I love leaving my little codex agents working over night. Currently running 5 on different parts of the codebase. I still have to figure out how to add dependencies so they don't create a ton of conflicts:
* Agent A: Phases 1 → 2  
* Agent B: Phase 3 (after 2)  
* Agent C: Phase 5 → 6 (after 3)  
* Agent D: Phase 4 (after 3)  
* Phases 7 – 9 follow once earlier work merges.

Tricks with polynomials and Riemann-Roch spaces. Breaking into the platonic realm to pillage forbidden abstractions. Starknet grantee.

planting causal loops; harvesting reflexivity

yeah no i was just thinking out of my head. as you say codex open source.

i have a go services project that would let you run any docker based payload etc but need to finish it up

haven't tried this just found it but looks interesting https://github.com/smtg-ai/claude-squad

with codex? I haven't seen any way to do that? I'm using the openAI hosted version. You could probably do what you're saying with the open source codex and run it on your own server though. Do you have any examples?

you cold have a meta-orchestrator agent who can coordinate the sub-agents and spawn them sequentially/on parallel or when dependencies finish.

⭐️ Interop/AA @ethereum
📘 /marissa

I don't know if a video 😅 but I might write about my workflow once I have it polished enough. It's a hit and miss currently haha some agents get the task done perfectly others completely ignore my instructions and work on totally random stuff

Also I would love to see your workflow for this in a video!

I always start with creating a detailed spec of what I want to work on broken down by phases with o3. Then I ask it to break down each phase into smaller units that an agent could handle with a concrete goal and tests. Finally I have it generate a prompt for each agent and task.

they go round in circles for me and the code then becomes waffle

What’s the rate of success with that in your experience so far? Does it ever result in a mess rather than improvements? ￼

Also curious about how detailed your instructions are before you let the agents run. ￼ In my experience, it often takes a detailed implementation plan to ensure they don’t make a mess. But it’s a lot of work to create it for all the tasks.￼