Friend sent me some PR that fixes compatibility in C code with latest versions of libraries and runtime which was done completely by AI agent. So you have autotests and some fails after update deps. It took few hours for agent to fix every issue, but it did the job. Mostly refactoring with renaming old var with new one. But sometimes additional code required.

It given me an idea how to evaluate quality of AI agent. Here is the rule: if junior developers would cost you same price as AI agent, would you still use AI instead of real humans? Or otherwise, if you need to pay same price for AI as salary for devs (including onboarding, vacations, etc) - would you still chose it over humans?