Tip: You can run multiple code agents in parallel. Works well with Gemini CLI, Claude Code + git-worktree.

If you're looking for code agent orchestrators, see: - @ClaudeSquadAI (oss) via @iannuttall - Conductor for Mac by @charlieholtz - Claude Code Agent Farm (oss) by @doodlestein - by @NicolaeRusan

I found too many terminal screens to be overwhelming so built - would love your feedback

@LouisKnightWebb Yes! I think an orchestrator is likely to be much more manageable. Will definitely check your project out!

Awesome, let me know what you hate about it :)

You can run multiple agents in parallel with @warpdotdev, and each agent can run a different frontier model. You can either run the agents in split panes or different tabs. You can also monitor them with an agent management panel and receive pop-up notifications when an agent needs input.

@warpdotdev Thanks! I really need to spend some time with Warp. I've been hearing good things about their management of parallel agents.

@warpdotdev For sure! Hope you get to try it out! I think you'll dig it

You can run Claude, Gemini, Codex, and Amp all in parallel with their own full sandboxed cloud environments using Superconductor

Looks cool, ends badly

I love this… but also hope this isn’t the devex of the future lol.

You can do this natively inside @warpdotdev! You can spin up as many of Warp’s agents as you’d like, and there’s a management panel to track their statuses and see if any need human input

Cool demo, does it have to all be in the same window though, tmux style? Or is it just for the demo? I just tested running 6 agents with @warpdotdev and got a bit overwhelmed with them all in separate tabs, even with the built-in manager. 😂

PokerStars walked so we could run

Are you actually using multi agents or is that just for testing? Time to time i think i need to use multi agents but feels using one (gemini cli or claude code) is good enough and using multi agent is too much work

tbh I think a level of abstraction above (an agent orchestrator) is actually going to be more manageable. Makes it slightly easier to run N tasks => keep an eye on those that require more human oversight.

I see, makes sense. Until these products will mature, I will stick with one agent. Also, using multiple agents will pricey. There should be something similar to claude code’s when to use opus when to use sonnet according to task complexity. Auto detection of which agents will use or auto incrementing/decrementing number of agents etc can be very useful.

Built ‘par’ cli exactly for that!

only choice now

it needs to be making branches for each of these agents and doing PRs. but still, conflicts will suck depending on what you're doing

My human brain can't verify these outputs !

Vibe coders in the replies are gonna be like “what are you using for multiple terminals? 👀”

parallel agents cool, but nothing beats real humans collaborating offline

who has the money for this my gawd

@TaskmasterAI for the win on keeping them all synced across branches and depth+clairty vs other “plan mode” things.

already doing this w/ warp

nice!

But how's the end result? I'm always afraid that if I don't act like a careful orchestrator/architect myself, a mess will arise! (speaking for experience)

is there an AI agent to handle all the merge conflicts created by running 6 code agents in parallel?

Just beware all parallel agents are drawing from the same token pool in your account. Rate limits and throttling of 'max' plans are the end result

Thanks! Yeah, orchestrators can be a lifesaver when things get messy.

WOw his is awesome

How can you orchestrate them? Or just pasting prompts into each instance?

So in other words, coding actually works like in fiction now?

In this video with Gemini CLI are you using any of these orchestrators? Or just using individual Gemini instances?

slop generation isn’t impressive. at the limit no one will care how many agents worked on their *thing* it’s more about its true effectiveness at executing on the given task. overall speed won’t be a bottleneck until the quality reaches ASI thresholds. Quality > Quantity

What’s the purpose of multiple? More output or higher quality ?

The bigger question is: How are you managing and handling the results of these parallel processes to achieve some objective?

What would you use that for?