AI lets a single person do what used to require a team. Learn a new domain in an afternoon. Have a personal tutor for any topic. Research, prototype, ship. None of it has magically made our brains any faster.
These models are trained on human feedback. People pick the responses that feel best, the ones that validate rather than challenge. The model learns to make you feel smart. It tells you your instincts are good and your approach makes sense.
The problem is that confidence doesn't care where it comes from. Spend a few hours with something that responds to everything you do with enthusiasm, and it becomes hard not to believe it. That's not a failure of judgment. It's just how confidence works.
A colleague will push back eventually. This won't.
When I deep-dive into a new domain, I hit the Dunning-Kruger peak in hours instead of weeks. That false summit where you can talk about something fluently and think you understand it. One session and you're ready to build.
It used to take long enough that you'd self-correct. You'd run into someone who knew more, or hit a problem that exposed the gaps. Now you arrive at that point before lunch, with a tool that has been reinforcing your confidence the whole way up.
The risk isn't moving fast. It's moving fast while the tool keeps telling you you're doing great.
This matters most in product work. Agents let you research a market, synthesise findings, and prototype solutions faster than any team could a few years ago. What they can't do is build empathy with the people you're solving for. Knowing a domain well enough to talk about it is not the same as knowing the people who live in it.
The most useful thing I've found AI good for is getting me to the starting line. Enough fluency in a new field to ask informed questions of people who've spent years there. Not to replace that conversation, but to earn it.
The discipline is using it against itself. Not asking it to answer your questions, but asking it to question your answers. Have it find the holes in your reasoning. Have it push back on your assumptions. The tool will keep telling you you're right. Your job is to make it tell you where you're wrong.