@poteto What's your breakdown on pr count by change type (new feature, bugfixes, skill enhancements, quality refactors, etc)? What's the split between PRs from agentic loops vs human-initiated/one-off prompting?
@jacobgold@poteto What's your guardrails stack look like? Is it pstack? I'm having trouble getting agents to keep PRs simple/focused/conventional etc, and it's a struggle to correct it without human review
@poteto@bot How do you ensure your agents are making the right architectural or design decisions? How do you stop them from adding excessive complexity?
Muse Code runs specialized background agents that stay active your whole session, so they build up context over time instead of starting from scratch on every task.
When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees. Your working copy is never touched. In testing we had it build six features for a game simultaneously with no collisions.
@skeptrune I'm trying to scale out something similar in my org. I think you need the AI approver looking for directional alignment/unambiguous improvements in addition to being bug-free, and it helps to start with low-risk PRs. You can be more flexible if you have high trust in the team
@fredrikalindh@dreamsofcode_io I still have to make corrections & guide the agent to get what I want, even with the latest models. And since I'm accountable for what it does, I spend a lot of time trying to get on the same page with it, especially for more open-ended research tasks. Is it different for you?
@PaulTassi I'm really hoping marathon leans deep into PvE. We know they've got the world class talent to make the best shooter PvE content in the world