framework gets.

The most interesting cases are when a single actor runs both moves at once. The same company that published Machines of Loving Grace in 2024 announced in April 2026 that its most capable model, Claude Mythos, was too dangerous to release to the public and would go only to a hand-picked group of about forty corporate customers under a program called Project Glasswing. The framing is what to avoid (this model could exploit critical infrastructure). The practical effect is what should be (Anthropic’s products are this powerful, so use them). Critics in the mainstream press noticed and called it safety theater deployed as marketing. When the safety claim doubles as the sales pitch, picking the right lens stops being academic.

The same problem shows up at the what’s possible end. In October 2025, OpenAI publicly claimed GPT-5 had cracked ten previously open Erdős problems. The mathematician Thomas Bloom, who maintains the canonical Erdős problems list, looked at the work and called the framing a dramatic misrepresentation; the model had retrieved existing solutions from the literature, not produced original proofs. The executive who posted the claim eventually left the company. Then on May 20, 2026, OpenAI announced a genuine result: an internal reasoning model disproved an Erdős conjecture from 1946 about how many pairs of points on a plane can sit at exactly unit distance from each other. This time the proof came with external verification from Bloom himself, Noga Alon, and Melanie Wood. The contrast is the lesson. The first announcement was a what’s possible claim dressed as a done-deed and collapsed in days. The second was a real result that survived the same scrutiny. The lenses, plus the social process of verification, are the difference between the two.

What’s missing from most public AI talk is the what’s likely in the boring sense. The view that AI is closer to electricity or the internet than to a new species, transformative over decades, diffused unevenly, mostly experienced as faster spreadsheets and better customer service and gradual shifts in which jobs exist, gets less airtime because it doesn’t make a thumbnail. Arvind Narayanan and Sayash Kapoor have argued exactly this in their Princeton-Knight AI as Normal Technology paper, explicitly positioning the view as the alternative to both utopian and dystopian framings that treat AI as a separate kind of entity. Even the careful doomers are quietly updating in this direction. Kokotajlo, whose AI 2027 went viral on that title, has since revised his median AGI estimate to around 2030 as the underlying capability curves came in slower than expected. That’s the move you want to see from anyone making predictions, and it’s mostly absent from the loudest voices on both sides.

The lenses don’t tell you which camp is right. They tell you what each camp is doing, which is most of what you need to stop being shoved around by them.

Now point all of this at your own life

The same four lenses work on the only future you steer: your own.

Run them on your next decade instead of the world’s:

• What’s possible: What’s genuinely open to you? Most people underestimate this one because they only ever saw the demo-reel version of other people’s lives. Imagine it for yourself.

• What should be: Whose voice is the “should” in your head? A parent, a peer group, an algorithm optimizing for your attention? A should be you didn’t choose is the most expensive thing you can carry, because it spends your years on someone else’s vision.

• What’s likely: What’s the honest base rate, given your real trajectory? If you keep doing exactly what you’re doing, where does that land?

• What to avoid: What’s the future you’re quietly drifting toward that you’d regret, and what’s the early signal you’d see first? Name it now, while you can still cheaply change course, and attach a then what so it’s a plan and not just a fear.

Then apply the boring-Tuesday test to yourself. Don’t picture the promotion or the finish line. Picture an average Tuesday five years out: what you do in the morning, who you spend the day with, what you’re tired of, what you look forward to. If that ordinary day is good, the plan is good. If you can only make the case with a highlight reel, the plan is a what’s possible in disguise.

Pick one real decision you’re sitting on. Run it through the four lenses; write a sentence for each. Put a probability on the two outcomes you’re worried about. Even a rough 60/40 beats a vague “probably.” Then put a date on the calendar for when you’ll know whether you were right, and go back and check.

You won’t predict the future. Nobody does. But you’ll stop outsourcing it, and that’s the part that was always yours to do.

I remember a monk answering a question on how to fall asleep for someone who was having problems sleeping.

His advice was: don’t think about sleep.

He suggested thinking about how comfortable the bed felt or how nice the room was. He explained that the moment you think about sleep, you’ll think about how you’re not sleeping. Then you’ll start to think about how long you’ve been not sleeping. Then you’ll think about tomorrow’s meeting, the run you’ll have to skip, and the slow drift toward exhaustion that never quite arrives. You toss and turn and repeat. It turns out thinking is the problem. Sleep doesn’t come because you want it. It comes because you stopped trying.

A lot of people in that room laughed. There is lots of wisdom there.

Despite how minor this topic was compared to other headier topics discussed in many sessions, I think about that talk often. Not at 3am, mostly. Any time I’m trying to feel a way other than how I feel.

The trap of good and bad

Most of us grew up with a vague spectrum of feelings. Joy, calm, gratitude, love, these are the good ones. Anger, sadness, jealousy, grief, fear, these are the bad ones, the ones we were quietly taught to manage, suppress, fix, or hide. The good ones we welcome and the bad ones we treat like uninvited guests.

This ends up with the same trap as the one with sleep.

You can’t push a feeling away by deciding it’s bad as it becomes another layer of the feeling. Now you’re not just sad, but you’re sad and also ashamed of being sad. You’re not just angry, you’re angry and worried that your anger makes you a bad person. The story you tell yourself about the emotion becomes heavier than the emotion ever was.

Subscribe now

What we usually try

When a feeling shows up that we’ve decided is wrong, we try a few tactics:

We distract ourselves. We open a phone, pour a drink, or start scrolling. The feeling goes quiet for a while and we count that as a win, even though it’s still there, demoted to background noise.

We hide the feeling. We tell ourselves we shouldn’t feel this way and push the feeling somewhere it won’t be seen. It goes down, but it doesn’t leave. It comes out later as tight shoulders, a short fuse with someone you love, a bad night’s sleep.

Finally, we vent. We tell someone, share, or post about it. There’s a small relief in this, like opening a valve, but the pressure builds again because the feeling itself never got met. Worse, we hand a piece of it to someone else, and over time the people closest to us start carrying our load.

None of these are wrong, but there is another way I have found liberating: letting the feeling be there.

Sitting with it

I’ve learnt, for myself, to let “bad” emotions or thoughts be in the body. I don’t argue with them, don’t moralize about them, don’t try to reframe them into something more presentable. I feel the heat in my chest, the heaviness in my body, the tightness in my jaw, and let it move through.

It’s much less dramatic than it sounds. The feeling becomes ordinary as soon as you stop feeding it a story. The shame of being sad falls off when you stop calling it shameful. The fear of being afraid stops replicating once you stop fighting the fear.

What surprised me, the first time I really tried this, is that big feelings are usually three or four feelings stacked on top of each other.

If I sit with anger long enough, sadness shows up underneath it. If I sit with sadness long enough, fear is usually underneath that. If I sit with fear long enough, there’s something quieter under it, something that feels like an old belief about whether I’m okay as I am.

You don’t get to the bottom layer by attacking the top layer. You get there by being patient with what’s actually present.

Don’t just do this for “bad” emotions though. As much as we welcome “good” emotions, I have found that sitting with “good” emotions exposes their hold on me; while it might feel good to revisit “good” emotions and memories, it is even more liberating to let them go.

The present is the only place this works

“Live more in the present” sounds like a wellness slogan or mindfulness mumbo jumbo until you notice what it really means. It means meeting what’s here without trying to swap it for something else.

The present moment, by definition, is the only place a feeling can actually be experienced. The minute you start labeling it, you’ve left. You’re now in a story about the feeling, not the feeling itself. And the story is where the suffering lives.

Anger isn’t bad. Anger is information. Sadness isn’t bad. Sadness is information. Fear isn’t bad. Fear is information. The category we filed them under was ours all along.

When you stop rationalizing, when you stop trying to feel a way you don’t feel, the emotion does the thing it was always going to do anyway. It moves. It changes. It softens. Sometimes it stays a while, but it stops running the show.

You fall asleep when you stop trying to fall asleep. You feel better when you stop trying to feel better.

May you be present and happy.

Waterloo at 17. PhD at 27. Google in 2007. Director of engineering by 2018.

Each one a rare achievement. Each one took years of work I treated, at the time, as obvious.

I didn’t know they were rare. That’s the part I keep returning to. At 17, applying and getting admitted to a Canadian university from Singapore felt routine. The PhD felt routine. The Google offer in 2007 felt routine. The director promotion eleven years later did too. I was working hard, I was inside it, and I had no reference point for how thin the path I was walking was.

Here’s how I viewed this initially. Not knowing how rare these events were was how I even was able to get ahead. If I’d understood the odds at 17, I might have hesitated. If I’d felt the weight of what a PhD means before starting one, I might not have started. If I’d benchmarked the Google offer against alternative timelines, I might have second-guessed it into something worse. If I’d seen how few engineers make director-level, I might have played the years between safer. Beginner’s luck isn’t a side benefit of youth. It’s how being inside something works.

I half-believe that but the truth is that I was extremely lucky… and foolish. And being foolish let me cash the luck without flinching. There were peers who took equally bold swings, equally oblivious, and didn’t land them. They don’t write Substacks about it. The 17-year-old version of me wasn’t brave. He was unburdened and that’s different.

Now I’m in my 40s.

Subscribe now

I write on Substack. I’m training to break 20 minutes for 5K. I have children whose lives I’m responsible for shaping in ways that will take decades to unfold. I sit every morning, watching my breath, doing a practice I’ve done long enough to trust and not long enough to stop wondering whether I’m fooling myself.

Some might consider these to be smaller stakes than getting into university or starting a career. They are also harder.

The 5K is the cleanest example. On an ultramarathon I could hide inside grit and aerobic base. A 50K is long enough that I could bare my teeth and grind through the miles. A time-trial 5K has nowhere to hide. I know exactly how much it will hurt because I’ve been timing intervals for months. I know my threshold pace to the second. The 17-year-old version of me wouldn’t have been afraid of a 5K. He would have run it. The current me is afraid of it, because I understand precisely what’s being measured and how easily I might fall short.

That’s the observation that I have been sitting with. Back then, I reached without knowing what I was risking. Now, I reach knowing exactly what I’m risking, and I reach anyway.

The fears are specific. I’m afraid the writing and the vlog will fail and I’ll have to admit I’m bad at my hobby. I’ll never see sub-20, and the failure won’t be lack of work or grit. It’ll be lack of talent. I’m afraid I’m making the wrong calls as a parent. I’m afraid my meditation practice is a sham and I’m delusional.

I’m afraid I can’t handle life.

These thoughts arise. They aren’t me, but they arise. I give them space. I look at them. I let them be what they are without arguing with them and without believing them.

That’s what decades of practice have given me, and it is smaller and more useful than the version of equanimity I once thought I was working toward. The fears don’t go away. The grip loosens. The reaching becomes possible, not because I’ve stopped being afraid, but because the fear stops being the thing in charge.

What I once called luck was something else. It was a path I was placed on, by parents and teachers and the timing of where the world was going. I walked it well, but it was given. I worked hard and took risks but was too inexperienced to understand that I was reaching. The reaching I’m doing these days is a path I’m authoring with eyes wide open.

The first kind of luck delivered me somewhere. The second kind, if it exists, has to be built.

And I’d rather try to build it than settle into the version of life the first kind earned me. Not because I have something to prove. Because the conditions that make a life feel like it means something don’t arrange themselves. Someone has to do the arranging, and the older I get, the less willing I am to wait for that to be someone else. Life, as it turns out, is about the journey, not the destination.

Eyes open. Reaching anyway.

That’s the work now.

He pulled over and rolled down the window.

“Come on in, I’ll give you a ride to the top.”

She smiled and shook her head.

“Thanks but no. If I don’t do it today, I can’t do it tomorrow.”

He drove on. She kept walking.

What do you want to still be doing when you’re her age?

That’s it. Sometimes there isn’t much more to say.

Running. Writing. Meditation. Climbing the hill behind your house. Whatever it is, you do it today. You do it tomorrow. Even when you’re tired. Even when it’s uncomfortable.

Same course. Same race. Mostly the same legs.

The difference came down to a sentence I wrote at the end of my last post about the race execution. Open at 4:40-4:45/km, not 4:37/km. Hold pace. Finish strong.

I committed to that plan in writing, then trained for five weeks to make it the right plan.

Today both showed up. The legs and the discipline.

The plan, executed

KM 1 was 4:44/km (7:37/mi). HR 154. Inside the planned window. A large pack streamed past me in the first 200m and I let them go. That was the plan, that was the plan, that was the plan.

KM 2 dropped to 4:36/km (7:24/mi). HR jumped to 170. I knew that was a touch too fast, the same trap as last time. But the body felt steady, the breathing was controlled, and I held the rhythm.

KM 3 was 4:45/km (7:39/mi). The slowest split. HR 173, legs starting to talk. This is the kilometer that broke me last time.

KM 4 was 4:41/km (7:32/mi). HR 175.

KM 5 was 4:41/km (7:32/mi), with a 4:33/km kick over the final 76 metres. HR 177.

A 9-second spread from fastest to slowest km, against the 36-second spread last time. By halves: 11:42 for the first 2.5km, 11:44 for the second. A 2-second positive split. That’s the controlled fade I was going for. Open conservative, settle in, fight for it on the back half. Last time the second half was 36 seconds slower than the first. Today it was 2.

Honestly, I was hoping for a clean negative split. A 2-second positive is close enough to even, and I’ll take it.

The pace tells one story. The parkrun stats round out the rest.

The headline number is age grade: 57.75% to 60.06%. Age grade adjusts your time for age and gender, so it’s the closest thing parkrun has to an apples-to-apples score across runners. The 60% line is parkrun’s informal “good” threshold. I crossed it today. Yay me. Pat on the back YJ.

A note on the course: Sunnyvale Bay Trail parkrun is gravel and trail, never going to be a fast course. My Garmin’s auto-detected 5K split landed at 23:24, with the full course measuring 5,076m end-to-end on GPS. Official parkrun time is 23:47 over the course as run. The slope matters more than the absolute number, and the slope today was kind. (Yes I know I’m hedging a bit here but cut me some slack.)

Subscribe now

What got me here: 6 × 1080m on the track

The training between parkruns was four sessions of 6 × 1080m repeats on the track, each Wednesday afternoon. Same idea as the 6 × 800m block from January, but with longer reps to build the kind of threshold endurance a 5K demands.

Mar 25, four days after the parkrun: 4:37/km average. A check-in run, more or less.

Apr 1: 4:33/km. Four seconds faster per km, same heart rate.

Apr 8: 4:33/km. A plateau. The body taking stock.

Apr 15: 4:26/km. A breakthrough session. Six reps between 4:23 and 4:31, average HR still 151. Same heart rate as the first session of this block, eleven seconds per km faster. The fitness was finally where it needed to be.

The pattern over four sessions: −11 seconds per km at the same effort. That projects to roughly 22:10 over a flat 5K, all else equal. Now I know how big an “all else equal” that is.

A nod to the 20-year-old me

Three days before the race, I ran a 2.4 km tune-up. Same distance as the IPPT 1.5-mile test from my Singapore army days. I came in at 11:05, average pace 4:37/km, average HR 157.

For comparison, I ran 9:40 for the same distance at age 20. So I’m 1:25 off my 20-year-old self forced to run all around in the training camp all day long for months. I’ll take it.

The tune-up wasn’t about the time. It was about practising race effort once before race day, finding 4:35-4:40/km again so my body knew what it felt like in the legs and lungs.

Same fitness, mostly. Different race.

The 57 seconds didn’t come from one place.

Some came from real fitness gains over five weeks of intervals. The 4:26/km session on April 15 was eleven seconds per km faster than the same workout in March, at the same heart rate. The 2.31-point age grade jump confirms it.

But most came from pacing. From not redlining at km 2, not crawling at km 5. Look at where the two lines diverge in the chart. Through km 2, they’re within three seconds of each other. Both runs were at the same fitness through that point. From km 3 onward, parkrun #1 falls off a cliff while parkrun #2 holds the line. The 32-second gap at km 5 is mostly a pacing gap, not a fitness gap.

Pacing is fitness’s most underrated multiplier.

Sub-20, eventually

Sub-23 is the next gate. 47 seconds away. At my current shape that means a 4:35/km even pace, no fade, HR under 175 until km 4. The plan: another 2-3 sessions of 6 × 1080m at 4:25-4:30/km, then maybe a session of 8 × 400m to sharpen the top end. Then race again in 4-6 weeks.

Sub-22 next. Sub-21 after that. Sub-20 is still the north star.

When I started this whole 5K experiment back in September, some part of me quietly believed I’d be sub-20 by spring. I had it earmarked as a fast project, the kind of self-improvement arc that would resolve cleanly inside a year, with a tidy graph at the end and a victory post to write. Half a year in, I’m 47 seconds short of sub-23, and the timeline has stretched past anything I’d planned. The math is honest now in a way it wasn’t in September.

Sub-20 will take as long as it takes. Could be six months. Could be eighteen. Could be never, in which case sub-21 will do nicely. Forty-something legs don’t pick up speed the way twenty-something legs did, and the climb is patient, whether I am or not.

I’m slowly getting there. That part I’m sure of.

I went from a 50K at 14:25/mi (8:58/km) to a 5K at 7:39/mi (4:45/km). 1080m repeats at 4:23/km. 60% age grade, finally. The direction is right. The process is working. The pace will keep moving as long as the work keeps showing up.

You train for what you want to do. Today I wanted to pace correctly. Next time, I want to do it 47 seconds faster with both eyes still on the Sub-20 goal.

Two weeks ago I wrote about what came into focus when I stopped moving. Years of tea I had drunk without tasting. Years of conversations I’d half-attended. Years of treating my body like a server I just needed to keep online.

Last week I wrote about tombstone goals. The goals worth having are the ones you never finish.

Both pieces circle the same question without naming it though. Why did it take leaving work for any of this to become visible?

The system I was inside wasn’t built to let me see. It strangely allowed me to selectively think critically about many parts of life and work, while simultaneously numbing me to others. And I was an enthusiastic participant in keeping it that way.

I know how this reads. Old me would have dismissed it. I don’t have a better way to put it. But if you’ve felt the shape of your life isn’t quite yours, and couldn’t name why, this is what I’ve got.

You would rather be busy than face yourself

The things that matter are the things that scare you. Finding your truth in life. Being a present parent. Writing something honest. Starting a project that might fail in public. They scare you because the moment you actually attempt them, you risk discovering you’re not as good as you’d hoped, and that you might not get there in the time you have left.

Work productivity hacks do not scare you though. Inbox zero does not scare you. Calendar tetris does not scare you. Reorganizing a doc does not scare you. Learning how to manage and lead projects and organizations does not scare you. None of these expose you, so you choose them. Gladly. You tell yourself you are being productive and responsible, and you will get to the real thing once the small things are handled.

You almost never get to the real thing.

I did this for years. I called it “clearing my plate” or “leveling up,” but it was simply avoidance.

Subscribe now

The work expands to fit

Even if you shake the distraction habit, the structure is built against us, or in a more sinister framing, precisely for us.

Work used to be task-shaped. You planted until the wheat field was planted. You baked until the bread was baked. The day ended when the thing was done, and the thing was visible.

The factory flipped that. Work became time-shaped. You were not paid to finish something; you were paid to be present for a block of hours, and the measure became how much you could produce inside that block versus yesterday’s block. Slowness became theft. Idleness became sin. The clock on the wall became the boss, and free time existed to recharge you for the next shift.

Knowledge work inherited all of it and lost the one honest part. The clock stayed. The shame around slowness stayed. What disappeared is the boundary of the wheat field, the point where you look up and say done.

In place of done, we built metrics. Launch counts. OKRs. Levels and promotions. Quarterly numbers that roll into annual numbers that roll into your career. The metric is not a measurement of the work; over time it becomes the work, because the work has no boundary of its own.

Metrics do two things a wheat field never did. They are gameable; you learn quickly which numbers move your career and which do not, and you route your effort there. They are comparable; everyone’s score is visible to everyone, which turns a job into a standings table. You are not doing the work. You are placing in it, and in a standings table, there is no done. There is only the next score.

Which is why finishing early does not end the day. New tasks appear. Things you didn’t think were important suddenly seem worth doing because you have the room for them. The vacuum doesn’t stay a vacuum. You fill it because not filling it feels like failure.

Speed is a ratchet

Once you can do something faster, you cannot go back to doing it slower. Your colleagues recalibrate. Your manager recalibrates. You recalibrate. The new speed becomes the floor, not the ceiling.

This is why the productivity premise breaks from the start. Save thirty minutes and you do not get thirty minutes back. You get thirty minutes of new work that wasn’t there before, plus a recalibrated expectation from everyone around you that you will keep producing at the new pace.

Years of this and the floor is somewhere near the moon. Now AI is being marketed as the next 10x. If the pattern holds, and it always holds, the floor moves up another order of magnitude and we will be sprinting against it within a quarter. The marketing is “more time for what matters.” The lived experience will be the same backlog with more capacity to grow it.

The only people I know who escaped this ratchet either left, got fired, or got sick. I have not yet met someone who stayed inside the system and quietly opted out of the acceleration. AI is not a way out of this. It is the disease dressed as the cure.

What it costs you

The first cost is guilt. You are never getting to the things you are “supposed to,” because the list is infinite, so you live in low-grade shame about your output. You try to fix the shame by working more, which produces more shame. The loop sustains itself.

The second cost is that you stop getting to what matters. When everything is on the list, nothing is prioritized. The big things, the ones that would actually move something, get buried under small things that yield instantly. You spend your career doing the urgent and never the important.

The third cost is the present. You trade today for a future state that never arrives, because the future just becomes another today with the same backlog. You are always paying down a debt that compounds faster than you can service it.

The fourth cost is isolation. The fastest way to protect your time is to remove people from it. Skip the coffee chat. Decline the dinner. Close the door. Get to the office before everyone else arrives. You optimize yourself into solitude and call it focus.

The fifth cost is the one I felt most where you lose the capacity to notice. Attention itself atrophies. You stop tasting food. You stop hearing what your kid is actually saying. You stop registering that the season has changed. The machinery is running but the operator has left the building.

The system works

The thing to understand about the system is that it works. That is what makes it a system. Companies grow. Shareholders are served. Metrics are hit. Products ship. It is not broken from its own point of view; it is doing exactly what it is built to do.

The human cost is not in the spreadsheet. It is not meant to be. While you are inside it, the cost feels small because the rewards arrive on a faster timescale than the damage. You get the promotion this year. The loss of attention, the atrophied body, the friendships that thinned out, those come due later. Each trade, in isolation, looks reasonable. The aggregate is not.

And the aggregate arrives at a time the system has engineered you to accept. By the time you retire, you are tired. You have spent the instrument you would need to object with. The deal is closed. You tell yourself it was worth it, because saying it wasn’t would be unbearable. Most people take the deal. The system counts on it.

None of this requires malice. No one designed it this way. It is what happens when millions of people optimize for the same legible metrics over a long enough stretch. We are the system. We keep it running by showing up.

What helps

I do not have a clean fix and I am suspicious of anyone who sells one. A few things have started to help, in the order I tried them. The previous two essays are really two halves of one answer. Pick goals the system cannot complete for you, which is the tombstone piece. Stop moving long enough to notice you are alive, which is the tasting piece. What follows are the small moves that make both of those possible from inside an ordinary week.

Stop expecting to win the game. The first move is admitting the math. You will never get to the bottom of the list. There is no clearing it. Once you accept this, you stop running toward a finish line that doesn’t exist, and you can start asking the only useful question. What do I want to spend today on?

Stay in the discomfort. When I am working on something that matters and the pull to check social media or not get stuck in a rut gets strong, that pull is the signal I am close to something real. The urge to distract is the tell. Sit longer.

Pick the important thing first, before anything else can fill the slot. If I wait for the energy and motivation so I can write, I will not write. If I wait for the perfect conditions to run, I will not run. The slot fills with something else, every time. So I work on my priority first and let the small things absorb whatever is left. Most of them turn out not to need me.

Cap your in-flight projects. I keep three. New one in, old one out. This sounds restrictive. It is the most freeing constraint I have used.

Build in time with no purpose. A run with no metric goal. An hour with a book that won’t make me money. A conversation that isn’t about anything. These were the first things I cut while working and the first things I had to deliberately put back. This is where the tasting comes back.

I am fortunate. I had the conditions to step away, and I have the conditions to stay away on my terms. Most people don’t. I don’t want to pretend otherwise.

I don’t think the answer for everyone is to leave their job. But while you are still in it, what is the system you are inside making invisible to you, and what would you have to stop doing to see it?

Most people can’t answer one simple question: what do you want out of this life? You’ll get a shrug, or a canned answer, or somebody else’s answer dressed up as their own.

I’ve come to think this is the only question that matters. Not because the answer is hard to find, but because we’re allergic to sitting with it long enough to hear ourselves think.

Here’s what I’ve learned about goals after spending my life getting to a place that was very good at handing them to me, and some time outside trying to write my own.

One thing up front: this isn’t about work goals. Work goals are a different conversation, and frankly an easier one. Someone will always be happy to hand you the next rung. The hard, important conversation is the other one; the one about what you want out of your actual life, which nobody’s going to hand you at all.

Start with the tombstone

Start with the destination. You can’t see the path from here anyway, and the destination is easier; it’s how we used to do it as kids when adults asked what we wanted to be when we grew up and we answered without flinching.

The goals worth having are the ones you’d want on your tombstone. Good father. Kind. Wise. Present. A son my parents are proud of. A friend people can count on. Have at least one of these. If you don’t, you’re climbing a ladder and you don’t know whose.

The test for whether you’ve found the right one is whether you sit up a little straighter when you write it down. If the words feel performative, or like something you’d say in a job interview, they aren’t yours yet. Keep going.

For me, one version of it is the hike. Specifically: being the kind of father my kids remember taking them into the hills with no agenda. No summit. No mileage target. No lesson. Just walking until somebody wants to stop and look at a bug. If that’s on the tombstone, I’ll have done something right.

Here’s the thing about tombstone goals that took me a while to appreciate: you never finish them. And that’s the feature, not the bug.

Work goals are achievable, and that’s what makes them feel productive, but it’s also what makes them hollow. You hit VP and the next morning it’s “now what”. You hit the number and it crumbles in your hand like sand. Anything you can finish turns out not to be the thing. Tombstone goals don’t have that problem. There’s no finish line for being a good father. You just keep showing up, and you get to keep showing up for the rest of your life. The practice is the point.

Subscribe now

Can you practice it today?

Not every life goal passes this bar. “I want to be happy” sounds like a tombstone goal but it isn’t, and the reason is worth sitting with. Happiness is a mood, not a practice. You can’t sit down and do happiness; you can only set up the room and hope it shows up.

The test is this: can you embody it today, in some small action, with nothing but your own behavior?

Being a good father means taking the kids on the hike this Saturday. Leave the phone in the car. Being kind could mean sending the message you’ve been putting off. Don’t post the angry thing. Being a good son might be calling your parents tonight, not “soon”.

In any case, happiness turns out to be what shows up when you’re busy doing the others well.

Write it down. Or don’t.

There’s something about putting words on paper that makes them real. You can’t hide from a sentence you’ve written. The vague longing in your head becomes embarrassingly thin once it’s sitting there in black and white, and you’re forced to either sharpen it or admit you don’t mean it.

So write it down. A page, a paragraph, five sentences. Whatever fits.

But if you already know what you want, deeply and without flinching, the writing is optional. Some people need the document. Some people need the clarity. Don’t let the ritual become the point.

The point is action.

Use the goal as a filter

You can have the most beautifully articulated life vision on the planet and it will do nothing for you if you don’t move. Goals don’t change your life. Decisions do. And decisions only count when you make them with your hands and your calendar, not in your head at 11pm.

Once you know what you want, the goal becomes a filter. Every opportunity, every invitation, every shiny object floating past gets held against it. Does this serve the thing? Yes, lean in. No, decline. Especially if it’s flattering. The Saturday hike stays on the calendar. The prestigious-sounding thing that would quietly eat the day doesn’t.

This sounds obvious. It is not what most people do. Most people say yes to whatever’s loudest, then wonder why their year disappeared.

The compounding benefit of filtering well is what people call luck. Luck is preparation meeting opportunity, and the preparation half is almost entirely under your control. Knowing what you want is preparation. Saying no to the wrong things is preparation. Showing up consistently in the area you care about is preparation. Do that work for long enough and opportunities you couldn’t have planned for start looking custom-wrapped for you. They were always there. You couldn’t see them before because you didn’t know what you were looking for.

You can only change yourself

The goal tells you what to do. Not anyone else. You can’t write down “I want my kids to love me” and expect them to cooperate on a schedule. You can’t decide you want a better relationship and force the other person to meet you there. You can’t will your parents into understanding you.

Life happens through give and take with the world. Everyone is operating inside the same system, with their own goals, their own incentives, their own bad days. The only lever you control is your own behavior: what you learn, what you practice, who you show up for, how you act when nobody’s watching.

So focus there. Prioritize the learning. Send the message. Do the rep. Let the world respond however it’s going to respond, and adjust.

Wanting something from someone else is not a strategy. Becoming someone worth being close to is.

The middle is mostly discomfort, and that’s the good part

The journey to anything worth having is mostly doubt, grief, and quiet fear that you’ve made a terrible mistake. Not occasionally. Mostly. The triumphant arrival photo at the end is maybe 1% of the experience. The other 99% is sitting with “what if this doesn’t work”.

You’ll feel less supported than you expected, too. The friends who cheered you on at the start get bored. The people you thought would understand don’t. You’ll be alone with it more than feels fair.

There’s a “whatever it takes, grind through it” framing for this part, and I won’t pretend that mentality isn’t real or useful. But I much prefer the other one: the discomfort is here right now, and I’m enjoying it anyway.

I’m writing this, in part, on a trail run. Lungs burning, legs heavy, recording between climbs. I have no idea where this particular thread leads. That’s a big part of why it’s fun. The discomfort isn’t the tax I’m paying for some future payoff; it’s the texture of doing the thing I want to be doing. Growth happens because you’re moving toward something you care about, and the burn is how you know you’re moving.

If it felt comfortable the whole way through, it probably wasn’t a stretch worth making. So when the doubt hits, don’t grit your teeth and white-knuckle it. Look around. Notice the trees. You’re already in the part that matters.

When the doubt hits, don’t panic

The instinct when you feel stuck is to thrash. Work harder. Pivot. Quit. Burn it down. Try five new things at once.

Don’t.

When the fear shows up, the move is the opposite. Slow down. Re-read what you wrote, or what you would have written. Ask yourself, calmly, whether the destination still makes sense. Most of the time it does. Your circumstances haven’t changed; your nervous system is having a bad afternoon. You had one bad hike. You didn’t become the wrong kind of father.

If the destination still makes sense, the answer is almost always the same. Heads down. Next small action. Then the one after that.

That’s the whole technique. Pick a goal you can practice. Use it as a filter. Take action. Don’t panic when it hurts.

What do you want?

Sit with that. The answer is the only one that matters.

Three weeks after I left Google, I was drinking tea over the kitchen sink, and my eyes filled up.

Not because I missed the job. Because the tea was extraordinary, and I had no memory of the last time I’d tasted it. Not “the last good one.” The last one. I’d been drinking tea for years without tasting it.

That was the first thing I noticed about leaving: the world had been there the whole time. I just wasn’t.

The world isn’t busy. You are.

The story I told myself for two decades was that life was loud and fast and I was doing my best to keep up. The truth, which I can only see in retrospect, is that the loudness was inside my head. The calendar wasn’t the problem. The voice running underneath the calendar was the problem; the one keeping a tally of unread messages while I was supposed to be reading my kid a book, the one rehearsing tomorrow’s presentation while I was supposedly on a date with the Mrs.

When that voice finally got quiet, the world didn’t slow down. I started catching up to it.

Subscribe now

What came into focus

My kids’ faces. I could not have described, with any confidence, the exact shape of my younger one’s smile. I’d been looking at him every day. I had not been seeing him. The first month after I left, I kept catching expressions on my kids’ faces I’d somehow never registered, and the feeling wasn’t joy. It was a kind of grief for the years I’d watched them on autopilot.

My parents getting older. This one stings more. There’s a particular slowness to the way my mum reaches for her cup now that wasn’t there a few years ago, and I don’t know when it started, because I wasn’t around for the start. Slowing down meant I finally had the bandwidth to do the math on how many visits we probably have left. The number is small enough to be useful.

The trail. I run a lot. I ran a lot before, too. But I used to run through places. Now I run in them. The light coming sideways through the redwoods at 7am isn’t a backdrop anymore; it’s the whole reason I’m out there. The 5K times still matter to me, the speed work still hurts, none of that has gone away. What’s changed is that the run isn’t a transaction with my future self. It’s something happening right now, to a body that won’t always be able to do this.

My own body, generally. Hunger. Tightness in my shoulders. The specific quality of being tired versus being depleted. When you’re moving fast enough, your body becomes a vehicle you complain about and ignore. When you slow down, it turns back into the thing you live inside.

The harder seeing

Some of what came into focus was less fun.

I started noticing how much of my old identity had been outsourced to a logo. I’d known this in theory for years; everyone at a big company half-knows it. Knowing it in your body, on a Tuesday morning with no meetings on the calendar and nobody waiting for your reply, is a different thing entirely. There’s a stretch of weeks in there that I’d describe, generously, as withdrawal.

I also started seeing how much of my busyness had been a kind of hiding. If you’re always responding to the next thing, you never have to sit with the question of whether the things you’re responding to are the right things. Speed is a wonderful anesthetic. It only stops working when you stop.

And I started noticing how little I controlled. Not in a dramatic way; in a daily way. The market does what it does. My kids will become who they become. My parents will age regardless of how many flights I book. There’s a humility that arrives, uninvited, when you finally have the time to look at your life clearly. It’s not defeat. It’s more like finally putting down a bag you forgot you were carrying.

What I’d tell the version of me still inside it

I’m not going to tell anyone to quit their job. That wasn’t the lesson, and it isn’t transferable. Plenty of people I love and respect are deep inside companies right now and doing the best work of their lives. The job wasn’t the enemy.

The enemy was the gear. I had exactly one speed for twenty years, and I mistook it for who I was.

If you’re reading this from inside a calendar that owns you, the thing I’d offer isn’t slow down. It’s find one place where you have a different gear, and protect it like it matters. A run with no watch. A meal where the phone is in the other room. A walk with your kid where you have nowhere to be after. Something the optimizer in your head can’t get its hands on.

You’ll be amazed at what’s been waiting there.

The tea was there the whole time. You just have to be the kind of person who can taste it.

I spent nearly two decades at Google. Good salary, good benefits, good trajectory. The conventional playbook said to keep climbing, keep earning, keep upgrading. Bigger house, nicer car, fancier vacations. More money equals more options equals more freedom. That’s the pitch.

Here’s what happened instead: my spending grew alongside my income, and my sense of freedom stayed flat. Every raise got absorbed. A slightly nicer car. A slightly better restaurant habit. A slightly more expensive hobby. None of it was extravagant on its own. All of it added up to the same feeling: I’m earning more and I don’t feel any freer.

A decade ago, the FIRE movement (Financial Independence, Retire Early) had an answer for this. Save aggressively, invest in index funds, retire in your 30s or 40s. The math was sound. But FIRE seems to have lost its energy, and I think it’s because the movement got stuck on the retire part. People optimized for an exit date and then discovered that “not working” isn’t a life philosophy. It’s a void. The interesting question was never “how do I stop working?” It was “how do I stop needing so much that work becomes a trap?”

That’s the question I’ve been living with. When I left Google to focus on family, writing, and the pursuits that matter to me, I didn’t have a grand financial masterplan. What I had was a philosophy I’d been assembling for years from books, from watching others, and from my own mistakes. Three rules.

1. Avoid lifestyle creep

2. Place asymmetric bets

3. Strive for less.

The reward is peace of mind. Not the Instagram version but the unglamorous kind where you wake up on a Tuesday and realize that nothing about your day is dictated by money.

Subscribe now

The trap no one warns you about

Money is time. The economy makes that literal. Every dollar you spend represents time you traded away. Finite time. Time you’re not getting back. You don’t want to trade money you don’t need with time you do not have.

That $200 dinner? Five hours of your working life. That car upgrade? Four months. You stop asking “can I afford this?” and start asking “is this worth the hours I traded for it?” Most people never make that switch. As income rises, they face a choice they don’t even realize they’re making: save the difference, or spend the difference. Almost everyone spends it. Not deliberately. The culture treats spending as the default, and defaults are powerful.

This is lifestyle creep. Nobody sits down and decides they need a multi-million dollar house, multiple cars, and a $5,000 couch. It just happens, the way weeds take over a garden. Through inattention.

Seneca, a Stoic philosopher, had a practice for this. He’d deliberately live in poverty for a few days: rough clothes, less food, a hard surface to sleep on. Not as self-punishment but as calibration. He wanted to know: how small is the actual gap between “enough” and “comfortable”? Smaller than you think. Much smaller.

I stumbled into my own version of this when I started training for ultramarathons. I stripped back a lot, not for philosophical reasons but because the training demanded it. Early mornings on trails. Simple meals. Fewer social obligations. More sleep. And then a funny thing happened: I didn’t miss any of it. The stripped-back version of my life felt better than the padded one. That was unexpected. It also made me wonder how much of my spending had been solving problems I didn’t have.

Asymmetric bets

The second rule sounds like finance jargon. It’s not. An asymmetric bet is any decision where the worst case is small and survivable, and the best case is big. I talk about this more in The 80/20 rule for your whole life.

Leaving Google looks risky from the outside. But think about the actual downside: I had savings, I had marketable skills, and if everything went sideways I could go find another job. That’s a bounded downside. The upside? Time with my kids during years that don’t come back. Creative work that builds on itself. Space to think without a performance review shaping what I think about. Unbounded.

William Green, in Richer, Wiser, Happier, describes how the best investors think this way. They buy assets when the price is well below what they’re worth. Buy low, who would’ve thought eh? Pay $60 for something worth $100 and you have a $40 cushion. It can drop and you’re fine. But if it rises to true value or beyond, you do well. The key insight: great investors don’t chase big returns. They avoid catastrophic losses and let compounding handle the rest.

This extends way beyond the stock market. Every skill you learn is an asymmetric bet. Fix your own plumbing once, and you save $75 an hour for the rest of your life. Learn to cook properly, and you save thousands a year, forever. Each skill widens the gap between what you need to earn and what you need to spend. That gap is freedom.

The same logic shapes my crypto and options positions, my Substack writing, and turning parts of it into a book. None of these are moonshots. They’re small bets with long tails: bounded downside, compounding upside, and the patience to let them run.

Striving for less

This is the hardest rule. The entire culture pushes against it. “Striving for less” sounds like giving up. I’d argue it’s the most aggressive financial move you can make.

The math is embarrassingly simple: a dollar saved is worth more than a dollar earned. Saved money isn’t taxed. In the 30% bracket, you need to earn $1.43 to keep $1. A dollar you don’t spend is a full dollar retained. Frugality literally has a better return than a raise.

But the math isn’t the hard part. The hard part is knowing what “enough” looks like for you, specifically, before the culture tells you it should be more.

It’s funny how “giving up” and “waking up” can look exactly the same from the outside.

The old personal finance wisdom said happiness plateaus around $75,000-$100,000 in income. But recent research from Matthew Killingsworth at Penn says otherwise: happiness keeps rising with income, with no clear ceiling, even well past $500,000 a year. The data is hard to argue with.

But look at why money keeps making people happier. The biggest driver isn’t nicer stuff. It’s control. Security. Freedom over how you spend your time. That’s the finding that matters here, because it’s exactly what financial independence gives you, and you can get there much faster by needing less than by earning more. (I wrote more about the relationship between money and happiness in Money won’t fix your life. It reveals it.)

The trap is confusing the two. More income spent on more stuff doesn’t buy you more control. It buys you more complexity. A bigger house means more furniture, more cleaning, more insurance, more worry about property values. Each purchase spawns costs that weren’t in the original plan. The thing you bought to simplify your life made it more complicated. And now you need the income to maintain it all, which is the opposite of control.

The wealthiest investors I’ve read about, Buffett, Templeton, Howard Marks, all live well below what their net worth would allow. Not because they’re cheap but because they understand that peace of mind compounds too. And anxiety about money doesn’t go away when you have more of it. It just changes shape.

Financial independence works the same way. It’s not about hitting a number. It’s about reaching the point where money stops being the thing that decides what you do with your day.

What it looks like

I should be honest about what financial independence looks like in practice, because it’s not what the blogs promise.

I wake up. I run trails. I write. I spend time with my family. I mess around with code projects that interest me. I read. I sit and think, sometimes for a long time. None of this is glamorous. None of it would make a compelling LinkedIn post. But every day feels like it’s mine, and I spent enough years where that wasn’t true to know it’s worth protecting.

The three rules again:

Avoid lifestyle creep. Question every recurring expense. Periodically reset your baseline by stripping back to less, on purpose, and noticing how little you miss.

Place asymmetric bets. Learn skills, build assets, take positions where the downside is small and the upside compounds over time.

Strive for less. Define your “enough” before the culture defines it for you.

You don’t need to live on $7,000 a year. You don’t need to retire at 33. You need to close the gap between what you earn and what you spend, widen it over time, and invest the difference in things that grow quietly while you’re not watching.

Share What YJ Thinks

Instead, I get up. I walk to my cushion and meditate.

Not because I feel like it. Not because some guru told me to. Because I have a rule: I meditate at least once a day. That’s it. No caveats about sleep quality or how I’m feeling. The rule doesn’t care about my allergies.

After the sit, I make tea. I’m standing in the kitchen, cupping the warm mug, when I hear it: the unmistakable splattering of raindrops on the roof. I look outside and the street is already darkening with wet. A Bay Area spring morning, grey and cold.

This is the part where most people would say, “I’ll run tomorrow.” And look, I’ve said it too. Many times. But I have another rule: I will at least suit up for my run, even if I don’t end up running. So I change into my running gear. I lace up my shoes. And then, because I’m already dressed and standing by the door, I step outside and run in the rain.

Later that evening, I’m at the dinner table with my family. My phone buzzes. Then buzzes again. WhatsApp, probably. The little dopamine merchant, vibrating on the counter. I leave it there. Another rule: devices go away during mealtimes.

Three rules. Three small acts of defiance against the path of least resistance. None of them heroic. All of them made up by me. For myself.

Rules without enforcement

Here’s the strange part: these rules carry no external authority. Nobody imposed them. There’s no penalty for breaking them, no boss checking in, no accountability partner sending disappointed texts. They’re rules I invented for myself, in moments of clarity, and then chose to follow in moments of weakness.

And that’s exactly why they work.

The conventional wisdom on habits and behaviour change focuses on goals. Run a sub-20 5K. Lose ten kilograms. Read fifty books this year. Goals are fine as directional signals, but they have a fatal flaw: they expire. You hit the goal, or you don’t, and either way the motivation evaporates. The runner who trains for a single race often stops running after race day. The goal was a destination, not a direction.

Rules are different. Rules are ongoing. A rule like “I meditate once a day” has no finish line. It doesn’t expire after some target is met. It’s a statement about what kind of person you are, not about what you want to achieve. The person who meditates daily isn’t chasing a goal; they’re expressing an identity.

James Clear puts it well in Atomic Habits: the most durable behaviour change comes not from setting goals, but from deciding who you want to be and then gathering evidence for that identity through repeated action. Identity-driven habits don’t stop working when you achieve something, because there’s nothing to achieve. There’s only someone to be.

When I sit on my cushion at 5am with congested sinuses, I’m not pursuing a meditation goal. I’m being the person who meditates. The rule is the bridge between aspiration and identity.

Subscribe now

What makes a rule worth keeping

Not all rules are worth making. “I’ll do 200 push-ups every morning” sounds impressive, but if you haven’t done a push-up in three years, you’re writing a rule you’ll break by Tuesday. A good rule has three qualities.

It reflects who you want to be, not what you want to get. “I put my phone away during meals” is an identity statement: I’m someone who is present with my family. “I want to reduce my screen time by 30%” is a goal. The identity version survives because it’s grounded in values, not metrics. When the metric is met, the goal dies; you already reached it. Values persist.

Ask yourself: what would the person I want to become do in this situation? That question generates better rules than any “smart” goal framework.

And once you know the identity you’re aiming for, the next question is whether you can state the rule clearly enough to act on it.

It’s specific enough to be unambiguous. “I’ll exercise more” is not a rule. It’s a wish. “I’ll at least put on my running shoes” is a rule. You either did it or you didn’t. There’s no negotiating with yourself at 6am about whether a brisk walk to the mailbox counts as “exercise.” The binary nature of a clear rule eliminates the mental overhead of deciding in the moment. And deciding in the moment is where discipline goes to die.

Figure out the smallest, most concrete version of the behaviour and commit to that. My rule isn’t “I will run every day.” It’s “I will suit up.” The genius is that once you’ve laced up your shoes and you’re standing at the door, running becomes the default. The hard part was never the run; it was getting ready for it.

But specificity alone isn’t enough. The rule also needs to survive contact with reality.

It’s something you can do on your worst day. It’s way too easy to design rules for our best selves: well-rested, motivated, caffeinated, clear-skied. But rules exist precisely for the days when conditions are terrible. The 5am allergy morning. The rainy run. The buzzing phone when you’re tired and tempted.

If the rule can’t survive your worst day, it’s too ambitious. Scale it back until it can. “Meditate for forty-five minutes” becomes “meditate once.” “Run ten kilometres” becomes “put on my shoes.” “No screens after 8pm” becomes “no screens at the dinner table.” You can always do more. The rule is the floor, not the ceiling.

Designing rules that stick

You’re engineering a behaviour that needs to fire reliably across varying conditions of motivation, energy, and circumstance. A few design principles help.

Anchor it to something you already do. The most reliable trigger for a new behaviour is an existing behaviour. “After I brush my teeth, I meditate.” “After I finish my morning tea, I change into running clothes.” “When dinner is served, the phone goes on the counter.” Atomic Habits calls this “habit stacking”: attaching the new behaviour to a cue you never miss.

A free-floating rule like “I’ll meditate sometime today” leaves too much room for procrastination. A rule chained to a specific anchor has a time, a place, and a trigger built in.

Frame it as what you’ll do, not what you’ll avoid. “No phone at dinner” is weaker than “I put my phone on the counter when dinner is served.” The first version creates a void; the second fills it with an action. Your brain needs something to execute, not something to resist. Resisting is exhausting. Executing is automatic.

Commit to the process, not the outcome. “I will meditate” is a process rule. “I will achieve inner peace” is an outcome fantasy. You control the process. You don’t control the outcome. Some meditation sessions are serene. Most aren’t. On my allergy morning, the sit was laboured breathing through clogged nostrils. It was not peaceful. But I did it, and the rule held.

You can’t reliably control motivation. But you can improve your chances by keeping the behaviour tiny and anchoring it to a routine you were going to do anyway. Design for the variables you control.

Staying in the game

Making a rule is easy. Keeping it on a rainy morning when you slept four hours, that’s the test.

Lower the bar, raise the consistency. The two-minute rule is powerful: if you can get yourself to do the first two minutes, the rest often follows. I don’t commit to a forty-five minute meditation; I commit to sitting down. I don’t commit to a 10K; I commit to putting on shoes. The absurdly low bar means I almost never skip, and on most days I end up doing far more than the minimum. Consistency compounds. Intensity doesn’t.

Track, but lightly. Keeping a visual record of your streak has real power. Marking a calendar, ticking a box; these tiny acts of acknowledgment feed the cycle. You see the unbroken chain and you don’t want to break it. But track only one or two rules at a time. Over-tracking can become a chore.

Expect to fail, then return. You’ll break your rules. Everyone does. The difference between people who build lasting habits and people who don’t is not perfection; it’s the speed of return. Miss a day? Fine. Do it tomorrow. The rule isn’t invalidated by a single exception. It’s invalidated by a pattern of exceptions you stop noticing.

This is how breath meditation works. You focus on the breath. Your mind wanders, inevitably. But instead of clinging to the distraction, you gently return your attention to the breath. The practice isn’t the focus. The practice is the returning.

The danger is all-or-nothing thinking: “I missed yesterday, so the streak is broken, so what’s the point.” The point is that you’ve been meditating for 200 of the last 210 days, and that’s a fundamentally different life than meditating zero days.

Let the identity solidify. This is where compounding works its magic. Every time you follow your rule on a hard day, you deposit evidence into your identity account. After enough deposits, you stop needing willpower. You meditate because you’re a meditator. You run in the rain because you’re a runner. You put your phone down because you’re the kind of person who’s present at dinner. The rule becomes who you are, and then it’s no longer a rule at all. It’s reflexive.

Rules as architecture

I sometimes think of my self-made rules as the load-bearing walls of a house. You can rearrange the furniture, repaint the rooms, knock out a cosmetic wall here and there. But the load-bearing walls stay. They’re what keeps the structure standing.

The meditation, the suiting up, the phone on the counter: these are small, invisible, unglamorous acts. Nobody gives you a medal for putting on running shoes. Nobody applauds when you leave your phone on the kitchen counter. But these tiny daily commitments are the architecture of a well-lived day, repeated enough times to become a well-lived life.

The rules are made up. Of course they are. But they were made up by a version of you who was thinking clearly, who had the luxury of deciding what mattered, who wasn’t tired or tempted or distracted. That earlier, clearer-eyed version of you designed the rules as a gift to the version of you who would be all of those things.

Trust the gift.

Make up your rules. Make them small, specific, and survivable. Anchor them to what you already do. And then follow them, especially on the days when you don’t want to.

Because the person you become is the sum of the rules you keep.

And somewhere around mile 25 (40 km), doing math in my head to stay awake, I realized something uncomfortable: at this pace, a 100-miler would take me over a day of non-stop running.

That’s not a race. That’s a hostage situation!

In ultrarunning you can grind out longer and longer distances, but if your base speed doesn’t improve, you’re adding hours, not miles. My average pace at Crystal Springs was roughly 14:25/mi (8:58/km). Extrapolate that to 100 miles and you’re looking at multiple days of continuous movement.

I don’t want to crawl through a 100-miler. I want to run it.

So before going further in distance, I’m going backwards. Way backwards. To the 5K.

The goal: sub-20 minutes this year. Fingers crossed.

You train for what you want to do

This sounds obvious, but it took me a while to internalize. For months, my training was simple: run long, run slow, run more. That approach got me through a trail half marathon in September 2025 and the 50K in January. But it wasn’t making me faster. It was making me better at being slow. It turns out that zone 2 training isn’t all I needed to do.

The shift required a different kind of discipline: cutting back mileage and adding speed work. Periodization, the coaches call it. You can’t train everything at once. You pick a quality, you build it, you move on. The idea is that your body can only adapt to so many stresses at once. If you chase endurance, speed, and strength simultaneously, you get mediocre at all three. But if you focus on one for a concentrated block of weeks, you get real adaptation. Then you shift focus, and the gains from the last block don’t disappear; they become the foundation for the next one. It’s sequential, not parallel. Think of it as compounding interest for your legs.

For the past few months, that quality has been “speed” (relatively fast for me, of course).

Most people assume the 5K is a sprint. It isn’t. A 5K race is roughly 90-95% aerobic. Three systems matter:

• VO2 max: the maximum rate your body can consume oxygen during hard effort. Your aerobic ceiling.

• Lactate threshold: the pace at which lactate accumulates faster than your body can clear it. The intensity you can sustain before things start to burn and fall apart.

• Running economy: how much energy you burn at a given pace. The efficiency of your stride. Two runners with identical VO2 max can have wildly different race times if one wastes less energy per step.

All the same systems that power an ultra, compressed into 20-something minutes.

Growing up in Singapore, the IPPT fitness test used by the Army (among others) includes a 1.5 mile (2.4 km) run as the endurance benchmark. At my peak, age 20, I ran it at about 9:40. Fast, but not blazing. A sub-20 minute 5K works out to roughly the same intensity. I’m now two decades and change past that peak.

I’m under no illusion this will be easy, or even possible. But if you don’t try, you don’t know. And here’s the upside: every second I shave off my 5K pace makes my ultras faster by default. Speed feeds endurance.

Subscribe now

Finding the floor

I started in September 2025 with tempo runs at 5K+ distances. No intervals, no track work. Run hard for 3-4 miles (5-6 km), see where I am.

Here was where I started: 9:44/mi (6:03/km) pace. A 30-minute 5K.

Over four months of consistent tempo work, the pace dropped steadily. 8:50/mi (5:29/km) by October. Sub-8:03/mi (sub-5:00/km) by November; 8:02/mi (4:59/km), my first PR of the journey. A slight regression to 8:15/mi (5:07/km) in December, which I’ll chalk up to holiday laziness.

The trend was clear though. Four sessions, 90 seconds per mile faster. My body was responding to the training.

Enter the interval

In January 2026, I shifted to structured intervals. The concept is simple: instead of running one continuous hard effort, you break the work into shorter, faster segments with recovery jogs in between. You run each segment harder than you could sustain over the full distance, rest enough to partially recover, then go again. The recovery lets you accumulate more total time at a faster pace than you could in a single effort. Over weeks, your body adapts to that speed, and what felt unsustainable starts feeling normal.

My format: 6 x half-mile (6 x 800m) repeats with recovery jogs. The goal: build top-end speed that I could eventually sustain over longer distances.

Session one: 7:34/mi (4:42/km) average across the reps. Consistent enough. HR manageable at 153.

Over seven weekly sessions through March, the pace came down like clockwork: from 7:34/mi down to 7:02/mi (4:42 to 4:22/km). Thirty-two seconds per mile faster in seven weeks, and my cadence climbed from 182 to 186 SPM. My legs were learning a new rhythm.

By the final session, I was running each half-mile rep at 7:02/mi (4:22/km) with a spread of only 11 seconds across six reps. The speed was there. Time to race. Or so I thought.

Race day

March 21, 2026. Sunnyvale Trail Running parkrun. My first timed 5K.

For the unfamiliar: parkruns are free, weekly, timed 5K that happens every Saturday morning in parks around the world. No entry fee, no qualifying time, no pressure. You show up, you run, you get a time. Volunteers make it happen. It started in the UK in 2004 with 13 runners in a London park and has since grown to millions of runners across 23 countries. It’s the on-ramp to racing for people who’ve never raced, and a no-stakes speed check for people who have. You register once online, print a barcode, and that barcode is your lifetime passport to any parkrun anywhere.

I went in targeting sub-23 minutes. I came out... not as expected, in 24:44.

The pacing chart tells the whole story:

See it? I thought I was running intervals. The first km I ran in 7:26/mi (4:37/km), which was interval pace. I went out like I was running a half-mile rep, not a 5K race. My heart rate spiked 16 bpm between km 1 and km 2, and by km 3 I was redlining at 176. The last two kilometers were damage control.

A 58-second positive split per mile from first to last km. Classic beginner racing mistake.

It was a good race. I finished, I learned, but it wasn’t the time I wanted.

What the numbers say vs. what my legs said

Here’s what’s frustrating: the fitness for sub-23 was there. My interval training had me comfortably holding 7:02-7:26/mi (4:22-4:37/km) in workouts. An even pace of 7:23/mi (4:35/km) would have given me 22:55 but I most likely should have adjusted my target given how I felt that day. Instead, I blew up because I overestimated the fitness I was in and couldn’t resist the adrenaline of the first kilometer.

Here’s what’s encouraging, since I’m a glass-half-full person: 24:43 with terrible pacing on a first-ever 5K race. Imagine what happens when the pacing is right.

What’s next

We keep going.

A few days after the parkrun, I ran my first session of 6 x 1km (6 x 0.62 mile) repeats at threshold. The shift from 800m to 1km reps is deliberate: the longer interval teaches the pacing discipline I clearly lacked on race day.

I managed to keep six seconds of spread across six reps. That’s the pacing consistency I needed during the race but didn’t have.

More importantly, while my parkrun first km was 7:26/mi (4:37/km), today’s average across 6 full kilometers of work was 7:26/mi (4:37/km). Same pace, completely different outcome. In the race I blew up after one kilometer at that speed. In training, I held it for six. The difference in this case is pacing strategy, not fitness.

The plan: 3-4 sessions of these 1km repeats, then another parkrun. This time, the strategy is different. Open at 7:31-7:39/mi (4:40-4:45/km), not 7:26/mi (4:37/km). Hold through kms 3-4, and finish strong. Target: sub-23:30 as a stepping stone.

Sub-20 is still the north star. That’s 6:26/mi (4:00/km) pace, and I’m at 7:26/mi (4:37/km) in training. There’s a long way to go. More interval blocks, probably some 400m work for raw speed, and more racing experience.

But I ran a 50K at 14:25/mi (8:58/km) and now I’m running 1km repeats at 7:26/mi (4:37/km). The direction is right. The process is working.

You train for what you want to do. Right now, I want to get fast.

If you are interested in my fitness journey, click here to follow me on Strava.

You can learn a new language in a year. You can switch careers at 40. You can leave the relationship that’s slowly draining you. You have neuroplasticity, opposable thumbs, and access to more information than any generation in history. The raw capability is there. It’s always been there.

And yet.

You keep meaning to start the project. You stay in the job you hate. You tell yourself next month will be different. You watch someone else do the exact thing you wanted to do and feel that specific ache of recognition: that could have been me.

So what separates doing from not doing? It’s not talent, resources, or even motivation, which is the most overrated force in the self-help universe.

As I introspected, I realized that the gap between action and inaction is almost always fear.

Fear doesn’t announce itself

Fear rarely shows up as “I’m afraid.” This surprised me. A lot. Fear shows up as procrastination. As perfectionism. As staying busy with low-stakes tasks so you never have to confront the high-stakes ones. As picking fights with your partner the week before a big opportunity. As saying yes to everything so you never have to commit to the one thing that matters.

These are protection mechanisms. Your brain identified something threatening: failure, rejection, loneliness, the vulnerability of wanting something and not getting it. And it built an elegant little system to keep you away from that threat.

The system works perfectly. The problem is that the threat it’s protecting you from is the same door you need to walk through to grow.

These are patterns where your brain links a situation to a past pain and then steers you away from anything that resembles it. You had a parent who was emotionally absent, so intimacy feels dangerous. You failed publicly once, so visibility feels like a trap. The original wound heals, but the avoidance pattern stays active, running in the background, often subconsciously.

The tricky part is that these patterns feel rational in the moment. You’re not cowering in a corner. You’re “being realistic.” You’re “waiting for the right time.” You’re “keeping your options open.”

Fear is fluent in the language of reason.

Subscribe now

The difference between fear and intuition

Not every impulse to avoid something is self-sabotage. Sometimes your gut is telling you something real. The key is learning to tell the difference.

Fear is loud. It makes predictions. It catastrophizes. It sounds like “this will definitely go wrong” or “everyone will judge you” or “you’re not ready.” It’s future-oriented, wrapped in certainty about outcomes you can’t possibly know.

Intuition is quiet. It deals in the present tense. It sounds like “this doesn’t feel right” or “I’m drawn to this” without needing to justify itself with a five-point argument. It doesn’t panic. It doesn’t generate adrenaline. It settles.

When you feel the urge to avoid something, pause. Ask yourself: is this a prediction about a future disaster, or is this a calm recognition of what’s true right now? The answer tells you whether you’re protecting yourself or holding yourself back.

In The Gifts of Imperfection, researcher and author Brene Brown frames it well: fear is the avoidance of uncertainty. Intuition is the willingness to hold space for uncertainty and act anyway, trusting the knowledge and experience you’ve already accumulated. The difference isn’t what you feel. It’s where the feeling is pulling you: toward contraction, or toward growth.

The emotional skills gap

The reason you default to fear instead of intuition isn’t because you’re weak. It’s because somewhere along the way, you didn’t get the chance to develop certain emotional skills. Self-confidence and self-trust, sure, but also something more basic: the ability to sit with discomfort without immediately reaching for a numbing agent. Your phone, food, alcohol, another person’s approval.

Self-confidence and self-trust are as learnable as cooking or coding. They develop through practice, not revelation. You don’t wake up one day feeling confident. You make one small decision that scares you, survive it, and build from there.

Ryan Holiday argues in The Obstacle Is the Way that difficult situations aren’t detours from growth; they are the growth. Failure shakes you out of your default patterns, reveals options you couldn’t see before, and forces you to develop capacities you didn’t know you had. The trial-and-error process is the most effective learning mechanism humans have. Every time you avoid it, you’re choosing comfort over capability.

This reframing matters. If discomfort is evidence that something is wrong, you’ll avoid it forever. If discomfort is evidence that you’re expanding, you’ll learn to move toward it.

How to close the gap

You don’t need a complete life overhaul. You need clarity, honesty, and a willingness to act on what you already know.

Name the pattern. Look at the areas of your life where you’re stuck. What behavior is keeping you there? Procrastination? People-pleasing? Staying in situations that make you miserable because leaving feels scarier than staying? Write it down. Be specific. “I avoid applying for jobs I want because I’m afraid of getting rejected” is useful. “I need to be more motivated” is not.

Trace it to the root. Every avoidance behavior protects you from a specific fear. Every fear points to a specific emotional skill you haven’t fully developed. Fear of rejection points to shaky self-worth. Fear of commitment points to a lack of self-trust. Fear of being alone points to an inability to meet your own emotional needs. Jack Canfield recommends being precise here: don’t write “I’m afraid of being alone.” Write “I’m afraid of managing my emotions without someone else to lean on.” The specificity turns a vague dread into a solvable problem.

Build the skill, not the avoidance. Once you know which emotional skill is underdeveloped, practice it directly. If you lack self-trust, start making small decisions without consulting five people first. If you lack the ability to sit with discomfort, start sitting with discomfort in low-stakes situations: a cold shower, an awkward conversation, a day without your phone. The muscle develops through use.

Create rules for yourself. Not goals. Rules. Goals are aspirational. Rules are operational. A goal says “I want to travel more.” A rule says “I transfer $200 to my travel fund on the first of every month, and I don’t touch it for anything else.” Rules remove the need to make decisions in the moment, which is when fear is loudest and your judgment is weakest.

Let your emotions inform you, not control you. Anger tells you what you care about. Jealousy tells you what you want. Regret tells you what you missed. These aren’t problems to suppress; they’re data. The next time you feel a strong emotion, don’t react to it. Interpret it. Ask: what is this telling me about what I need? Then act on the information, not the impulse.

The uncomfortable truth

Nobody is coming to rescue you from the gap between who you are and who you could be. No book, no podcast, no coach, no revelation at 3 a.m. will do the work for you. The information you need is already in your hands. It’s been in your hands for years. The question was never “how do I change?” The question is “why am I choosing not to?”

And once you’re honest about the answer, the path forward gets simpler. Not easier. Simpler.

You already know what to do. Start doing it. Start small, start scared, start imperfectly. But start. Because the distance between where you are and where you want to be isn’t measured in miles or years. It’s measured in the decisions you keep postponing.

Make one today.

Every January, millions of people adopt the same handful of habits: wake up at 5 a.m., meditate, journal, cold plunge, meal prep on Sundays. By February, most have quit. The problem isn’t laziness. The problem is that these habits were designed for a different person’s brain, body, and life.

The self-improvement world treats habits like universal prescriptions: follow these steps and you’ll transform. But you’re not a generic machine. You’re a specific person with specific wiring. And the fastest path to lasting change is understanding that wiring before you try to rewire it.

The advice trap

When a successful person shares their morning routine, they’re sharing what works for them. A night owl CEO who credits her success to 5 a.m. wake-ups is telling you about a correlation in her life, not handing you a universal law.

The same goes for me, or any influencer or creator or guru giving you a framework. Frameworks are starting points. They become useful only after you’ve filtered them through self-knowledge.

You have to do the prerequisite work that makes advice land. And that work is knowing yourself well enough to choose the right advice from the buffet.

You respond to expectations differently than I do

One of the most useful lenses for understanding yourself is noticing how you respond to expectations, both external ones (deadlines, social commitments, a boss’s request) and internal ones (your own goals, resolutions, promises to yourself).

People vary wildly here. Some are naturally disciplined and self-directed, but risk becoming rigid. Others won’t commit to anything until they understand the reasoning behind it; they need evidence, not instructions. Some are great at showing up for other people but terrible at showing up for themselves; they need external accountability to bridge the gap. And some resist expectations entirely; they’ll only stick with a habit if it feels like their own idea, not someone else’s mandate.

These aren’t personality flaws. They’re deep patterns in how you relate to rules, commitments, and motivation. And you can’t build the right habits until you understand which pattern is yours.

If you’re someone who struggles with internal commitments and you’re trying to build a solo meditation practice with no accountability structure, you’re fighting your own wiring. If you’re someone who resists being told what to do and you’re following a rigid 90-day program because some influencer said you should, you’ll quit by week three. Not because you’re weak, but because the approach doesn’t fit.

Your ingrained traits aren’t problems

Beyond how you handle expectations, you have deep traits that shape which habits will stick.

Are you a morning person or a night person? If you come alive after 9 p.m., building your most important habit around a 5 a.m. alarm is a losing strategy. A large study from Imperial College London found that night owls scored higher on cognitive tests measuring intelligence, reasoning, and memory. Work with your chronotype, not against it.

Do you prefer slow-and-steady or intense sprints? Some people chip away at projects a little each day. Others need the pressure of a deadline to produce their best work. Neither approach is superior. But trying to force yourself into the wrong one creates friction that kills momentum.

Do you crave novelty or consistency? If routine bores you, a habit that looks the same every day will feel like a prison. You might do better rotating between different forms of exercise each month, or varying your creative practice, rather than grinding through the same thing forever.

The point isn’t to use these traits as excuses. It’s to design your habits around your actual personality instead of an idealized version of yourself.

Subscribe now

Frame habits as things you’ll do, not things you’ll stop doing

Here’s a finding worth paying attention to: a large-scale study tracking over 1,000 people’s New Year’s resolutions found that people with approach-oriented goals (”I will exercise three times a week”) were significantly more successful than those with avoidance-oriented goals (”I will stop eating junk food”). At the one-year mark, 59% of the approach group considered themselves successful, compared to 47% of the avoidance group.

This matters because so much habit advice is framed around stopping, quitting, cutting back. “Stop scrolling. Quit sugar. Cut out distractions.” These are avoidance goals. They define success as the absence of something, which gives your brain nothing to move toward.

Instead, reframe: “I’ll read for 20 minutes before bed” beats “I’ll stop looking at my phone at night.” “I’ll cook dinner three nights a week” beats “I’ll stop ordering takeout.” Same intent, different framing, better outcomes.

Start with awareness, not ambition

Before you adopt a single new habit, track how you’re spending your time right now. Not to shame yourself but to see clearly.

James Clear makes a related point in Atomic Habits: your current habits form chains, where the end of one behavior naturally cues the beginning of another. Mapping these chains reveals opportunities. You don’t need to overhaul your life. You need to find the seams where a new habit can slip in.

Tracking also reframes progress. When you can look back and see that you stuck to your habit 22 out of 30 days, the 8 missed days stop feeling like failure. They become context as you build momentum, not failing.

People who value improvement over perfection are more resilient after setbacks. I’ve seen this in my own life, and the research backs it up. They keep going because they measure themselves against where they started, not against some flawless ideal.

The habits that make other habits easier

If you’re going to take one piece of prescriptive advice from here, make it this: fix the foundations first.

Sleep, movement, nutrition, and physical order (or whatever version of “order” your brain needs to function). These aren’t glamorous habits. Nobody’s posting their consistent 10 p.m. bedtime on Instagram. But they create the energy and willpower that make every other habit possible.

In The Power of Habit, Charles Duhigg calls these “keystone habits“: change one core habit and it creates a ripple effect. Not because it magically transforms your life, but because the reward of one positive change motivates you to make another.

The specific foundation habit that matters most varies by person. For some, it’s sleep. For others, it’s a daily walk, or clearing clutter from their workspace, or cutting back on alcohol. You’ll know which one it is because fixing it will make everything else feel slightly easier. My foundation is running more (and more).

Quitting a bad habit isn’t a simple matter of willpower. Many bad habits are responses to anxiety or discomfort. You scroll your phone not because you love social media but because you’re avoiding an uncomfortable feeling. You eat chips not because you’re hungry but because you’re stressed.

Instead of white-knuckling your way through cravings, get curious about them. Notice what you actually feel when you indulge in the habit. Often, you’ll find it’s not even satisfying. You’re eating mindlessly, scrolling mindlessly, drinking mindlessly. The habit isn’t delivering what it promises.

This awareness alone can loosen a habit’s grip. Not overnight. But consistently.

And it helps to know your own pattern with moderation. Some people do well with “everything in moderation”; knowing they can have a treat on Saturday keeps them disciplined all week. Others find that one cookie becomes the whole box, and they’re better off cutting the habit entirely. Neither approach is more virtuous. The right one is the one that matches your brain.

Prepare for failure (because it’s coming)

You will slip. The question isn’t whether; it’s what you do next.

The worst response is shame. Research consistently shows that self-compassion after a setback predicts better recovery. People who beat themselves up tend to spiral: they feel terrible, so they seek comfort, which often means returning to the bad habit. People who treat themselves the way they’d treat a friend pick themselves up faster.

The second-worst response is all-or-nothing thinking. You ate a cookie, so you might as well eat ten. You missed a workout, so the whole week is ruined. This is self-sabotage disguised as logic.

A better approach: design three versions of every habit. Your ideal version for high-energy days. A medium version for normal days. And a minimum version for the days when you’re exhausted, sick, or overwhelmed. On your worst day, doing the minimum version keeps the chain alive. And keeping the chain alive matters more than any single peak performance.

That same New Year’s resolution study found something else worth noting: the group that received some support (a named accountability partner and monthly check-ins) outperformed both the group that received no support and the group that received the most support. The researchers suspect that too much structure (SMART goals, interim deadlines, detailed exercises) actually backfired by creating more opportunities to feel like a failure. Sometimes a light touch works better than a heavy program.

By the way, you also have to watch out for your own excuses, as your brain is creative when it wants to let you off the hook. Some greatest hits:

• “I’ll start Monday.” (You won’t.)

• “I can’t because of [weather/travel/work/other person].” (You have more control than you’re admitting.)

• “I’m skipping it this one time.” (Maintaining the streak matters as much as the habit itself.)

• “Life’s too short not to enjoy yourself.” (Good habits are how you enjoy yourself long-term.)

Your brain perceives minor discomfort (boredom, effort, social awkwardness) as danger, and manufactures excuses to help you escape. Recognizing these excuses as automatic defense mechanisms, not rational arguments, takes away their power.

You don’t have to fight the excuse. You have to see it for what it is. Label it (”that’s my brain trying to protect me from discomfort”), and then act anyway.

What to do with advice (including mine)

So here’s my framework for taking advice, including everything in this article:

Filter it through self-knowledge. Does this match how you respond to expectations? Does it work with your energy patterns, your temperament, your life?

Test it as an experiment, not a commitment. Try a habit for two weeks. If it creates friction that feels productive (the discomfort of growth), keep going. If it creates friction that feels pointless (fighting your own wiring), adjust.

Steal the principle, customize the practice. The principle behind “wake up at 5 a.m.” is “protect time for your priorities before the world makes demands on you.” If you’re a night owl, that might mean blocking off 10 p.m. to midnight for deep work. Same principle, different execution.

Frame it as approach, not avoidance. Instead of “stop doing X,” find the positive behavior you’re moving toward. Your brain needs a destination, not a prohibition.

Stop the moment it becomes performance. If you’re maintaining a habit because it looks good on social media or because a guru said you should, but it’s not making your life better, drop it. Habits exist to serve you, not the other way around.

The people who sustain real change aren’t the ones with the most discipline. They’re the ones who know themselves well enough to pick the right habits in the first place.

Start there.

That thought kept our species alive. The ones who tracked where they stood in the group, who noticed when someone else found a better food source or earned more respect, survived at higher rates than the ones who didn’t care. Social comparison is how humans survived.

And now we’ve plugged that mechanism into the most powerful comparison machine ever built.

The slot machine in your pocket

Your brain has a status detection system. It’s constantly scanning the people around you, reading faces, voices, body language, possessions, trying to figure out where you rank. For most of human history, “the people around you” meant your village, your extended family, your trade guild. A few dozen faces. A knowable world.

Social networks took that system and made thousands, millions, and even billions of people your neighbor.

Every scroll is a status check. Every photo of someone’s launch, someone’s vacation, someone’s promotion, someone’s engagement ring fires up the same neural circuitry that once helped you figure out whether you were pulling your weight in the tribe. But the tribe was 150 people, and you knew them. You knew the guy bragging about his hunt was compensating for something. You knew the woman with the best clay pots also had a terrible temper. Context kept comparison honest.

Instagram, TikTok, and LinkedIn don’t give you context. They give you the highlights, stripped of every unflattering frame, served at the speed your thumb can flick.

And your brain can’t tell the difference.

Tech companies engineer these feeds around intermittent positive reinforcement. Post a photo, and you might get a flood of likes, or you might get silence. That unpredictability is the same principle that makes slot machines addictive. So you keep pulling the lever. You keep refreshing the feed. You keep checking the notification badge.

Each time, you’re asking your status detection system the same question: Where do I stand?

Each time, it gives you a worse answer.

The anxiety loop

Your brain’s prefrontal cortex, the part responsible for planning and rational thought, is designed to predict future threats based on available information. When it has good data, this works. When it doesn’t, it starts improvising. Spinning up worst-case scenarios. Running what-if simulations with no grounding in reality.

Social media gives your prefrontal cortex data that’s curated to look flawless. An endless stream of best moments, best angles, best outcomes, which your planning brain interprets as evidence that everyone else is thriving and you’re falling behind. So it starts planning for the threat of being left behind. But there’s no specific threat to address, no bear to avoid, no concrete problem to solve. There’s only a vague sense that you’re not enough.

That’s anxiety. And anxiety forms a habit loop.

First, you feel a trigger: the vague unease of not measuring up. Then you perform a behavior: you scroll, looking for reassurance, or distraction, or something that makes you feel connected. Then you get a result: a brief dopamine hit, followed by more comparison, followed by more unease. Your brain logs this cycle and automates it. Within weeks, you’re opening Instagram without conscious intention, the way you reach for a light switch in a dark room.

But it gets worse. Because anxiety doesn’t stay contained. The unease you feel about your status online becomes a trigger for secondary habits: stress eating, procrastination, drinking, online shopping, picking fights in comment sections. Each secondary habit produces its own anxiety, which triggers more habits. The loops multiply. They entangle. And when you’re stressed enough, your prefrontal cortex goes offline entirely, leaving you with nothing but survival instincts: fight (rage-tweet), flee (binge Netflix), or freeze (lie on the couch refreshing the same three apps for four hours).

To be clear, the individuals at Instagram, TikTok, LinkedIn, and the rest almost certainly don’t want to amplify your anxiety. But these loops do result in their engagement metrics going up. And so we continue to build and refine these loops to their limits. Unfortunately, the reward system, both of social media hijacking our brain and of tech’s performance reviews, makes this result inevitable.

Subscribe now

Now add AI to the mix

Everything I’ve described so far happened with human-curated content. People chose their best photos, wrote their wittiest captions, constructed their most impressive narratives.

AI removes even that friction.

Generative AI tools can now produce flawless images, compose eloquent posts, and fabricate entire personas. The highlight reel used to at least reflect a version of someone’s real life; edited and polished, but recognizably theirs. Now the reel doesn’t need a life behind it at all.

This changes the comparison game in two ways.

First, the floor rises. When everyone has access to AI-assisted content creation, the baseline for “impressive” shifts upward. Your unfiltered, unenhanced life starts to look not just ordinary but substandard, measured against a standard no human life was ever meant to meet.

Second, the signals break. Your status detection system evolved to read human signals: the tremor in someone’s voice, the forced quality of a smile, the gap between what someone says and how they carry themselves. AI-generated content has no such tells. There’s no body language to read, no inconsistency to notice. Your brain keeps scanning for status information, keeps trying to figure out where it stands, but the data it’s processing is increasingly synthetic.

You’re playing a status game against opponents who don’t exist.

The performance of wisdom

Now here’s where I’m supposed to give you the solution.

And there are solutions, or at least promising directions. Practice mindfulness to interrupt the anxiety loop before it becomes automatic. Take deliberate breaks from your devices so your dopamine system can recalibrate. Invest in face-to-face relationships that satisfy your social needs in ways a notification never will. Fill your time with demanding, hands-on activities that produce the kind of deep satisfaction scrolling can’t touch.

That paragraph sounds great, doesn’t it? Clean. Authoritative. The kind of advice you’d highlight and save and never quite follow through on.

I need to be honest about something.

Writing an essay about the dangers of social media and the importance of mindful living is a performance. A status game where you earn prestige by demonstrating wisdom, by packaging insight into shareable paragraphs, by positioning yourself as the person who sees through the illusion while everyone else is trapped in it.

So let me tell you what the curated version of me would leave out.

I scroll through LinkedIn and read about the latest AI results and feel a knot form in my stomach. Not intellectual curiosity. Fear. Fear about the money I won’t earn, the technical achievements I haven’t produced, the widening gap between where I am and where I told myself I’d be by measures I used to hold close to my heart. I close the app and open it again later because I can’t help thinking about it.

I worry about the mistakes I’ve made as a father. As a son. As a person. Not in a poetic, growth-mindset way. In a 2 a.m., staring-at-the-ceiling way where the specifics replay and I can’t find the version or interpretation of the memory where I come out looking decent.

On some days, I fear death. Not as an abstraction, but as a physical anxiety that sits in my chest and asks what I’ve done with my life, and whether any of it was enough.

And I worry that writing these essays is its own form of dishonesty. That I exaggerate how fully I embody the ideas I describe. That the distance between who I am on this page and who I am at my worst is itself a kind of lie. You’re reading about escaping the comparison trap from someone who has checked his subscriber count this morning.

I’m not different. Knowing the trap doesn’t always keep you out of it.

The builders, creators, and researchers whose products and ideas fill this post have done extraordinary work mapping the machinery of anxiety, addiction, and status. But none of them have fully escaped that machinery in their own lives. The neuroscientist studying mindfulness still gets irritated in traffic. The psychiatrist writing about habit loops still reaches for comfort food after a bad day. The journalist documenting smartphone addiction still feels the phantom buzz of a phone in his pocket.

What helps anyway

So what do you do with that?

Not what the self-help version of this essay would tell you. Not a numbered list of habits that fix the problem if you stick with them for 30 days. The problem doesn’t get fixed. The wiring is the wiring.

But you can learn to live inside the environment without letting it define you.

You notice when you’re scrolling without purpose and you put the phone down. Not every time. But more often than you did last month. You catch yourself mid-comparison and remind yourself that you’re measuring your interior life against someone else’s exterior. The reminder won’t stick permanently. You’ll need to repeat it tomorrow.

You call instead of texting. You show up in person when you can. You create things with your hands. You walk without earbuds. You sit with boredom long enough to discover that it transforms into something else: an idea, a memory, a strange and useful feeling you’d have missed if you’d reached for your phone.

Some days you fail at all of this. That’s not a setback, but the human condition.

The companies building these platforms employ thousands of engineers whose job is to keep you engaged, with algorithms as the delivery mechanism and our loneliness as the business model. You didn’t choose this environment. But you’re in it. And the only honest response isn’t to pretend you can escape it. It’s to keep choosing, each day, how much of your attention you’re willing to hand over.

This is the practice.

These two facts are related.

Around the same time, I was sitting in a room learning Tibetan meditation from a teacher who spoke in long silences. I didn’t know what I was looking for, exactly. I knew I hadn’t found it yet.

The 80/20 principle

In 1896, the Italian economist Vilfredo Pareto noticed that roughly 80% of Italy’s land was owned by 20% of the population. The pattern kept showing up: 80% of effects from 20% of causes. The Pareto principle became one of the most widely cited ideas in business and economics.

A century later, Google formalized a version of it as a management philosophy. Engineers were encouraged to spend 80% of their time on core projects and 20% on self-directed experiments. The policy produced Gmail, Google News, and AdSense. It also produced thousands of projects that went nowhere, and that was fine. The 20% wasn’t supposed to have a perfect hit rate. It was supposed to create the conditions for unexpected breakthroughs.

I used my 20% time aggressively. My experimental folder was enormous. My lines of code, even as a junior engineer, exceeded those of distinguished engineers; not because I was a better programmer, but because I was running more experiments. Most of them produced nothing. A handful reshaped my career.

It took me years to realize I’d unconsciously applied the same architecture to the rest of my life. The brokerage account. The meditation practice. The pattern was always the same: a stable foundation taking up most of the space, with a laboratory running alongside it.

That architecture; a stable 80% with a laboratory running in the other 20%; shaped three habits I’ve built my life around. The habits aren’t brilliant or original. They compound. And underneath all three sits a single principle that keeps the whole thing from falling apart: judge the portfolio, not the position.

The three habits are:

1. Expand your knowledge

2. Master your decisions

3. Build and grow

Here’s how they work.

Expand your knowledge

The most valuable asset you own isn’t in your brokerage account or your retirement fund. It’s the mental model you carry into every decision. That model either appreciates through constant learning or depreciates through complacency. There’s no neutral zone.

The practitioners, engineers, and investors I admire most share a common trait: relentless intellectual curiosity across domains. The deep kind where you follow a thread for weeks, months, and even years because it interests you, with no immediate payoff in sight.

I learned this first in the most personal domain of my life: the search for a contemplative practice.

I grew up moving through phases of Christianity; Protestant, Methodist, charismatic. Each tradition gave me something I kept. None gave me a reason to stop looking. So I kept looking. I took Tibetan meditation classes, read the texts, sat with teachers who spoke more in silence than in words. Then I encountered Theravada Buddhism, and something shifted. The approach was more direct, more empirical. It resonated in a way the others hadn’t.

The years of spiritual exploration weren’t wasted. They were the research phase. Every tradition I moved through sharpened my sense of what I was looking for, even when (especially when) I realized it wasn’t what I’d found. The Tibetan practice wasn’t a dead end. It was calibration: I learned what stillness felt like, which is how I recognized it when Theravada offered a more direct path to it.

The same curiosity-driven approach shaped my technical career at Google. The company had a formal structure for it: 20% time. One day a week, you could work on something outside your core job. Something experimental.

Most engineers used their 20% sparingly. I used mine like it was oxygen.

My first 20% project started because I was curious about public transit data. Singapore had bus schedules in no format Google Maps could ingest. So I ported them: converted the raw schedule data into the standardized format, built a proof of concept, and showed it could work. A small project. Nobody asked me to do it. I did it because the problem was sitting there and I wanted to see if I could solve it.

Then I got curious about structured data inside wikis. I built a Wiki markup parser using lex and yacc, the classic compiler tools, because I wanted to extract semantic information from wiki-formatted text. That parser folded into my main project, using the then-nascent translation technology to improve cross-language information retrieval. The same approach I’d built for one wiki became the basis for extracting structured knowledge from Wikipedia at a Google-wide scale when someone found my codebase and built on top of it.

Then deep learning caught my attention. This was before the current AI wave, when neural networks were still a niche interest among most engineers. I studied the fundamentals, ran experiments, and applied what I learned to help build the first query-less feed ranking system in Google Search; a system that could surface relevant content without the user typing anything.

Not all of them worked. Plenty of experiments in that overflowing folder went nowhere. I was disappointed every time. But I wasn’t staking my career on any single experiment, so the disappointments were tuition, not trauma.

The same pattern drove my financial education. In 2010, I opened a brokerage account and started buying $500 lots of individual stocks; not because I had a thesis about beating the market, but because I wanted to understand how different businesses worked. I read Ethereum’s whitepaper before most people had heard the term “smart contract.” I deployed tens of thousands of dollars across hundreds of peer-to-peer loans on Prosper and LendingClub because I wanted to understand credit risk from the lender’s side, and applied machine learning to invest intelligently in high-credit-risk loans. I invested in early-stage startups through crowdfunding platforms because I wanted to see how venture math worked with my own money on the line.

Each experiment taught me something no textbook could. The P2P lending account, which wound down as defaults crept up and the industry folded, taught me the difference between yield and return. Crypto taught me that conviction and position sizing are two different skills. Startup investing taught me that portfolio-level math can work even when most individual bets fail.

The common thread across all three domains: deliberate exposure to things outside your core. Read beyond your field. Study adjacent disciplines. Spend time with people who’ve already achieved what you want to achieve.

The varied inputs create connections between seemingly unrelated concepts, and those connections are where the breakthroughs live. A wiki parser built from compiler tools. A meditation practice refined by years of comparative exploration. A machine learning approach to peer-to-peer lending. None of these would have happened inside a narrow lane.

Subscribe now

Master your decisions

Expanding your knowledge is necessary but not sufficient. The harder skill is knowing what to do with it: when to push an experiment forward, when to walk away, and how to avoid staking your identity on any single outcome.

I learned this most clearly in my brokerage account.

In 2012, I went through an options phase: selling puts on Amazon and Goldman Sachs, buying calls, collecting premium, feeling sophisticated. Then I realized options require monitoring, and monitoring requires time, and time was the one resource a full-time engineer didn’t have in surplus. I stopped. Not because options are bad, but because I recognized the mismatch between the strategy and my life circumstances. A decision about fit, not about the instrument.

I traded Sears like it was a day job; buying and selling a few hundred shares here, then a hundred more later. The company went bankrupt in 2018. Whatever I made on the round-trips, I spent more in attention and commissions than it was worth. The lesson was clean: trading isn’t investing, and the line between them blurs when you’re not paying attention.

But I held ADP, the payroll services company. Bought 40 shares in May 2010 at $38. Added more in 2012. And then I did nothing. Nothing changed about the business that made me want to sell. So I didn’t. Fifteen years later, one position bought with experiment money was worth more than every failed experiment combined, many times over.

The decision to hold ADP wasn’t brilliant analysis. It was the absence of a reason to act. And that’s its own kind of decision mastery: recognizing when the best move is no move at all.

The options taught me to match strategy to circumstances. Sears taught me to distinguish trading from investing. ADP taught me the power of inaction. Three positions, three different lessons about the same underlying skill: calibrating when to move and when to stay.

The same calibration mattered at Google. My 20% projects required a constant stream of judgment calls. The Wiki markup parser worked as a standalone tool for my team. Should I push to expand it Google-wide? That meant advocating to leadership, navigating organizational politics, and risking the project getting killed by people who didn’t share my enthusiasm.

The spelling correction project drew the most skepticism. Google’s web spelling system was built for short queries; a few words at a time. Extending it to full documents wasn’t an obvious move. The first reaction from senior engineers wasn’t “how would this work?” It was closer to “why would we even do this?” Google Docs already had basic spell-check. The web-scale approach seemed like overkill for a productivity tool.

But I’d built the prototype, and the prototype made the argument I couldn’t. The web-based approach caught errors conventional dictionaries missed; proper nouns, technical terms, emerging slang; because it drew on the same corpus powering Google Search. Once senior engineers saw it working, the skepticism dissolved fast. It wasn’t a hard sell once there was a demo. It made the product obviously better. The gap wasn’t between a good idea and a bad one. It was between an idea that sounded unnecessary in the abstract and one that looked inevitable in a demo.

That experience taught me something about conviction: it’s not enough to believe you’re right. You need to build the thing that shows you’re right. Persuasion follows demonstration.

The meditation journey required the same calibrating skill. Moving from Tibetan practice to Theravada wasn’t a failure of commitment. It was an honest assessment that one approach resonated more deeply than another. There’s a version of spiritual seeking that treats every change as evidence of flightiness. I see it differently. Each transition was a decision informed by direct experience, made without guilt or identity crisis.

The connecting thread: good decisions require detaching your ego from your experiments. Uncertainty is the default condition, not an obstacle to overcome. Think rationally about probabilities rather than emotionally about outcomes. Reflect on past decisions to extract patterns, not to berate yourself for the ones that didn’t land. And above all, don’t confuse the decision with the result. A good process can produce bad outcomes, and a bad process can get lucky. Judge the process.

Build and grow

Knowledge and good decisions matter because of what they enable: compounding. Not financial compounding alone, though that’s the most visible kind. Compounding of skills. Compounding of relationships. Compounding of inner stability.

The principle is the same across every domain. Make consistent deposits into a foundation. Protect that foundation from catastrophic loss. Give it time. The returns arrive slowly, then all at once. It’s little wonder compound interest is the eighth wonder of the world.

In my career, each 20% project added to my toolkit in ways I didn’t fully appreciate at the time. Porting bus schedules taught me to build proofs of concept fast. The wiki parser taught me tools built for one context can be repurposed at scale. The spelling correction project taught me how to stretch a working system into a new domain, and that a working demo is worth a thousand arguments. The deep learning work taught me to place a bet on an emerging technology early, before the organization was ready to believe in it.

Individually, none of that sounds remarkable. But skills stack. A decade in, I wasn’t a specialist in any one of those areas. I was the person who’d done all of them, and that turned out to be more useful than being the best at any single one. The experimental folder wasn’t waste. It was compound interest on knowledge, paid out over years.

The financial compounding is easier to quantify. In 2010, I started feeding my 401(k) through Vanguard: total market index funds, international stocks, bonds, REITs. Then I did the hardest thing in investing. I stopped looking at it. Over the next decade, the account grew roughly 10x. Part of that was market performance; the S&P 500 roughly quadrupled over the same period with dividends reinvested. The rest was consistent contributions, compounding on top of compounding. I didn’t rebalance obsessively. I didn’t panic sell during COVID. I didn’t try to time anything. The returns came from three forces: consistent contributions, compound growth, and the stubbornness of not touching it.

That boring foundation made everything else possible. When 80% of your money is growing steadily, the remaining 20% becomes a sandbox where you can afford to take real risks, because the downside is capped by design.

The experimental brokerage account produced its own compounding story. ADP grew roughly a hundredfold over fifteen years. The magic was time and compounding, not brilliance. Meanwhile, the startup investments I made through crowdfunding followed venture math: most failed completely, but the winners more than compensated for the losers, and the overall multiple landed well above what I’d have earned in an index fund over the same period.

The meditation practice compounded too, though the returns are harder to measure on a statement. Years of exploration; moving through Christian denominations, studying Tibetan techniques, reading contemplative texts; distilled into a Theravada practice that became the non-negotiable foundation of my daily life. The exploration phase was the research. The daily sitting is the compounding. Each year of consistent practice builds on the last in ways that are subtle but unmistakable: steadier attention, less reactivity, a quieter relationship with my own mind.

And the principle that protects compounding is the same across all three domains: avoid catastrophic loss. In finance, this means living within your means, avoiding excessive debt, and keeping cash reserves so you never have to sell investments at the worst possible time. In a career, it means delivering on your core responsibilities before running experiments; the 80% has to be solid. In a contemplative practice, it means protecting the daily habit from the chaos of life, making it non-negotiable even when everything else is in flux.

You can’t compound what you don’t protect.

What I’d tell someone starting out

Secure the base first. In your career, be excellent at your core job before you start experimenting. In your finances, automate your savings and build a boring diversified foundation before you open a brokerage account. In your inner life, establish a daily practice before you chase peak experiences.

Size your experiments to learn, not to get rich. My early stock trades were $500 at a time. My early 20% projects were small proofs of concept. My early meditation experiments were evening classes, not retreats. Small enough that a total loss is tuition, not trauma.

Diversify your experiments, not your positions alone. I didn’t put my 20% into one area. In finance, I spread it across stocks, P2P lending, crypto, and startups. In my career, I explored data conversion, natural language processing, machine learning, and product advocacy. In contemplative practice, I moved through Christian traditions, Tibetan Buddhism, and Theravada. Each experiment taught me something different about risk, resonance, and my own psychology.

Keep notes. I have brokerage statements going back to 2010. In Google’s monorepo environment, the code from 20% projects is all saved. I have handwritten notes and journals from meditation classes and retreats. When I look at these records, I’m not nostalgic. I’m reading lab notebooks. Every entry tells me something about what I believed at the time, what I got right, and what I’d do differently.

And above all: judge the portfolio, not the position.

We’re wired to feel losses more intensely than gains. Behavioral economists call it loss aversion. A $110 loss on Converted Organics feels like a failure when you stare at it in isolation. A 20% project killed after three months feels like wasted effort. A meditation tradition you leave behind can feel like a spiritual dead end.

But zoom out. Check the aggregate. If the portfolio is growing; if your career is advancing, your net worth is compounding, your inner life is deepening; then the individual losses are doing their job. They’re the cost of exploration. They’re tuition.

The venture capital industry understood this decades ago. VCs expect most investments to fail. They don’t call those failures. They call them the cost of finding the winners. VCs invest other people’s money. I applied the same logic to my own life, with the safety net of a boring foundation underneath it all.

The experiments work only because the 80% exists. If you’re not building a solid career foundation before running side projects, the experiments are reckless. If you’re not maxing your tax-advantaged retirement accounts before trading individual stocks, the brokerage account is gambling. If you don’t have a consistent daily practice, the exploration is spiritual tourism.

The 80% keeps you solvent, employed, and grounded. The 20% keeps you curious, engaged, and growing. Without the 80%, the experiments are dangerous. Without the 20%, the foundation becomes a cage.

You need both.

The best investment framework isn’t the one that maximizes returns. The best career strategy isn’t the one that eliminates risk. The best contemplative practice isn’t the one that promises instant transformation. The best framework is the one you’ll stick with. For me, that’s been a boring foundation with a laboratory on top.

I still keep the brokerage statements, old code, handwritten notes from meditation classes. Not out of nostalgia. Every entry reminds me what I believed at the time, what I got right, and what I’d do differently. That’s the whole practice, really. Run the experiment. Write it down. Run the next one.

That’s the pattern I keep noticing across work, rest, and everything in between. Most of our dissatisfaction comes from fighting our own limitations instead of working with them. We fight what we have, how much we can do, how hard it feels, and how uncertain it all is. And the fighting is what exhausts us, not the living.

I’ve been circling this theme for a while. Not new ideas, exactly. More like ideas I already believed but hadn’t articulated together. Here are five worth keeping in mind.

Work with what you have, not what you wish you had

Life isn’t an optimization problem. We assume there’s one right path and our job is to find it, when in reality there are dozens of viable paths and the real skill is picking one and moving.

Most people fail to act not because they lack ability, but because they’re focused on the wrong version of their life. The imagined version. The one where they have more time, more clarity, more resources.

This is the trap I fell into for years in my career and life. I kept waiting for the right conditions before making changes. The right financial cushion. The right next step. The right moment of certainty. But certainty comes from movement, not contemplation.

The solution is simple: look at your life as it is. What skills do you have now? What time is available now? What’s within reach now? Then narrow your focus. Drop obligations draining you. Stop consuming information you’ll never use. Pick one or two things and give them your real attention.

This honest assessment is harder than it sounds. Our brains are wired to construct stories instead of examining facts, to overweight what’s loud and underweight what’s true. We overestimate our abilities in areas where we’re weak and inflate the value of what we already have. Seeing your life clearly takes deliberate effort.

This is removal, not addition. Flow comes from setting things down, not picking more up. Your limitations aren’t obstacles. They’re the edges giving your work shape.

Movement beats preparation

There’s a version of productivity that’s all planning and no doing. Designing systems. Organizing tools. Optimizing workflows. It feels like progress because it is effortful. But effort without output is a way of hiding. Most projects die in the comfortable middle of “maybe” because we confuse preparation with progress.

Planning is not action. The person who writes 200 rough words today is further ahead than the person who spent three hours perfecting their writing setup.

This connects to something I’ve noticed since stepping away from structured work. When nothing demands you, you discover how much of your “productivity” was motion without direction. Some days after I left, I’d spend hours preparing to do something and never do it. The preparation was a way of avoiding the vulnerability of producing something imperfect.

The antidote is embarrassingly small. Pick one goal. Break it into a piece boring enough to start without resistance. Do it today. Not after conditions improve. Today.

There are small nagging tasks we all carry around. The bill you haven’t paid, the email you keep dodging. Each one is trivial. Together, they create a low-grade hum of anxiety making everything feel heavier than it is. I started dedicating the first few minutes of my morning to clearing these, and the difference surprised me. Same principle as writing down what’s looping in your head before you start working. Free up the space so attention can go where it matters.

Humans can sustain deep focus for about three to four hours a day. Productivity researcher Anders Ericsson found this ceiling across top performers in various fields, and Cal Newport builds his own workdays around the same constraint. Everything after runs on fumes. I spent years pushing through long days and calling it discipline. It wasn’t discipline. It was a misunderstanding of how the mind works. Build around the constraint instead of pretending it doesn’t exist. Rest is as important as work.

Subscribe now

Most difficulty is self-inflicted

The difficulty of a task is rarely what stops us. The second arrow, the story we tell about the task, carries most of the resistance. Our anxiety and desire for perfection make our goals, problems, and barriers seem bigger than they are. The fix: pretend the goal is easy. Not because it is, but because the framing removes the psychological weight keeping you frozen.

This is where thinking harder makes things worse. Your brain craves narrative, so it invents complexity where there isn’t any. It latches onto whatever’s loudest, constructs stories from limited information, and gives your fears more weight than they deserve. You aren’t analyzing the task. You’re building a case for why it’s terrifying.

This is counterintuitive if you grew up associating effort with value, as I did. If something doesn’t feel punishing, it must not matter. But ease often follows understanding. Zone 2 running is easy and builds a strong aerobic base; the trick is to be consistent.

When a problem becomes clear, the path through it narrows. Fewer decisions, fewer things to hold in mind. What remains feels lighter, not because something important was skipped, but because unnecessary resistance fell away.

Perfectionism shows up here too. It’s procrastination in disguise. If you wait for the brilliant insight before starting, you’ll wait a long time. If you produce something flawed and keep going, you’ll produce something good. For software engineering, here’s my heuristic: separate building from polishing into different passes. First you make it work. Then you make it elegant. Trying to do both at once is how projects stall indefinitely. This trick works in many other facets of life.

Let go of control

We also need to let go, and this deserves its own section because it’s where I get stuck the most.

Letting go of the need to control other people’s emotions. I’ve killed ideas and held back honest conversations because I was managing someone else’s potential reaction. But you can’t control how people feel. You never could. Accounting for others matters; organizing your life around their imagined responses doesn’t.

Letting go of what isn’t working. The things draining your energy right now aren’t usually the things you haven’t started. They’re the things you haven’t ended. We avoid quitting because it feels like failure, but staying in the wrong situation isn’t persistence. It’s avoidance. Sometimes the most productive thing you can do is stop.

Letting go of the plan. There’s resonance in engaging with life as it comes, including the interruptions. Rolling with what shows up instead of white-knuckling your schedule. Flow isn’t something you force into existence. It’s what remains when you stop blocking it.

Letting go of the story. Your brain wants to narrate everything, assign meaning, predict outcomes. Sometimes the clearest thing you can do is stop the commentary and respond to what’s in front of you. Meditation taught me this: contentment isn’t something you achieve. It’s what’s already there when you stop adding noise.

Be here. This is the life.

I’ve spent large portions of my life treating the present as a rough draft. A rehearsal for the real version starting once conditions were right. Once I had more money. More freedom. More clarity about what I wanted.

I’ve written about this before. How stepping away from work didn’t produce a credits-ending montage. How the growth came not from the freedom itself but from learning how to meet ordinary days without rushing to fill them.

This is it. Right now. There is no future version of your life more real than the one you’re living.

If you want to be a writer, write today. If you want to be more present in your relationships, start tonight. If you want to be generous, act on the impulse when it arises instead of waiting for a grander opportunity. Your future self is built from what your present self does.

Let go of the desire for permanence. Stop trying to photograph every good moment to preserve it. Stop worrying about legacy. In the grand sweep of time, none of it echoes forever. And that’s not depressing. It’s freeing. When the pressure to leave a mark lifts, you’re free to do meaningful work for its own sake.

I’ve found this to be true in my own life since leaving Big Tech. I don’t have more time than before. But I have more attention. When I’m teaching my kid to make sourdough or running a trail in the fog, there’s no laptop open “just in case.” No phone buzzing with escalations. The activity gets my full presence. That’s the dividend of being here.

Happiness, meaning, and purpose aren’t states you achieve and hold. They’re rhythms you build. The person waiting for a breakthrough misses the small openings already in front of them.

The small, quiet differences you make in your immediate world matter. The friend you checked in on. The page you wrote. The morning you met without rushing to fill.

Those count. They’re enough.

Share

I didn’t grind through feature engineering tutorials. I didn’t spend years building intuition for gradient boosting. I built an agentic workflow that handles experiments and iterations for me. The AI does the data science. I do the steering.

This isn’t a story about how I’m clever. It’s a story about what happens when the floor rises.

What the AI did, and what I did

Let me be concrete. The cheating detection competition involved tabular data and a social graph of user connections; the task was predicting which users were cheaters. My final solution used 59 engineered features, 13 diverse models, and a stacking pipeline with label propagation and isotonic calibration.

Here’s what the AI handled: writing the feature engineering code, implementing SVD embeddings on the adjacency matrix, tuning hyperparameters via Optuna, setting up the neural network architectures (FT-Transformer, TabPFN, TabM), managing GPU instances, and preventing runaway cloud costs. Thousands of lines of code I never wrote.

Here’s what I did: I noticed that missing values were the signal after the AI built visualizations - users with certain features missing were cheaters about 98% of the time. I caught a leakage bug when computing neighbor statistics as cross-validation scores looked amazing but the leaderboard told a different story. I decided to mix algorithm families for ensemble diversity. I chose how conservative or aggressive parameters for the various algorithms should be based on intuition and a few experiments.

Implementation belonged to the machine while key insights and judgment belonged to me.

I built this system around four principles:

Guardrails. LLMs will happily wander off-task. They’ll optimize the wrong metric, chase interesting tangents, or spend three hours debugging a visualization nobody asked for. Scope the task tightly. My agent knows it’s doing Kaggle competitions, not general data science consulting. That constraint isn’t limiting; it’s focusing.

Skills. Generic prompting produces generic results. I wrote domain-specific knowledge documents: when to use target encoding versus label encoding, how to detect leakage in cross-validation, what hyperparameters work well for LightGBM on tabular data. The agent doesn’t memorize these. It retrieves them when relevant. Exploratory data analysis tips surface at the beginning. Hyperparameter guidance surfaces during training. Just-in-time knowledge, not a firehose of text and a waste of context.

Sandbox. An agent that can’t execute is an agent that hallucinates. The workflow can launch GPU instances, run experiments, manage cloud costs, and shut down resources when they’re not needed. The execution environment is as important as the reasoning. Without it, you’re paying for suggestions instead of results.

Memory. Each experiment builds on the last. The agent tracks what it’s tried, what worked, what failed. Without memory, you get an LLM that rediscovers the same dead ends repeatedly. With memory, you get compounding progress.

The key insight here is that knowledge documents are meta-code - instructions for writing instructions. Karpathy said English is the new programming language. What he didn’t say is that the grammar is constraint design and the vocabulary is domain expertise.

I don’t tell the agent “use target encoding on feature_007.” I teach it when target encoding is appropriate. The agent figures out feature_007. I notice that missing values are the signal in the first place.

This inverts the traditional programming relationship. I’m not writing code that solves problems. I’m writing knowledge that teaches a system how to write code that solves problems. One layer up. The leverage is enormous.

The scary part isn’t automation. It’s standardization.

When people worry about AI taking jobs, they imagine robots that think like humans. That’s the wrong frame. Job extinction has never required human-level intelligence. It requires something simpler: standardization and more supply of the core output.

The pattern repeats across industries and centuries:

1. A skilled task produces a repeatable output

2. Technology standardizes that output

3. Cost collapses and speed explodes

4. Employment contracts into a narrow premium niche

5. Survivors move up-stack into design, supervision, or systems thinking

Before electronic computers, “computer“ was a job title. Rooms full of people calculated ballistics tables, astronomical data, and engineering figures by hand. At NASA’s predecessor, African American women mathematicians (watch Hidden Figures if you haven’t yet) worked in segregated computing pools, doing the math that guided early spaceflight. They didn’t vanish because machines could think. They vanished because arithmetic became cheap. The work still existed but the workers became unnecessary.

Switchboard operators manually connected every phone call until direct dialing made them obsolete. Typesetters spent years mastering metal letterforms until Linotype, then phototypesetting, then desktop publishing compressed the craft into nothing. Elevator operators were considered essential for safety until push-button controls proved otherwise; their disappearance is now the textbook example of technological unemployment.

The Jacquard loom automated weaving patterns with punched cards in the 1800s. Skilled artisans who once created complex designs by hand found themselves tending machines or out of work entirely. The Luddites weren’t anti-technology. They were anti-unemployment.

Different industries. Different centuries. Same arc. None of these transitions required general intelligence. They required reliable automation of the core task.

My Kaggle results are a small data point in a larger story: the core task of data science is becoming standardized. So are other parts of the tech industry. Starting with coding.

Coding: the anchor case

Software engineering has always been layered.

At the bottom: syntax, APIs, boilerplate.

In the middle: architecture, data flow, abstractions.

At the top: strategy, problem framing, tradeoffs, accountability.

AI is collapsing the bottom layer.

The early data is messy. A randomized controlled trial from METR found that experienced open-source developers were 19% slower with AI tools while believing they were 20% faster. GitClear’s analysis of 153 million lines of code shows code cloning up 4x since AI assistants became widespread. Technical debt may be accumulating faster than we realize.

But the picture is more complex than “AI doesn’t help.” Anthropic’s internal research on their own engineers found the opposite: people delegating increasingly complex work to Claude, requiring less oversight over time, and reporting significant productivity gains. The difference may be in how work is structured. The METR developers were experienced with their codebases but had only moderate AI experience. They were using new tools on old workflows. Anthropic’s engineers have rebuilt their workflows around what AI is good at.

This is the pattern with every tool adoption. Productivity gains lag capability gains. The question isn’t whether AI helps today’s developers on today’s code. It’s whether tomorrow’s developers, building AI-native systems from scratch, will need the same headcount.

I’ve seen the other side. My Kaggle workflow isn’t faster because the AI writes better code. It’s faster because I’ve restructured the work around what AI is good at. The implementation layer became cheap. So I stopped treating implementation as my job.

There’s another pattern from history: pushback. The telephone operators’ union fought direct dialing. Typesetters resisted phototypesetting. Workers who’d spent years building expertise don’t surrender it gracefully, and sometimes their resistance buys time for adaptation. But it never reverses the trajectory.

The profession won’t vanish but the median role will. This suggests way fewer people writing raw code, but there will be more value placed on system design and ownership. Iteration cycles will become faster and there will be higher expectations per contributor.

Radiology: the canary that didn’t die

Radiology was supposed to be the proof case. In 2016, Geoffrey Hinton declared that “people should stop training radiologists now.” Digital images, clear benchmarks, repeatable pattern recognition tasks. If AI was going to replace knowledge workers anywhere, it would start here.

Nine years later, American diagnostic radiology residency programs offered a record 1,208 positions in 2025, up 4% from the previous year. Vacancy rates are at all-time highs. Radiologists are the second-highest-paid medical specialty in the country, with average income of $520,000, up 48% from 2015.

Did the predictions fail?

Not exactly. AI did automate significant chunks of the work. Over 700 radiology AI tools have cleared FDA approval. Systems now detect pneumonia, identify nodules, spot strokes, and flag breast cancer, often matching or exceeding human performance on narrow benchmarks.

But this is a case of Jevons paradox: when efficiency improvements increase total consumption rather than decrease it. Cheaper, faster imaging meant more imaging. An aging population, expanding screening guidelines, and the sheer availability of AI-assisted reads drove demand upward. Radiologists didn’t disappear. They moved up-stack. Less time scanning for nodules. More time on complex integration, clinical judgment, interventional procedures, and the cases AI flags as uncertain.

One study found radiologists spend only 36% of their time on direct image interpretation. The rest is consultation, integration, supervision. AI took over a task; humans moved to the layer above.

The lesson isn’t “AI won’t displace workers.” The lesson is that displacement and adaptation happen simultaneously. The workers who survive are the ones who move before the floor finishes rising. Radiology salaries are up because radiologists adapted. They moved up-stack before the stack collapsed beneath them.

The application gap

Here’s what I notice about the current AI discourse: everyone’s debating whether the next model will be transformative. Meanwhile, current models are radically underdeployed.

Frontier model progress may be slowing. GPT-5’s release was described as “evolutionary rather than revolutionary” by the Financial Times. The benchmark gaps are shrinking; on MMLU, GPT-3 scored 44%, GPT-4 jumped to 74%, but GPT-5 only reached around 87%, with smaller gains each generation. Meta’s Francois Fleuret declared: “We are seeing the plateau: scaling up is coming to an end.” Inside labs, the consensus is growing that adding more data and compute will not create the breakthroughs once promised.

But this is almost beside the point. We haven’t figured out how to use what already exists.

Less than 10% of firms in the overall economy report using AI regularly. The number rises to only about 20% in professional, scientific, and technical industries. The Yale Budget Lab found no discernible disruption in the broader labor market 33 months after ChatGPT’s release. Historically, widespread technological disruption takes decades, not months.

And this is where the opportunity sits. Claude Code isn’t a breakthrough model but it is a breakthrough application: an agentic wrapper around existing capabilities that restructures how coding work gets done. The people building these wrappers aren’t behind the curve. They’re building the infrastructure the rest of the economy will eventually run on.

Being a “GPT wrapper” isn’t shameful. As it turns out, with products like Cursor and Claude Code, it is one area where value is getting created. The frontier labs build capabilities. The application layer builds products. I hypothesize that there will be more economic impact that will come from the second group, not the first.

What moving up the stack means

I should be clear about what I don’t know.

There may be a phase transition coming. A point where automation stops respecting professional and societal boundaries, where moving up-stack stops working because there’s no stack left to move to. If general reasoning becomes cheap enough, the pattern I’ve described breaks down entirely. This becomes a societal question, not a career one.

I don’t know if that’s one or five or fifty years away or never. Neither does anyone else, despite their confidence.

But until that moment arrives, the historical lesson holds: progress doesn’t pause for philosophical certainty. The people who wait for clarity get run over by the people who build through ambiguity.

I don’t have a roadmap. I’m figuring it out myself. But I notice a pattern in who thrives during transitions.

It’s not the people with the best predictions about where things are going. It’s the people willing to build while uncertain. The ones who ship before they’re sure. The ones who treat “I don’t know” as a reason to experiment, not a reason to wait.

When you can have anything built, the bottleneck isn’t capability. It’s imagination. It’s curiosity. It’s the willingness to be wrong in public while you figure out what works.

What do you do at this point? Create new products powered by AI, or learn to be 10x with AI? Do you want to be a shovel maker or do more shoveling? I can’t answer that for you. It depends on your interests, your skills, your appetite for different kinds of risk. Nobody knows which specific skills will matter in five years. But both the shovels and what we’re digging for will look different than they do now.

Disposition matters. The future belongs to the audacious, the curious, and the people who can’t help but build things. When the floor is rising, the worst thing you can do is stand still and hope it stops.

The question isn’t whether AI will change your work. The question is: at which layer do you choose to remain valuable?

History favors those who answer early.

If you’re building in the space of data science agentic workflows, I’d like to hear from you. Find ways to contact me (there are lots). I’m interested to know what you believe is worth building and why. Let’s exchange notes.