The Human Maxima
Agents are relentless climbers. Choosing the hill is still our job.
I spend my days building infrastructure for AI agents. Ironically, I’m building the tools that should, in theory, automate me out of a job. So people assume all this acceleration makes me nervous. It doesn’t. I’m optimistic.
Agents are relentless climbers. Improving them is literally called hill climbing. Give one a direction and it goes further, faster and more patiently than any of us.
But that grit cuts both ways. It doesn’t stop. It never asks whether this is the right hill. That question is most of what we call taste.
If you’re starting a company now, this is the part to get good at. Agents will do more of the climbing every year. Some of your edge is knowing which models and harnesses climb fastest, and shaping your software around them. The bigger edge is noticing when you’re on the wrong hill, stepping into the agent’s work, and pointing it somewhere better.
Everything will feel like progress. That’s the trap. Without the judgment to make those calls and the autonomy to act on them, a hundred agents will pull you in a hundred directions, and nothing will get done.