claude being claude. I intructed it to iterate on the commands to find optimal prompts. LOL

ahaha i created a company. fml. just talk to it is still more effective. it was fun tho 😂

gotta fit some assistants and PMs in there that sit on the sidelines and scream.

We need a word for this... "agentic producivity trap"?

@steipete The automation is clever, but I wonder about the cognitive overhead of managing multiple subagents through tmux. The problem isn't just fixing the linter issues, but maintaining mental context across parallel execution streams. At some point, the coordination cost might exceed

@_junaidkhalid1 my adhd brain is really good at this.

@steipete Why don't you just use "biomejs"

@1baga doesn't do deep typescript linting. only per file which is super limiting.

@steipete Yeah that, and documentation generation are two of the most highly parallelisable tasks. But how much wall time did you save, really? Often I just leave this to the end of a session and leave it running over lunch or something.

@julianharris I'm doing this for fun, not because it's faster. Next time I have this issue I might save time.

@steipete I was doing similar but actually switched to a single threaded lint fixer that categorizes by problem type. I found that when looking at one file at a time, it will sometimes fix one thing and break another. Meanwhile, often multiple lint issues will have a common source problem.

@0xByron Claude does that. codex reads so much files that this is not an issue.

@steipete How are you doing the loop? Spawn 1 cc and have it call codex directly?

@0x_Sero worker - manager - boss - and a script that wakes up boss periodically.

@steipete Try giving them mail, then they coordinate even more effectively! https://x.com/doodlestein/stat...

@doodlestein 🤣

@steipete Great now just remove your self from the loop with a single overseeing LLM. Let us know how that goes.

@giffboake haha, for basic stuff that's totally doable, but where's the fun in that?

@steipete That’s backwards no? Shouldn’t the smarter gpt5 orchestrate and plan instead?

@garyfung Nah, gpt writes far better code. claude's a better agent. linting isn't planning features.

@steipete This is the way. At least until codex cli gets the missing orchestration features.

@s_streichsbier there's no missing featurs, it's just not a good agent. it's good at getting a task done.

@steipete Is there any UI for the Claude code or you need to do it all things in the terminal?

@Mateq007 cause I like that.