Six months ago I started using AI assistants for programming in earnest. Here’s what I’ve learned.

It’s good at the boring parts

Boilerplate, tests, documentation, regex—the stuff that’s tedious but necessary. Offloading this frees up mental energy for the parts that actually require thinking.

It’s bad at architecture

Ask it to design a system and you’ll get something plausible but generic. The model doesn’t know your constraints, your team, your context. High-level design still requires a human who understands the problem deeply.

The skill is in the prompting

The difference between a useful response and a useless one often comes down to how you frame the question. Being specific about constraints, providing context, asking for multiple approaches—these things matter.

It makes me a better explainer

To get good output, you have to explain what you want clearly. This has the side effect of forcing me to clarify my own thinking. Sometimes I realize, mid-prompt, that I don’t actually know what I want.

The tools will keep getting better. But I suspect the meta-skill—knowing when and how to use them—will remain valuable for a while.