Orchestrate teams of Claude Code sessions

Especially for what’s basically an experiment. Gas town didn’t really work, so there’s no guarantee this will even produce anything of value.

I mean what you get for Claude Code Max is insane its 30x on the token price. If you don’t spend that all it’s your own fault. That must be below elecricity cost

It's not like he was the only one who came up with this idea. I built something like that without knowing about GasTown or Beeds. It's just an obvious next step

https://github.com/mohsen1/claude-code-orchestrator

There seems to be a lot of convergent evolution happening in the space. Days before the gas town hype hit, I made a (less baroque, less manic) "agent team" setup: a shell script to kick off a ralph wiggum loop, and CLAUDE-MESSAGE-BUS.md for inter-ralph communication (Thread safety was hacked into this with a .claude.lock file).

The main claude instance is instructed to launch as many ralph loops as it wants, in screen sessions. It is told to sleep for a certain amount of time to periodically keep track of their progress.

It worked reasonably well, but I don't prefer this way of working... yet. Right now I can't write spec (or meta-spec) files quick enough to saturate the agent loops, and I can't QA their output well enough... mostly a me thing, i guess?

Compare both approaches to mature actor frameworks and they don’t seem to be breaking much ice. These kinds of supervisor trees and hierarchies aren’t new for actor based systems and they’re obvious applications of LLM agents working in concert.

The fact that Anthropic and OpenAI have been going on this long without such orchestration, considering the unavoidable issues of context windows and unreliable self-validation, without matching the basic system maturity you get from a default Akka installation shows us that these leading LLM providers (with more money, tokens, deals, access, and better employees than any of us), are learning in real time. Big chunks of the next gen hype machine wunder-agents are fully realizable with cron and basic actor based scripting. Deterministically, write once run forever, no subscription needed.

Kubernetes for agents is, speaking as a krappy kubernetes admin, not some leap, it’s how I’ve been wiring my local doom-coding agents together. I have a hypothesis that people at Google (who are pretty ok with kubernetes and maybe some LLM stuff), have been there for a minute too.

Good to see them building this out, excited to see whether LLM cluster failures multiply (like repeating bad photocopies), or nullify (“sorry Dave, but we’re not going to help build another Facebook, we’re not supposed to harm humanity and also PHP, so… no.”).

This is nothing new, folks have been doing this for since 2023. Lots of paper on arxiv and lots of code in github with implementation of multiagents.

... the "limit" were agents were not as smart then, context window was much smaller and RLVR wasn't a thing so agents were trained for just function calling, but not agent calling/coordination.

we have been doing it since then, the difference really is that the models have gotten really smart and good to handle it.

Honestly this is one of plenty ideas I also have.

But this shows how much stuff is still to do in the ai space

[dead]

At this point, if you're paying out of pocket you should use Kimi or GLM for it to make sense

Ah ok, same. I keep wondering about how this would ever accomplish anything.

I've had good results with Haiku for certain tasks.

You write a generic architecture document on how you want your code base to be organized, when to use pattern x vs pattern y, examples of what that looks like in your code base, and you encode this as a skill.

Then, in your prompt you tell it the task you want, then you say, supervise the implementation with a sub agent that follows the architecture skill. Evaluate any proposed changes.

There are people who maximize this, and this is how you get things like teams. You make agents for planning, design, qa, product, engineering, review, release management, etc. and you get them to operate and coordinate to produce an outcome.

That's what this is supposed to be, encoded as a feature instead of a best practice.

I agree, but I've found that making an "adversarial" model within claude helps with the quality a lot. One agent makes the change, the other picks holes in it, and cycle. In the end, I'm left with less to review.

This sounds more like an automation of that idea than just N-times the work.

Humans can't handle large tasks either, which is why you break them into manageable chunks.

Just ask claude to write a plan and review/edit it yourself. Add success criteria/tests for better results.

Exactly, one out of four or three prompts require tuning, nudging or just stopping it. However it takes seniority to see where it goes astray. I suspect that lots of folks dont even notice that CC is off. It works, it passes the tests, so it is good.

There is research[0] currently being done on how to divide tasks and combine the answers to LLMs. This approach allows LLMs reach outcomes (solving a problem that requires 1 million steps) which would be impossible otherwise.

[0]https://arxiv.org/abs/2511.09030

you need a reviewer agent for every step of the process - review the plan generated by the planner, the update made by the task worker subagent, and a final reviewer once all tasks are done.

this does eat up tokens _very_ quickly though :(

You definitely have to create some sort of PLAN.md and PROGRESS.md via a command and an implement command that delegates work. That is the only way that I can get bigger things done no matter how „good“ their task feature is.

You run out of context so quickly and if you don’t have some kind of persistent guidance things go south

I'm not anti-whimsy, but if your project goes too hard on the whimsy (and weird AI-generated animal art), it's kind of inevitable that someone else is going to create a whimsy-free clone, and their version will win because it's significantly less embarrassing to explain to normal people.

Where are the polecats, though? What about the mayor's dog?

I don't know what Gas Town is, but Claude Code Agent Teams is what I was doing for a while now. You use your main conversation only to spawn sub agents to plan and execute, allowing you to work for a long time without losing context or compacting, because all token-heavy work is done by sub agents in their own context. Claude Code Agent Teams just streamlines this workflow as far as I can tell.

yeah, seems like a much simpler design though (i.e. only seems like one 'special/leader' agent, and the rest are all workers vs gastown having something like 8 different roles mayor, polecat, witnesses, etc).

Wonder how they compare?

Yeah but worse

No polecats smh

>"Seems similar to Gas Town"

I love that we are in this world where the crazy mad scientists are out there showing the way that the rest of us will end up at, but ahead of time and a bit rough around the edges, because all of this is so new and unprecedented. Watching these wholly new abstractions be discovered and converged upon in real time is the most exciting thing I've seen in my career.

These two basically do what you want, let Claude be the manager and Codex/Gemini be the worker. Many say that Coder-Codex-Gemini is easier to understand than CCG-Workflow, which has too many commands to start with.

https://github.com/FredericMN/Coder-Codex-Gemini https://github.com/fengshao1227/ccg-workflow

This one also seems promising, but I haven't tried it yet.

https://github.com/bfly123/claude_code_bridge

All of them are made by Chinese dev. I know some people are hesitant when they see Chinese products, so I'll address that first. But I have tried all of them, and they have all been great.

You can accomplish this with https://github.com/AgentWorkforce/relay and make the Lead/Orchestrator any harness you want. At the core agent-relay is agent to agent communication but it unlocks quite a few multi agent orchestration paradigms. I wrote about some learnings here as well https://x.com/khaliqgant/status/2019124627860050109?s=46

I use opus for coding and codex for reviews. I trigger the reviews in each work task with a review skill that calls out to codex[0]

I don't need anything more complicated than that and it works fine - also run greptile[1] on PR's

[0] https://github.com/nc9/skills/tree/main/review

[1] https://www.greptile.com/

I think this is where future cursor features will be great - to coordinate across many different model providers depending on the sub-jobs to be done

You can run an ensemble of LLMs (Opus, Gemini, Codex) in Claude Code Router via OpenRouter or any Agent CLI that supports Subagents and not tied to a single LLM like Opencode. I have an example of this in Pied-Piper, a subagent orchestrator that runs in Claude Code or ClaudeCodeRouter and uses distinct model/roles for each Subagent:

1. GPT-5.2 Codex Max for planning

2. Opus 4.5 for implementation

3. Gemini for reviews

It’s easy to swap models or change responsibilities. Doc and steps here: https://github.com/sathish316/pied-piper/blob/main/docs/play...

At Augment' we've been working on this. Multi agents orchestration, spec driven, different models for different tasks, etc.

https://www.augmentcode.com/product/intent

can use the code AUGGIE to skip the queue. Bring your own agent (powered by codex, CC, etc) coming to it next week.

Unlocking the next order of magnitude of software inefficiency!

Though I do hope the generated code will end up being better than what we have right now. It mustn't get much worse. Can't afford all that RAM.

It'd be nice if CC could figure out all the required permissions upfront and then let you queue the job to run overnight

Anyone paying attention has known that demand for all type of compute than can run LLMs (i.e. GPUs, TPUs, hell even CPUs) was about to blow up, and will remain extremely large for years to come.

It's just HN that's full of "I hate AI" or wrong contrarian types who refuse to acknowledge this. They will fail to reap what they didn't sow and will starve in this brave new world.

People often compare working with AI agents to being something like a project manager.

I've been a project manager for years. I still work on some code myself, but most of it is done by the rest of the team.

On one hand, I have more bandwidth to think about how the overall application is serving the users, how the various pieces of the application fit together, overall consistency, etc. I think this is a useful role.

On the other hand, I definitely have felt mental atrophy from not working in the code. I still think; I still do things and write things and make decisions. But I feel mentally out of shape; I lack a certain sharpness that I perceived when I was more directly in tune with the code.

And I'm talking, all orthogonal to AI. This is just me as a project manager with other humans on the project.

I think there is truth to, well, operate at a higher level! Be more systems-minded, architecture-minded, etc. I think that's true. And there are surely interesting new problems to solve if we can work not on the level of writing programs, but wielding tools that write programs for us.

But I think there's also truth to the risk of losing something by giving up coding. Whether if that which might be lost is important to you or not, is your own decision, but I think the risk is real.

From an economic standpoint this is basically machines doing work humans used to do. We’ve already gone through this many times. We built machines that can make stuff orders of magnitude faster than humans, and nobody really argues we should preserve obsolete tools and techniques as a valued human craft. Obviously automation messes with jobs and identity for some people, but historically a large chunk of human labor just gets automated as the tech gets better. So I feel that arguing about whether automation is good or bad in the abstract is a bit beside the point. The more interesting question imho is how people and companies adapt to it, because it’s probably going to happen either way.

When I use Google maps, I learn faster.

And I haven't to solve real hard problems for ages.

Some people will have problems some will not.

Future will tell.

The crash and burn can't come soon enough.

Honestly my job is to ensure code quality and to protect the customer. I love working with claude code, it makes my life easier, but in no way would a team of agents improve code quality or speed up development. I would spend far too much time reviewing and fixing laziness and bad design decisions.

When you hear execs talking about AI, it's like listening to someone talk about how they bought some magic beans that will solve all their problems. IMO the only thing we have managed to do is spend alot more money on accelerated compute.

It would be tragically ironic if this post is AI generated.

I agree on all parts. I do not understand why anyone in the software industry would bend over backwards to show their work is worth less now.

>I'd bet anything these new models / agentic-tools are designed to optimize for token consumption.

You would think, but Claude Code has gotten incredibly more efficient over time. They are doing so much dogfooding with these things at this point that it makes more sense to optimize.

How Butlerian of you.

username checks out

Shaking fist at clouds!!

> Any self respecting engineer should recognize that these tools and models only serve to lower the value of your labor.

Depends on what the aim of your labor is. Is it typing on a keyboard, memorizing (or looking up) whether that function was verb_noun() or noun_verb(), etc? Then, yeah, these tools will lower your value. If your aim is to get things done, and generate value, then no, I don't think these tools will lower your value.

This isn't all that different from CNC machining. A CNC machinist can generate a whole lot more value than someone manually jogging X/Y/Z axes on an old manual mill. If you absolutely love spinning handwheels, then it sucks to be you. CNC definitely didn't lower the value of my brother's labor -- there's no way he'd be able to manually machine enough of his product (https://www.trtvault.com/) to support himself and his family.

> Using these things will fry your brain's ability to think through hard solutions.

CNC hasn't made machinists forget about basic principles, like when to use conventional vs climb milling, speeds and feeds, or whatever. Same thing with AI. Same thing with induction cooktops. Same thing with any tool. Lazy, incompetent people will do lazy, incompetent things with whatever they are given. Yes, an idiot with a power tool is dangerous, as that tool magnifies and accelerates the messes they were already destined to make. But that doesn't make power tools intrinsically bad.

> Do you want your competency to be correlated 1:1 to the quality and quantity of tokens you can afford (or be loaned!!)?

We are already dependent on electricity. If the power goes out, we work around that as best as we can. If you can't run your power tool, but you absolutely need to make progress on whatever it is you're working on, then you pick up a hand tool. If you're using AI and it stops working for whatever reason, you simply continue without it.

I really dislike this anti-AI rhetoric. Not because I want to advocate for AI, but because it distracts from the real issue: if your work is crap, that's on you. Blaming a category of tool as inherently bad (with guaranteed bad results) suggests that there are tools that are inherently good (with guaranteed good results). No. That's absolutely incorrect. It is people who fall on the spectrum of mediocrity-to-greatness, and the tools merely help or hinder them. If someone uses AI and generates a bunch of slop, the focus should be on that person's ineptitude and/or poor judgement.

We'd all be a lot better off if we held each other to higher standards, rather than complaining about tools as a way to signal superiority.

[flagged]