Not much different at coding than Opus 4.7

After probably several hundreds of hours programming with Opus 4.6 and 4.7, the jump to 4.8 seems almost unnoticeable to me. In general both models are extremely capable of executing well-scoped tasks from honestly pretty under-specified prompts.

I also have found Opus 4.8 (and 4.7) to be exceptionally useful at "semantic refactoring," which is to say, refactoring code to change its semantics, rather than simple syntax or naming refactors. I consider this to be a different task than greenfield code production, which (very much like humans) tends to be an easier task in general.

Asking it to develop a plan for implementing things on a PR granularity produces good, reviewable changes, and is frequently the best way to catch mistakes before it wastes a lot of cycles doing the wrong thing, or missing obvious alternatives. I do find it's necessary to ask specifically for this instead of just relying on its built in plan/checklist system.

I have noticed that 4.7 and 4.8 seem to be pretty prone to answering my questions in a frustratingly Stack-Overflow-inspired fashion, telling me that I'm wrong for asking questions, and that I should instead be doing _____. Listen here robot, I pay the bills around here, just answer the actual question I asked.