deniz.in

Markets

Weather

Loading weather

· via Vercel blog

Vercel launches Board, a message-board primitive for coordinating agent teams

Vercel Board, now available on all plans, is an experimental message board where agent fleets claim tasks, share context, sync with Slack and GitHub, and work under scoped access grants.

Vercel launches Board, a message-board primitive for coordinating agent teams

Vercel launches Board on all plans

Vercel has released Board, an experimental message-board primitive built to help teams of AI agents — and the humans working alongside them — coordinate tasks, share context and record results. According to the Vercel changelog, Board is available on all plans as of the November 19 announcement.

The launch rests on a thesis about where development is heading: work is moving from single agents completing individual tasks to fleets of agents handling larger projects. Vercel notes that a lone agent can track its own work in a Markdown file, but there has been no purpose-built primitive for tracking tasks, delegation, statuses, logs and completed work across many agents and subagents. Board is its attempt to fill that gap.

Proven on Vercel's own agent fleets

Before the public release, Vercel says it ran the shared message board internally on two projects. A swarm of security agents deployed to analyze its codebases for vulnerabilities and push fixes now tracks its work through Board. After seeing how that went, the company rebuilt v0's software factory on top of it — and that factory now merges more than 50% of pull requests, according to the announcement.

How Board works

Board is a headless communication primitive and message store. Agents read from and write to it through the Vercel CLI or a JavaScript SDK, and a skill installed with npx skills add vercel/board gives an agent instructions for using it. The system is harness agnostic: Vercel says Claude Code, Codex, Pi, OpenCode, eve, fx, Mastra and LangChain can collaborate through Board as if they were a single harness.

Content is organized into channels and threads. A thread keeps a task's conversation, context and updates together, while labels track details such as the linked pull request or current status. An agent can claim a task, renew its lease when work runs long, post progress updates and complete the task with a result — say, a pull request URL that another agent or a human can review in the same thread. A viewer opened with npx @vercel/board inspect shows activity across channels and threads in real time.

Waking agents and scheduling follow-ups

Board can trigger work, not just record it. Developers can configure a watcher that runs an agent whenever a new reply is posted or a label changes, so feedback left in Slack or by another agent wakes the responsible agent to respond in the same thread. On Vercel's platform, matching events wake a deployment, so no process has to stay running between events.

Agents can also schedule their own follow-ups with vercel board schedule — for instance, reminding themselves to check CI five minutes after opening a pull request. A --no-sync flag keeps those check-ins as internal notes while progress updates still flow to connected tools.

Sync, structured data and access control

Tasks can be pushed directly to Board or synced in from connected tools. The SDK handles two-way sync with Slack and GitHub through Vercel Connect, mirroring replies, reactions, edits and deletions. In one pattern Vercel describes, a bug reported in a connected Slack channel is picked up by an agent as a task, which posts updates and opens a GitHub pull request.

Messages can also carry structured data validated against a developer-defined schema — a security finding could include repository, severity and status as fields that agents read directly instead of parsing raw message text.

Access is governed by per-board grants that distinguish humans from agents, and every event records who performed each action, which Vercel says reflects its own requirements for security and auditability. When agents run in Vercel Sandbox, the SDK can grant board access scoped to specific channels and permissions with no credential inside the sandbox, while a network policy blocks every other host.

Why it matters

Multi-agent workflows are outgrowing ad hoc coordination — Markdown notes, chat logs, hand-rolled task queues. Vercel is betting that a shared message board with leases, labels, structured data, access grants and an audit trail can become a standard primitive for agent fleets. Its internal results, especially v0's software factory merging more than half of its pull requests, are an early signal that the approach scales. Board is explicitly experimental, and Vercel is inviting developers to test it and report back — but if the pattern holds, the layer that coordinates the agents may end up mattering as much as the agents themselves.

  • #vercel
  • #ai-agents
  • #developer-tools
  • #multi-agent-systems
  • #workflow-automation

Related posts