I've been building products for 15 years. VR platforms, immersive experiences at the World Bank, enterprise tools, and now an AI company. In all that time, nothing has changed how I work as much as building with AI.
Not what I build. How I build.
There's a lot of noise about this right now. AI is going to replace every developer. AI is a toy that spits out broken code. Both camps are loud, and both are wrong. So here's the honest version, from someone who ships with it every day.
AI is the fastest, most tireless teammate I've ever had. It never gets bored. It writes the boilerplate, wires up the integration, refactors the ugly function, and explains a codebase I've never seen in minutes instead of days. Work that used to eat a whole afternoon now takes a coffee.
A few months ago I took on something that would have scared me a year earlier: untangling a large, unfamiliar codebase that had grown messy over years. Normally that's weeks of careful reading before you dare change anything. With AI, I could interrogate the code, trace how a change would ripple, and move with a confidence that used to take much longer to earn.
But it is not magic. It doesn't know what you're actually trying to build. It will hand you something that looks right and is quietly wrong. It has no taste, no memory of why you made a decision three weeks ago, no instinct for the one edge case that pages you at 2am.
So the job didn't disappear. It moved.
I spend less time typing and more time deciding. Less time on how, more on what and why. Execution used to be the bottleneck. I had ten ideas and time to try one. Now the cost of trying an idea has collapsed, so the scarce thing is judgment: knowing which idea is worth building, and being honest about when the output is genuinely good versus just good-looking.
The trap is real, too. It's easy to let AI hand you code you don't understand and ship it anyway. That's how you end up with a product you can't debug and a team that can't own it. The discipline I've had to relearn is simple: I don't merge what I can't explain. AI writes the first draft. I still have to understand every line.
What surprised me most is that it made building feel less lonely. Founding a company can be isolating, especially early, when it's just you and a blank file at midnight. Having something to think out loud with, to argue an approach against, to hand the grunt work to, changed the texture of the day. I ship more. I experiment more. I kill bad ideas faster, because they're finally cheap to test.
After 15 years, I didn't expect a tool to change how it feels to build. This one did. It didn't replace the craft. It stripped away the parts that were never the craft in the first place, and left me more time for the part that was: understanding the problem, and deciding what to make.
If you're building something and trying to figure out where AI actually fits, without the hype or the fear, I'd be happy to compare notes.