Take the optimism pill

I finished DHH's Rails World 2026 opening keynote feeling relieved, which is a strange thing to feel after an hour about the total transformation of your profession. But that's the talk. He says optimism is the only sane option and he clearly means it: not as a mood, as a conclusion.

The only sane bet

With predictions this unreliable, pessimism buys you nothing. If the future turns out fine, you spent your time scared for nothing; if it doesn't, you should have spent your last day a little more cheerful. He says it as a game-theory joke, but I felt something loosen hearing it said straight instead of tiptoed around. Yes, stay cautious: security is real, nobody knows what the economy looks like after a shift like this. But caution and dread are different things, and I've been carrying the second one without noticing. So: caution where it counts, optimism everywhere else. That part of me is done apologizing.

Everyone gets a camera

The part that rearranged something in me was the camera. A portrait used to mean sitting for months in front of a master painter, affordable only to royalty. Then the camera, then the Kodak Brownie at roughly a dollar a picture, and suddenly everyone could have their picture taken. Two trillion photos a year now.

But the part I keep returning to is what happened to the artists. Realism, reproducing exactly what's in front of you, stopped being a viable job the moment a machine did it cheaper. And they didn't vanish. They pivoted. Picasso went cubist, the impressionists chased light instead of detail. Art got less realistic and more interesting precisely because the realistic part got automated.

I keep turning that over because it tells me exactly where I stand: the machines take the reproducing. We keep the making.

He calls this bigger than computers, bigger than the internet, and for once the superlative doesn't feel cheap. I was young when the internet arrived: it was just there. This is the first before-and-after I am conscious inside of. Everything before November 24th, everything after. His Kodak Brownie moment, and I was here for it.

Faster, not stalling

The claim that models will plateau, that there's a ceiling to how good this gets, is the one I believe least, because I can feel the evidence arriving weekly. Releases went from once a month, once every two months, to a rhythm where a week with no model from any big lab is a silent week. And silent weeks feel weird now.

Open weights especially. DeepSeek 4.1 Flash, MiMo v2.6 with its transparent RL runs, Muse from Meta catching up on a lot. None of these were close to this good three or four months ago. I got my own taste of frontier intelligence using Opus 5 recently, and Opus 5.5 is even better. Whatever sits inside the labs right now is probably the generation we'll see in two or three months. Every stall gets declared the wall, then the wall ships a new version. I'm bullish because the evidence keeps outrunning the skepticism.

All the things I'm not doing

Then there's the clip of 2005 DHH: "look at all the things I'm not doing." It made me laugh and then made me quiet, because that has quietly become my job description too: not the code I write, the configuration I don't write, the things a small team "couldn't" maintain that now just get built.

And I love how he reframes the work: give a task to an agent the way you'd give it to a coworker, send them off, come back and review when something's ready. You stop being the person who chisels the code and become the person steering the intelligence. A professional maker of things. That reads less like a threat and more like a promotion.

The part of that which clicked hardest for me is the how: asynchronously. Don't sit in front of a chat window babysitting the loop, nudging each step along like it's pair programming with a slow intern. Hand over a goal, a problem, an outcome, not a task list, and walk away. Check back when there's something to review. The more synchronous you make it, the more you're still doing the job yourself, except now you're doing it through a text box. The whole point is the handoff. I want the loop as async as it can possibly get, because that's where the leverage lives: they work, I think, and neither of us waits on the other.

The guilt that left me

The deepest change is a guilt I didn't know I was carrying. The older guard always says you must review every line, check the AI code, never trust it. I'd always felt slightly bad skipping the diff. DHH's version of it: he doesn't know any Rust and considers that a feature. He evaluates the box from the outside, exactly the way any business owner who ever commissioned programmers evaluated theirs. Same phenomenon, new hands.

So I stopped confessing. A few months ago if you'd asked me if I'm a vibe coder, I'd have said yes and hesitated, a little ashamed, like you were judging me. Today the positions reversed: if someone tells me they write every line of production code by hand, I file them as an oldie. Fine as art, fine as a hobby. If it relaxes your nerves, enjoy it. I don't hate anyone for it. But it isn't economically viable and it isn't going to be. Just don't try to defend hand-writing every line as the future. His phrasing is harsher than mine: we had a wonderful run chiseling code by hand, and it's over. Same conclusion.

That, beyond the arguments, is what I actually want out of this: an age of abundance like none we've seen, and lives with more happiness than we can think of.