Amnesiac

When people talk about AI coding performance, they are more than likely talking about “individual task performance”, meaning, I tell AI to code up a feature and it does it.

What I’ve seen is, as the project gets bigger and decisions accumulate, not just big architectural decisions but even smaller ones, and bugs show up and get fixed, I see that AI is rather amnesiac.

A new feature might benefit from an old bug fixed but AI doesn’t remember about that and implements the same bug, to be told to fix it again afterwards.

It’s not a memory issue. When told about the previous bug, “ah I see the issue” and remembers and applies the same fix to the new situation

Btw this is not a criticism per se, this sort of thing happens with Junior devs too. That’s why I say AI is a junior dev who happens to know a lot about a lot, or can do tasks very fast, but needs a tech lead to keep tabs on it to make sure it’s doing the right thing at a high level