Building at AI speed: the real bottleneck is deciding
Today, in a single session, I built more than I would have put together in a week a year ago: articles in two languages, a manifesto, a new public page, the rewrite of a small content cluster and — in the background, while I wrote — a team of agents diagnosing and fixing three bugs in a separate project. I'm not writing this to brag. I'm writing it because, at the end of the day, the thing stuck in my head isn't the speed. It's something else: producing is no longer the bottleneck. Deciding is.
TL;DR
- At AI speed the limit isn't how much you can make, but how fast you can choose well.
- Every quality jump of the day was a «no, not like that» — mine, not the model's.
- It's the same double-edged sword as always: AI multiplies both quality and chaos. What changes is who holds the handle.
- Discipline (real sources, the right words, verified numbers) doesn't vanish with speed: it becomes the only thing that matters.
Speed wasn't the point
For years the job was: have an idea, then spend hours turning it into code or text. The slow part was production. Today that part is almost free. I can dictate a confused thought and get it structured; ask for a translation and have it run in parallel across six files; spin up a team that investigates a bug while I write something else.
And here's the flip: when producing costs almost nothing, the only scarce resource becomes judgment. Not "how much can I push out", but "what deserves to exist, and in what form". The bottleneck moved from the fingers to the head.
Every quality jump was a «no»
When I look back at the moments where today's work actually got better, the model never decided it. A blunt refusal of mine did:
- «I don't like that word, it's not professional.» And a phrase tossed off became a precise concept.
- «Cut the commercial part from the manifesto.» And a muddled document split into two — one high, one blunt — and both worked better.
- «It's not the past: a law is an idea, and the data tells you what it became.» And a vague thesis became a backbone.
- «Those numbers are old, check them.» And they were old indeed.
The model produced the volume. The value came from those «no»s. That's the thing nobody automates for you: taste, and the nerve to throw things away.
The same double-edged sword, lived
This isn't something I made up today — I'd already written it: AI can raise quality or multiply chaos, and it depends who uses it. The difference is that today I lived it for hours, and felt both sides in the same afternoon. The bright side: you build in a day what you used to postpone for months. The dark side: the same ease with which you produce one right thing makes you produce ten wrong ones, and nobody notices they're wrong until you really look.
Speed without someone who really looks isn't a shortcut. It's debt piling up in silence.
Discipline doesn't disappear: it becomes everything
There's a thread tying together everything I wrote today, even pieces about different things. On one side, the conviction that Git and versioning discipline separate serious projects from chaos. On the other, the idea that a serious tool must answer with real sources, not opinions.
They're the same thing, seen from two sides. At AI speed, discipline — using the right word, citing the source, verifying the number before writing it, saying «no» when it rings false — isn't a brake on productivity. It's the only part of the job that's still entirely mine. The rest gets delegated. This doesn't.
What I take home
That the craft is changing shape, not disappearing. You write less and decide more. You become less «the one who produces» and more «the one who holds the line»: on quality, on truth, on taste. It's lighter in the hands and heavier in the head. And that's fine — as long as you remember that the machine goes as fast as you like, but you set the direction.
FAQ
Does AI replace writers or developers?
It replaces the mechanical part — production, translation, the first draft. It doesn't replace judgment: what's worth doing, in which words, with which sources, and what to throw away. That part, today more than ever, is human.
How do you work well at this speed without piling up chaos?
With discipline: verified sources, chosen words, numbers checked before publishing, and the willingness to say «no» and redo. Speed amplifies both care and sloppiness — you decide which.
You write less, you decide more. If you're moving fast with AI but fear you're piling up chaos, that's the problem I help keep under control: let's talk. And if you like, read the flip side: I dropped Git for speed and why discipline still matters.
Related articles
- Understand a country by cross-referencing its open data, with AI agentsA statute is an idea; the data tell you what it actually became. An agentic system that cross-references every public database — law, cadastre, statistics, archives — to understand reality by analysing the data, not the opinions.
- Why I Use Git as a Litmus Test When I Judge a TeamVersioning discipline is still the fastest way to tell whether a dev team is serious. Here is why — and where Git actually comes from.
- I Dropped Git for Speed (and AI Let Me Get Away With It)Under pressure I gave up my sacred Git+CI workflow for instant deploys. AI made the shortcut rational. Here is what I learned — and what scares me.