Speed Is Not Velocity

To my former physics students: AI has me thinking again about the difference between speed and velocity.

Speed is how fast you’re going. Velocity is speed with direction. I drilled that distinction for 15 years. I never expected to need it in software.

Now I run 100+ AI agents, and I can tell you: the industry does not have a speed problem. It has a direction problem.

Cursor, Copilot, Claude. They all make you faster. My agents can execute a full feature spec in 10 minutes. That is extraordinary speed. But early on, I gave agents objectives without enough constraints. The result was the software equivalent of driving 120 mph with the wrong GPS directions: fast, impressive, and ending up at the wrong house.

The fix was never faster models. It was better direction: constraints, architecture, and specifications.

That is what I tell anyone who asks about working with agents. The hard part is not getting them to move. The hard part is knowing where to point them. That part, the figuring out, is still slow, still messy, and still deeply human.

Speed is getting cheaper. Direction is the scarce resource.

So the real question is not how to make AI go faster. It is where direction comes from, and how much of it you need before speed becomes useful.