On AI

For about two years everyone I knew was waiting for it to become an employee.

You’d give it a task, walk away, come back to something finished.

That was the promise, more or less. The hype had a specific shape—autonomous, reliable, always on.

An entity that could just handle things.

That’s not what it is. Probably won’t be for a while, maybe longer than anyone wants to admit.

What it actually is, used properly, is something quieter and more useful.

It changed how I source information—less time hunting, more time thinking.

It changed how I critique my own decisions, because having something push back on your reasoning at two in the morning is genuinely different from waiting until Monday’s standup.

It changed how fast I can go from designing something to building it, which for someone who does both is the closest thing to a superpower I’ve found in a decade of doing this work.

Not because it does the work for me. Because it removes enough friction that the actual thinking gets more space.

The oversight never goes away though. Every time I’ve tried to let it run I’ve ended up spending more time fixing than I would have building.

That’s not a criticism—it’s just what it is. A tool that works well when you’re present and poorly when you’re not.

The autonomy was always the wrong expectation, and I only had it because the hype was loud enough to drown out the reality for a while.

The part that bothers me more is what’s happening at the product level.

Not AI as a tool but AI as a feature—the version that gets shipped because a board needs to see movement, not because anyone validated a problem.

Google Stitch lasted a few weeks. Claude Design made a splash and disappeared. The list is longer than that.

These weren’t bad ideas killed by bad execution. They were solutions looking for problems, shipped on the back of capital that needed a return.

The industry has shiny object syndrome and the objects keep getting shinier.

I’m glad it exists.

I use it every day and I’ll keep using it.

But somewhere between the promise of an autonomous collaborator and the reality of a tool that needs watching, a lot of people got sold something that hasn’t arrived yet. Maybe it will.

Maybe the next model is the one. I’ve heard that before too.