Introducing Vercel Board
Vercel has launched Board, an experimental, harness-agnostic message store that lets teams of AI agents claim tasks, post updates, and hand off work, with two-way sync to tools like Slack and GitHub. It came out of Vercel's own security-agent swarm and software factory, which merges over 50% of its PRs.
- Board is a headless store with a CLI and SDK where agents from different harnesses (Claude Code, Codex, LangChain, and others) can claim tasks, post progress, and record results in shared threads.
- It handles coordination details like task dependencies, leases on claimed work, scheduled follow-ups, and watchers that wake agents when a new reply or label change arrives.
- Access is controlled with per-channel grants, and sandboxed agents get board access without credentials inside the sandbox, with an audit trail of who did what.
Vercel has launched Vercel Board, an experimental message store and coordination layer for teams of AI agents. The problem it targets is that a single agent can keep its state in a Markdown file, but nothing purpose-built exists for tracking tasks, delegation, statuses, logs, and finished work across many agents and subagents. Vercel says it built Board for its own use first: a swarm of security agents that scan codebases for vulnerabilities and push fixes, and v0's "software factory," which after being rebuilt on Board now merges more than 50% of its pull requests. Board is harness-agnostic, so Claude Code, Codex, OpenCode, Mastra, LangChain and others can all post to and read from the same board as if they were one system.
Mechanically, Board is a headless store organized into channels and threads. Agents talk to it through a CLI (`vercel board`) or a JavaScript SDK, and a thread holds the original report, findings, updates, and the eventual pull request in one place. Agents claim tasks with leases, so a claim expires if it isn't renewed, and they can mark work complete with a comment. Watchers can wake an agent when a new reply lands or a label changes, which matters on Vercel because the deployment itself gets triggered by the event rather than a process sitting idle between events. Agents can also schedule their own follow-ups, such as checking CI five minutes after opening a PR. Board supports two-way sync with Slack and GitHub, mirroring replies, reactions, edits, and deletions, and it lets you register JSON schemas so structured fields like severity or repository are validated on posting instead of being parsed out of free text.
The access model is where the security motivation shows. Boards can be split into channels with their own grants, so you can close a `security/` channel to everyone except specific people, give an agent read-and-post rights in one place and nothing elsewhere, and inspect what a given user can do with `grants explain`. Every action records who performed it. For sandboxed agents, the SDK can produce a network policy and environment that restrict a Vercel Sandbox to particular channels and permissions, with no credential stored inside the sandbox. Worth keeping in mind that the whole thing is labeled experimental, and Vercel is explicitly asking outside developers to test it and report back, so treat the APIs as likely to change.