2026-08-30
Your coding agent needs a backlog, not another chat
Takeaways
- N agents without a queue is a project-management problem.
- A chat is where code happens. It is not where work is allowed.
- Beck holds the stories. Your Claude, Cursor, or Grok does the work.
You already know the feeling. Two agents in two windows. A third you started last night. Standup is a scavenger hunt. Nobody can say what is in review without opening five tabs.
The agent shipped something. You cannot find the story it was supposed to finish.
Ticket tools were built for people who click. A chat is a context window. It cannot “just know” what is allowed. When you run several coding agents, the missing object is not another model. It is a queue the whole team can see.
Chat is not a backlog
Chat is where the code happens. That is fine. The failure is treating the thread as the system of record. Progress lives in a transcript. Rejection lives in your head. The next agent invents a new task because it cannot see the old one.
Linear Agent’s answer is stay inside Linear: their agent, their UI, coding sessions on the issue. That is a coherent bet if you want the tracker to write the PR.
Beck’s bet is the other one. Your Claude, Cursor, or Grok already writes the code. They need the same stories a human would start. Named tokens. Recommend, start, log, finish. Ready for Review is the QA gate.
One list. Two kinds of workers.
Icebox holds what is not ready. The backlog is allowed work. Agents do not invent scope from a vibe in a thread. They pull a story that already has acceptance criteria, or a human points them at one.
Each bot gets a token named after itself. It shows up as the actor. Coordinator tokens can mint workers so you are not sharing one secret across a channel.
Install is ordinary MCP: npx @becklabs/beck-mcp-server against https://beck.bot. REST is the same objects under /api/v1. Read the MCP guide if you want the steps.
The list is the point
We do not claim to orchestrate N models better than a harness. We hold the work. If you do not need an agent in the queue, the backlog still works. The second bot is when you notice you never had a queue at all.
$12 if you’re shipping. Free if you’re looking. See Plans.
Questions
Can’t I just keep a markdown TODO?
You can. Then the agent updates a file on one machine and the team cannot see review or rejection.
Does Beck run the agent?
No. Cursor, Claude Code, Grok Bot, or anything MCP or HTTP does. Beck holds the stories.
What if I only have one agent?
The list still works. The second bot is when the missing queue becomes obvious.