Every agency's website now says "AI-powered" somewhere. We'd rather tell you specifically what these tools are good at, because we use them daily and the honest answer is more interesting than the slogan.
Here's the map as we see it from inside real projects.
Where AI genuinely earns its keep
The first draft of almost anything. Boilerplate code, a migration script, a test suite skeleton, the plumbing of a form. Work that used to take an afternoon of typing now takes minutes plus review. The review isn't optional, but the afternoon is gone and nobody misses it.
Reading more than a human can. Point a model at a 40-plugin WordPress install, an inherited codebase, or a decade of server logs and ask what's there. As a summarizing, pattern-spotting assistant, it's superb. We find issues faster than we did three years ago, full stop.
Migrations and translations. Old jQuery to modern JavaScript, one template system to another, a CSV of chaos into clean structured data. Mechanical transformation with a clear before and after is the sweet spot. This used to be the most soul-crushing work in the industry, and we're glad to lose it.
Tests nobody would have written. Edge cases, weird inputs, the fifth and sixth scenario a tired human skips. Test coverage on our projects is simply better now.
Where it doesn't
Knowing what to build. A model will happily build the wrong thing with perfect confidence. Understanding that the client's actual problem is the approval workflow, not the dashboard they asked for, still comes from conversation and experience.
Architecture. The decisions that make a codebase cheap to change for ten years are about the specific business, its data, and its future. Models optimize for plausible; architecture needs true.
Taste. The difference between a generated layout and a designed one is the difference between text set in a font and typography. Fine, until you put them side by side.
Accountability. When something breaks at 2am, a model has no pager. Someone has to own the outcome, not just the output. That's the part of this job that was never really about typing.
The shape of it
The pattern across all of these: AI compresses the distance between deciding and having. It's a phenomenal engine for producing what you've already specified. It contributes almost nothing to knowing what to specify, and it takes no responsibility for the result.
Which is why our answer to "do you use AI?" is yes, everywhere it earns its keep, and why the studio stays founder-led anyway. The hours we save on boilerplate go into the parts the tools can't touch: understanding the problem, designing the thing properly, and answering for it once it's live.
The typing was never the hard part. It just used to take longer.