GitHub Next opens up Chopin: planning with agents should be a team effort, not a monologue in the terminal

By GitHub Next · October 21, 2026.

GitHub Next, GitHub's research lab, has published and open-sourced Chopin, a real-time multiplayer planning environment with AI agents. It is authored by Krzysztof Cieślak and Maggie Appleton. According to the team itself, it is a very early research prototype: they do not recommend using it yet and expect to have alpha testers in about a month.

The idea is for several people and an agent to write a plan together in an editor based on MDX (an extension of Markdown that lets you embed interactive React components). Plans are organized around GitHub repositories and are visible to anyone with access to the repo.

According to GitHub Next, the prototype includes visual (WYSIWYG) editing with tabs, charts, diagrams and formulas, real-time user presence and comment threads for directing the agent. It also has a persistent decision log, a "research" block where the agent writes a report inside the plan, and an MCP server (Model Context Protocol, the standard for connecting tools to agents) that lets you use any agent. The agent behaves like any other user of the document.

Chopin is also the successor to Ace, a more ambitious multiplayer agentic engineering product that, according to the post, is on hold until further notice. The team notes that another group within GitHub is working on bringing collaborative planning documents to github.com, and that Chopin aims to influence that effort.

Their starting diagnosis is sound: with agents capable of implementing, development time shifts toward planning and reviewing. Their four criticisms of current tools are the lack of interactivity (a plan can only be accepted or rejected), of team collaboration, of traceability of who decided what, and of visual explanations. Their speed comparison, using GitHub Next's own figures, is illustrative: we read about 238 words per minute, while a frontier model generates about 4,500 tokens per minute, a ratio of 1 to 15.

Our take: the most valuable part of Chopin is not the editor but the third criticism. A decision log that distinguishes what a human decided from what an agent silently decided is governance applied to the day-to-day of software. When nobody reads the diffs, responsibility shifts to the plan, and a plan without authorship is a risk that is hard to audit.

Caution is also warranted. For now it is an early demo, and the post is a statement of intent by its authors, not evidence that this way of planning produces better software. The claim that senior developers have stopped reading diffs is, moreover, a description of part of the industry, not a measured figure.

Broadly, the move fits an industry trend: agent tools are shifting from the individual assistant to team collaboration, and competitive advantage is moving from the model to the "plumbing" (coordination, context, permissions, traceability). It is a thesis we already hold: whoever controls that fabric wins, not just whoever has the smartest model.

In the short term, this does not remove the underlying tension: if implementation gets cheaper, value and risk concentrate in those who know how to specify and judge, and that will change who is indispensable on a team. In the long term, humans and agents reasoning together about a problem, in a legible and auditable way, is the kind of infrastructure needed to delegate ever-larger tasks without losing control. Chopin is a sketch, but it points in the right direction.