AGENTIC COMMONSAI industry briefings

繁中EN

TopicAI Chips & ComputePublished

All English articles

Vercel Board: a Shared Message Board for Agent Teams

Summary: Vercel launched Vercel Board on November 18, 2026 as an experimental agentic primitive: a headless message store that lets teams of agents claim tasks, post updates, and hand work to each other across tools. Agents reach it through a CLI or JavaScript SDK, and Claude Code, Codex, Pi, OpenCode, eve, fx, Mastra, and LangChain can all use the same board. It matters because multi-agent teams have been coordinating through Markdown files and Slack threads with no purpose-built primitive underneath.

What actually changed

Vercel’s own framing: a single agent can track its work in a Markdown file, but there was no primitive for tracking tasks, delegation, statuses, and logs across many agents. Board fills that gap with channels and threads. A thread keeps the original report, findings, and pull request together; labels track status; and an agent claims a task, posts progress, and settles the result in the same thread:

vercel board tasks claim --ready dev --label kind=bug --wait 10m
vercel board tasks complete <id> "Can I get a review? <pr-url>"

Two internal deployments prompted the build. Vercel ran a swarm of security agents to analyze its codebases for vulnerabilities and now tracks their work on Board. After that, Vercel rebuilt v0’s software factory on it and reports it “now merges over 50% of PRs” — a vendor’s self-reported figure with no stated methodology or baseline.

The mechanics that make coordination work

Three features do most of the heavy lifting.

Watchers and scheduled follow-ups. A watcher runs an agent when a new reply lands or a label changes, so feedback from Slack or another agent wakes the right agent automatically. On Vercel, matching events wake the deployment, meaning no process has to stay running between events. Agents can also schedule their own follow-ups — after opening a pull request, one can tell itself to check CI in five minutes with vercel board schedule and --no-sync, keeping internal notes off connected tools.

Two-way sync. Tasks can start in Slack or GitHub and mirror into Board, with replies, reactions, edits, and deletions synced both ways. A teammate reports a bug in Slack, an agent claims it, opens a GitHub pull request, and Board keeps the whole trail in one thread.

Structured data. Messages can carry schema-validated fields — a security finding can include repository, severity, and status — so downstream agents read fields directly instead of parsing message text. Schemas are versioned, like security-finding@1.

Limits before autonomy

The access model is the part Vercel says it built for its own security needs, and it maps cleanly onto what we’d want before giving agents autonomy: per-user and per-agent grants scoped to channel trees (security/=read,post lets a production agent read and post and nothing else), and every board’s events capture who performed each action. Sandboxes get board access with no credential inside them, limited to chosen channels, with an explicit network allowlist.

Task dependencies round this out: a review task stays unavailable until its fix completes, while independent fixes proceed in parallel — which is what a swarm needs to know what can start and what must wait.

Our take: If you run two or more agents on different harnesses today and coordinate them through shared files or Slack, this is worth a scoped experiment now: it is usable today via CLI and SDK, not a waitlist, and the grant-and-audit model already meets the limits-before-autonomy bar we’d otherwise bolt on ourselves. Assume the coordination overhead you’re replacing (hand-off slips, duplicate claims) is real — that’s the bet. Treat the “merges over 50% of PRs” figure as a reason to test, not a reason to switch: it is Vercel’s own number on Vercel’s own workload. Teams running a single agent, or none, can ignore this — a shared board solves a problem they don’t have. Two things would change the verdict: an independent team matching these coordination gains on a plain issue tracker plus webhooks, or critical features ending up tied to Vercel’s platform in ways that make the board hard to leave.

Where it fits

Board is the latest piece of Vercel’s agent stack, and the pattern it targets — agents claiming work, posting results, waiting on approval — matches what the company has been building toward across its platform. To start, Vercel suggests installing the Board skill (npx skills add vercel/board), reading the docs, or pasting its getting-started prompt into an agent. Vercel built Board for itself and explicitly invites curious developers to test it and give feedback; the sources give no pricing, no usage limits, and no independent test of the coordination gains, which is exactly why a small reversible pilot on one real workflow beats a migration.

Sources

Original post
AGENTIC COMMONSOperated by PHLEGON LABS
ShareXEmail
Support us

Related reading

  1. Mercury 3 Hits Vercel AI Gateway at Launch Pricing

    Vercel added Inception's diffusion-based Mercury 3 to AI Gateway with 50% off launch pricing, 1M-token context, and 1,000+ tokens per second.

    Vercel

  2. Vercel Drives Goes GA: Agent State That Outlives Sandboxes

    Vercel Drives reaches general availability, giving agents persistent, mountable, forkable storage across sandbox runs.

    Vercel

  3. Cloudflare Web Search API in AI Gateway

    Gateway users should consider search integration: three launch partners share billing and access controls, but server tools are still pending.

    Cloudflare