Blogmark

To master these tools you need to learn how to get them to prove their changes work as well.

I’ve been using Claude Code to automate a series of dependency upgrades on a side project. I was surprised by the number of times it has confidently told me it fixed such-and-such issue and that it can now see that it works, when in fact it doesn’t work. I had to give it additional tests and tooling to verify the changes it was making had the intended effect.

My surprise at this was because the previous several upgrade steps it would deftly accomplish.

Almost anyone can prompt an LLM to generate a thousand-line patch and submit it for code review. That’s no longer valuable. What’s valuable is contributing code that is proven to work.

The step beyond this that I also view as table stakes is ensuring the code meets team standards, follows existing patterns and conventions, and doesn’t unreasonably accrue technical debt.