@kunchenguid Do you use review bots in your flow post-push? In a lot of my PRs, the over engineering happens after I push when the bots go back and forth for a while.
Fable 5 is an interesting one. I finds it hallucinates less and can come up with much more "correct" plans than the other models , but I hate its output and it's a constant battle between me and it on its output. Trying various things like telling it to ELI5, use ASD STE 100, or that I have ADHD even.
@illscience shopping probably is going to be huge at some point. even for simple recurring cases like planning out your week's meals and ordering your groceries.
@gregisenberg And don't forget security! Agents need guardrails to prevent themselves from going rogue in the face of prompt injection and the like. As well as to protect themselves from supply chain attacks.
@coco_ruski@trq212 I've removed superpowers from mine personally starting at opus 4.8. Felt like it was doing a pretty good job with a much less bloated and me-specific system prompt.