Why I build AI-first, and what it actually changes
The speedup is not typing. It is that the cost of trying the second idea dropped to almost nothing, which changes what you are willing to attempt.
I came into development from digital marketing, which means I never had the luxury of pretending code was the interesting part. Code was always the thing standing between an idea and a live page. So when a tool arrived that removed most of that distance, I did not have a craft identity to defend. I just used it.
People assume AI-first means the model writes the code and I paste it in. That is the least useful version. What changed for me is the cost of the second attempt. Before, the first architecture you chose was the one you shipped, because rewriting it meant a week you could not bill. Now I will build a feature two ways in an afternoon, look at both running, and delete the worse one. I make more decisions with evidence and fewer with taste. That is the whole gain, and it is bigger than typing speed by a wide margin.
Day to day, the shape of the work looks like this. I write the specification first, properly, in prose, including the things I am not building and why. That document is now load-bearing: it is the brief for me and the brief for the model, and a vague sentence in it produces vague code within minutes instead of within a sprint. Bad requirements used to fail slowly. Now they fail fast and loudly, which I count as an improvement.
Then I hold two lines hard. Everything generated gets read before it gets committed, and anything I cannot explain to a client does not go in. Not because the model is untrustworthy in some abstract way, but because I am the one on the call in six months when it breaks. Ownership does not transfer to the tool. The teams I have seen get burned all skipped that step, shipped a system nobody understood, and then discovered that debugging code you never read is slower than writing it yourself would have been.
The uncomfortable part is what it does to pricing. If a build genuinely takes a third of the time, hourly billing quietly punishes you for being good at your job. I moved to pricing the outcome instead. Clients do not care how the work gets done. They care that the storefront is live, that it converts, and that somebody picks up the phone when it does not.
AI-first is not a stack choice. It is admitting that the bottleneck was never the code.