Vibe Coding and Self-Driving Cars

Both self-driving cars and AI coding get destroyed in the discourse for the same reason - they make stupid mistakes that humans wouldn’t. Autonomous vehicles drive into fire trucks, Claude writes functions that don’t handle edge cases. Easy dunks.

But the car safety argument is compelling - even with occasional failures, autonomous systems could reduce overall accidents because human drivers are inconsistent. Some people are great drivers, others text while merging. The aggregate matters more than individual failures.

The coding version of this isn’t about error rates though. It’s about energy allocation. Most developers spend 80% of their cognitive budget writing the first draft of code, then have whatever’s left over for debugging, testing, refactoring. That’s backwards.

AI vibe-coding flips this. Instead of grinding through implementation details until you’re mentally fried, you burn through the first draft quickly (bugs and all), then spend your fresh mental energy on the stuff that actually makes code good - finding edge cases, improving architecture, writing tests that matter.

Right now we treat debugging like cleanup work. But debugging is where you actually understand the problem deeply. It’s where good software gets built. If AI can handle the mechanical translation from idea to running code, we can spend our energy on making that code robust instead of just functional.

This might be wrong. Maybe the debugging phase needs the deep familiarity you get from writing every line. But the energy reallocation feels promising enough to experiment with.

This post was written with AI assistance.