Running five AI coding agents means five terminals and copy-pasting context between them. MyPeople replaces that with one board and a persistent Boss agent: you give it the outcome, it writes the cards, assigns an owner, and reviews the work before it reaches you. The Terminal Graph shows every agent's live terminal wired to the card it owns. The HUD shows who is working and lets you attach to any of them from the browser. Claude Code, Codex and Grok run side by side, self-hosted in Docker.

I'm Dane, I build at Plow. For months my day looked like this: I was the only place my agents' work was tracked. Nothing was assigned, nothing had a status, and when a session ended the reasoning went with it; every morning I re-explained yesterday's task to a fresh agent. So I built the manager instead of getting better at being one. You write a card on a board. A Boss agent that stays alive triages it, assigns it to an engineer, and that engineer owns that card for its lifetime. Status, comments and proof land on the card as the work happens. A HUD shows every agent across every machine. And you can open any agent's terminal in a browser tab and read exactly what it's doing right now. It runs in Docker on your own machine. You log into your own Claude Code, Codex or Grok; Access the page to receive access!

One thing I'd love here is a way to pause or queue a card from the HUD without it auto-resuming, especially when I need a minute to think before an agent starts working. A simple hold toggle would let me batch up intent and then release it when I'm ready, instead of feeling like I have to babysit the board.

love that framing! The card ownership model is exactly how this should work. Congrats on shipping, Dane.

MyPeople

Put a boss on your Claude Code and Codex agents

Put a boss on your Claude Code and Codex agents