About Development Bottlenecks

Yesterday OpenAI published a rather interesting post about how they developed the Sora Android app using their Codex. I don't want to retell everything, it's already been done well here, and in general the talking points remarkably closely repeat the content of my channel and my talks 🙂

I want to latch onto one quote:

Our bottleneck in development shifted from writing code to making decisions, giving feedback, and integrating changes.

Always has been, as they say. They're obviously being quite disingenuous, since this idea looks great for marketing purposes.

Coding itself is almost never the bottleneck. Behind 80% of resources spent on development come another 80% of resources spent on analytics, QA, deployment and various marketing things. Or at least that's how it should have been. If it wasn't, then something's wrong at the conservatory.

Simply put, if your project has somehow established development approaches and everyone understands what to do, you start working much faster than things can be reviewed, tested and then regression-tested. And if during development you don't think about how not to break anything extra and how difficult it will be to test, the bottleneck will be one step further.

AI hasn't really changed anything, just showed it a bit more clearly. Just as without AI we started piling on automated checks and tests on top of development, we'll have to do the same here. But if without AI we can live without all this for quite a long time, with it you simply can't work any other way — it will produce nonsense, and very quickly. Next it won't be a bottleneck, but a bottleneck-zilla.