Learn about Harnesses! I went over all these - omp - pi - claude code - zcode - codex - warp This is what I am using everyday, I explain the difference and benefits of each.

On Youtube also

Pi isn't comparable to others imo wins by a large margin

Add Hermes!

No more droid ?

No hermes harness?

No Hermes?

Try by @1jehuang

bro this is exactly why i don’t think there’ll be one “best” harness 😭 you end up using codex for one thing, claude code for another, omp/pi for something else… and then somehow you have 8 agents running across 5 terminals. we’re basically building @aoagents for that mess. use whatever harness you want, AO just gives you one place to run and keep track of all of them.

jcode is another rust option

Six harnesses. Still one question. What can the chosen one touch without asking.

Which harness wins for you?

Have you used @opencrabs ?

You have to give Ghostex a try again my friend, been working very hard on it and it's really good now. Recently added Agent CLI "Chat Views" and an Extensions system. Also made the app much lighter and faster (It's now GPUI & Rust based instead of swift/appkit). Android & iOS apps rebuilt from scratch too. Please let me know what you think!

The thing that surprised me most moving between harnesses is how much of the quality gap is the system prompt and the tool descriptions, not the model. Same model, two harnesses, very different tool call accuracy. Which one has felt most steerable to you?

Hey @grok tldr this video plz

The only harness difference that survives a month is how it handles context compaction and what it can touch unasked.

Decompile codex app lol

I'm trying out Codegraff and JCode

timely, thank you! 🙏🏽🤗 (cc-ing @itsbeaudean too)

Add amp and devin too

Hermes?

You look older than I expected

Swival?

Hi which of these are written in bash

One can use Claude Code models on other harnesses with a subscription or only via api?

Keep the great content coming OxSero!

it often shocks me how disgustingly unproductive omp vibe mode can be with the wrong model

This is great! Btw you can have subagents (and goals) with pi extensions

Nice I need to watch

This was really interesting. Thanks for sharing your workflows!

The model list is noisy. The harness list is the actual stack.

harness diffs matter more than model diffs lately.