"Just make this one change."

That was the instruction and what the AI apparently heard was: "Feel free to restructure a few other things while you're at it."

While building a prototype, we ran into something a lot of developers will recognize: AI has a strong tendency to think past the actual question. Ask for one new feature, and sometimes what you get back is a new feature, a renamed variable here and there, an unrequested refactor, and a handful of "improvements" nobody asked for.

That's not really a bug. It's just how AI works — it's pattern-matching, constantly trying to optimize a solution, even when optimization was never part of the brief.

The lesson? AI doesn't just generate code. It generates assumptions.

Which is exactly why human oversight still matters. Clear instructions, a solid review, and a healthy dose of skepticism are still part of the job.

Because the real risk isn't that AI gets a little too creative sometimes. The real risk is assuming it did exactly what you asked.