Built by Hand

The author of The Shape of the System lives on a small farm in rural Lithuania. We talked about forgetting, about why the manifesto is really a set of handrails he built for himself, and about the life behind it. Lightly edited.

Let's start somewhere honest: what did you personally break?

The first one is in the foreword, so I won't go over it again, except to say it involved a mail server and it was twenty-five years ago and it still turns up at odd moments. But the truer answer is smaller and it happens every day. I lose my own work all the time. I forget to commit, so things sit in worktrees I can't ever find again. Repositories that never got pushed. Folders on the desktop that end up buried inside the next "New Folder", which is inside the next one, all the way down. This isn't a beginner I'm describing. I've been doing it for years. The manifesto isn't me lecturing other people about discipline. It's me building handrails for myself, because I already know exactly how I fall.

So "engineering for bounded cognition" isn't an abstraction for you.

No. It's autobiography with the names changed. I have ADHD. I was diagnosed a few years ago, and the treatment that actually works only reached me recently, and that's a big part of why all this writing has shown up at once. There's a harder piece too, which I'll say once and not make a thing of: I don't remember most of my life. Not just childhood. Near enough the whole of it, forty-odd years, just gone, and I'm told that's part of the same wiring. When your own memory doesn't keep the receipts, you stop trusting any design that asks a person to remember the important thing at the important moment. I don't build for forgetting because it's a clever principle. I build for it because I am the forgetting.

The title is The Shape of the System, and you chose to write it as a numbered manifesto. Why those two things?

Take the shape first. In the end it's the shape that wins, not the people. Every team has good engineers who care. What actually settles whether the thing falls over on a Friday is duller than that - what can reach what, where the boundaries sit, what happens by default when nobody's watching. You can be the sharpest person in the building and the shape will still beat you, because it's working the whole time and you aren't. The manifesto bit - I wanted to be pinned to something. A book argues you round gently and you put it down agreeing with the mood and none of the specifics. Twenty-five numbered claims are something you can jab a finger at and say no, not here, not for this. I'd rather you disagree with one of them precisely than nod along to all of it vaguely. Numbering it was me committing, in public, to things that might turn out wrong.

The series keeps describing minds that are "tired, distractible, ordinary, and now partly machine." What does "partly machine" mean?

The language model is just the newest bounded mind at the table. It has a context window the way you have a working memory, and it falls off the end of it the same way you do, quietly, without telling you it's stopped holding the earlier thing. The part I find almost funny is that everything we learned writing software for a distractible human turns out to be exactly what a model with drifting attention needs too. Make the dependencies explicit. Keep the relevant thing close to where it gets used. Don't rely on anyone, carbon or silicon, to remember the bit you mentioned four screens back. The audience got bigger. The design didn't have to change at all.

There's one move that repeats across nearly every tenet: the fix is never "be more careful."

Because carefulness is a battery, and it's always lower than you think, and it's lowest exactly when the stakes are highest. The whole book is really one idea said twenty-five ways. Find the thing you're quietly relying on a human to remember, and change the shape so they can't forget it, because the forgetting was never a character flaw you can train out of people. It's the specification. People forget. Build as though that's a law, because it is.

Who are you when you're not at a keyboard?

Ten years ago my wife and I drove around Europe in an old military truck, working as we went. Somewhere in the middle of that we decided to stop driving and start building something, and we found a small farm in Lithuania. Most of what's here we've put up with our own hands. There's a sauna we built ourselves, and a pottery she built. I make hay for the horses, who are honestly more lawn ornaments than working animals at this point, and my favourite stretch of the year is the bit where I'm on the tractor and the weather is, for once, not actively against me. It's a deliberately low-tech life, and I think that's the point. When you heat your own home by hand, resilience stops being a word in a slide deck.

Does the farm teach you anything that the software didn't?

That nothing is ever finished, and everything you build you also have to maintain, forever, or it goes back to the field. A fence, a roof, a system - it's all the same bargain. You don't get to walk away from the thing you made. A lot of software is written by people who've never had to live downstream of their own decisions for ten years. Out here you always do. The hay you cut badly in July is the problem you've got in January.

You write about systems. Are you a great programmer?

I can hold my own, and cleverness was never the bottleneck, I'll happily own that. I joke that I know more than twenty programming languages but can barely speak my native tongue. But great and clever aren't the same word, and mistaking one for the other cost me years. The only skill I'd genuinely defend now is writing the least code that works, and that code is usually pretty dumb. Clever is a loan taken out against the attention of whoever reads it next, and the next person is nearly always me, six months on, tired, with no memory of why I was so pleased with myself. Dumb and obvious is a kindness to that person.

You wrote twenty-five blog posts to draw people toward the manifesto. Why not just point them at the manifesto?

Because the manifesto is written for someone who already trusts me, and nobody starts there. So the posts are doors into the same house. Each one takes a single scar - a backup that was never once restored, a clock that put the refund before the sale, a program that opened the vault for the wrong person - and just tells it as a story, stripped of the vocabulary and the tenets, with nothing to look up. If it works, you finish it nodding, and only then does it mention, almost in passing, that there's a longer argument behind it. You don't lead with the thesis. You lead with the wound everyone recognises.

What do you want someone to take from all this?

Relief, mostly. The whole thing is permission to stop pretending you'll be sharp when it counts. You won't. I won't. It's just the kit we were issued, no more a moral failing than poor eyesight, and once you accept it you can build for the person you actually are instead of the one you keep promising to become. People hear all this as pessimism, and they've got it backwards. Pessimism says people are careless and nothing can be done. I'm saying people are ordinary in completely predictable ways, and predictable is something you can build for. If you want one concrete thing to do on Monday: pick a single resilience feature you've never once watched work - the failover, the restore, the alert you're sure fires - and make it happen on purpose, in daylight, with a stopwatch running. By the end of the morning you'll know one true thing instead of believing four comforting ones. Trade a belief for a measurement. Then do it again next week.