Discussion
Prepare Claude agent-team layoutsin one terminal.
bwestergard: Looks like a great implementation. I want to question the basic user story, which seems to be: "I am a software developer who wants to improve productivity by running multiple simultaneous agents that are roughly isomorphic to a human software developer team."I am burning a lot of tokens every day at work and on personal projects. It's helpful. I generally work in tmux with github copilot in one pane, and a few other terminal panes showing tests and current diff.I find it really important to avoid the temptation to multi-task by running multiple agents. For quite varied tasks, productivity gains from multi-tasking have proven to be illusory. Why would it be different with writing software?https://en.wikipedia.org/wiki/Human_multitasking
theturtletalks: I'm also trying to build something similar for agent orchestration where one terminal is controlling multiple terminals. I tried using tmux but it's very good at sending the initial text to the tmux sessions, but I've not been able to get an agent to have a proper back and forth controlling multiple tmux sessions. I know we can use send-keys, but reading the session or knowing when that session is complete is kind of up in the air. And then if the main orchestrator terminal has checked all the sessions to see if they're actually working and doing things, the main session kind of stop so I've kind of been thinking about a cron that periodically checks in and nudges it to check the sessions again. Are they still working? Do they need more guidance? Essentially having one terminal control others, but having that back and forth with the terminals has been pretty challenging to achieve. Have you gotten anywhere with this?
mlboss: Can somebody develop a mobile app that natively supports tmux
deadbabe: For all the hype of AI agents, you never see people taking on real challenging projects like this. Just low hanging fruit.
cyrusradfar: Congrats on getting this out. What was the most surprising part of the build?
0dayman: https://cmux.com/
ekropotin: So basically tmuxinator?
operatingthetan: If this supported Gemini and Codex I would find it useful. I never run more than one Claude, it's always a mix.
AtxWrk70: the token costs are real. we switched to smaller models for 80% of tasks and barely noticed
selixe_: Interesting, but I wonder if this shifts too much complexity onto the user.tmux is powerful, but not exactly approachable, and "multi-agent orchestration" on top of it feels like something that could get hard to reason about quickly. Curious how you think about UX here.
j45: Tmux is pretty easy to pick up and build muscle memory by learning a few keyboard shortcuts from a basic youtube video and it's handy when you don't want to switch screens between multiple terminals just for one thing.The abilty to split and divide the screen pretty simply with a few keys is handy, and the abilty to save that layout for the shell items you're using to load up easily again the next time is valuable too.Multi agent orchestration likely just means keeping track of a few different windows all on one screen.
thijsverreck: Good points and indeed thinking about this quite a bit. Currently leaning towards a CLI first approach so that Claude/Cursor/[insert coding agent] can configure and control the ide. Feels a bit meta, but also makes it extremely user-friendly.
