Building with AI - Dreaming and Incrementality

The more I build with AI tools, the less and less I ask them to do. When I started using Cursor, I had gotten caught up in the demo hype. I asked it to build big, complex features in my product. Features I didn’t define very clearly and that would take serious architectural thinking across many files to make work.

I quickly found myself spending more time chasing hallucinations and weird AI decisions than actually building new features. I learned that it was a lot cleaner and faster to ask the AI to take smaller, more focused, steps. This change of approach allowed me be productive again.

But what’s the cost of conceding to this kind of pragmatism?

In this morning’s Stratechery, Ben Thompson took a contrary view of Tesla’s recent “We, Robot” robotaxi event that has been widely mocked among a lot of people in tech. While Thompson acknowledges a lot of the weaknesses in the demo, he pointed out that what Elon Musk does well is communicate and pursue a big dream. By doing that, you’re forced to break through the current constraints that exist with the big step changes required to make the dream feasible.

This approach is what allows you to make big leaps and tends to be proven right in the end. On the other hand, more incremental, pragmatic approaches tend to win in the short term but then dramatically underperform the big dreaming approaches in the long run.

It’s not quite the same thing as the AI coding problem, but it was a good reminder for me. It’s okay in the short term to think incrementally and work pragmatically. After all, you need to actually release something now. And, unlike Elon Musk, most of us don’t have billions of dollars to bridge the gap between the present and the dream.

But it’s important not to lose sight of the initial dream and the more hopeful applications of AI that you had when you first learned about these tools.

Eventually they’re going to get good enough to handle those dreams. If you get stuck in the pragmatic mode of thinking, you’re going to limit what the AI can do.

Pragmatism is good, but sometimes it’s important to dream a little every now and then.