@steipete The natural next step is creating a loop designer that prompts your agents.

Wondering when my agents will cut me off because I am dead weight.

@InderosD I have my claw supervising my codex’es.

@steipete yeah easy for you to say lol

@LexanderBrouwer Who still uses VS Code?

@steipete Can you explain your workflow in detail? Would love a blog post about it

@gauthampai Don’t worry it’ll take 3 months until it’s there.

We’ll be talking about fleets that design your loops then.

@sabastod Yeah that’s one of the OpenClaw features, I use that when we have meetings. bidi > sst/tts tho

@mosyaseen I use a VISION.md for my projects

@weswinder welllllll

@steipete but you kinda still need to be in the loop to test stuff unless you literally have infinite tokens

@steipete @steipete i just made a voice agent that instructs my claude/codex and talks back to me.

Basically smaller local model with STT/TTS pipelines that sends tmux keys and instructions to my claude/codex session to build what I want.

Does that count?

@MyMoonEnt Correct. Is your time really not worth more?

@steipete Oh god, LinkedIn will now start a new fad, "Loop Engineering".

Harness Engineering is so last year. Loop Engineering is what you should be doing.

@steipete loop i follow on a daily basis: /grill-with-docs -> /to-prd -> /to-issues -> /tdd -> /diagnose( resolving bugs ) -> /teach ( understand/learn each slice as you build )

thankyou @mattpocockuk

What does this loop like in practice? Like Claude -p in a for loop? /goal?

@wicus_g Who still uses Claude Code?

@steipete the jump from "write me a function" to "design the feedback loop that writes the functions" is where most people stall out for about six months

@steipete If you wonder how to do it, you’re not alone: I’ve built AgentBox to streamline controlling tens of agents from Claude/Codex/Opencode

check http://github.com/madarco/agentbox

@egebasoner ngmi

@steipete totally agree you should use https://github.com/cgarrot/zob-harness