Wink Pings

Vercel Built a Task Board for Agents, but Don't Rush to Ditch Slack

Vercel released an experimental product called Board, a secure message board for humans and agents. It puts tasks, status, permissions, and PRs into the same channel. It is already used internally for security vulnerability scanning and the v0 software factory, the latter of which has merged more than 50% of PRs.

On November 18, Vercel released something called Board. Its positioning is straightforward: a secure message board, for humans and agents. It is currently an experimental product. Vercel is the deployment platform behind Next.js; this time it didn't ship a framework, it shipped a board.

The problem it aims to solve isn't complicated. A single agent can use a Markdown file to track tasks, but when multiple agents and sub-agents work together, there are no dedicated primitives for tasks, delegation, status, logs, and completed results. Vercel ran into this coordination problem internally too, so it tried treating a shared message board as a foundational component.

Two internal use cases are worth a look. One is a security agent cluster that analyzes codebase vulnerabilities, pushes fixes and updates; it now uses Board to track work. The other is v0's software factory. After being rebuilt on Board, it currently merges more than 50% of PRs. This figure comes from Vercel's blog post, not third-party statistics.

Board takes the form of a headless communication primitive and message store. Agents read and write through a CLI or JavaScript SDK. Each Board is organized by channels and threads; threads keep conversation, context, and updates together, while tags can record associated PRs or current status. Agents can claim tasks, post progress, and record results. For example, a coding agent can find an open bug, claim it, post a PR link, and then mark the task complete.

It also supports watchers. When there is a new reply or tag change, it can trigger an agent run. On Vercel, matching events wake up a deployment, so no process needs to stay running. Agents can also use vercel board schedule to schedule follow-ups, such as checking CI 5 minutes after opening a PR. This design makes agents seem more proactive, without waiting for a human to send another message.

Sync is another focus. Board can do two-way sync with Slack and GitHub; replies, reactions, edits, and deletions are all mirrored. Messages can also carry structured data validated by a schema, so agents can read fields directly instead of extracting them from natural language. On permissions, Board supports issuing grants to different people and agents and records who performed each action. When paired with Vercel Sandbox, you can avoid putting credentials in the sandbox and only expose specified channels and read/post permissions.

Real feedback is more interesting than launch copy. After Vercel's post, X user micah replied with profanity and a video, essentially about shipping something at 6 p.m. on a Wednesday. Salim said Claude Code refused to install Board and posted a screenshot. Rick Blalock said his first reaction was dislike; he wanted a tool that only surfaces decisions and gets him out of the weeds, but after looking into it he thought the "for agents" part was useful and planned to try it. eddie oakwood asked whether it was open source, saying he didn't see a GitHub link in the docs. David Sandy said his rule is that agents can post on the board but cannot have the company card. Draxx asked whether agents not hosted on Vercel can join.

Claude Code 拒绝安装 Board 的截图

These reactions show two things. Board is not just another human chat tool; it is more like a task ledger plus permission layer for agents. Humans can watch, audit, and intervene, but the core users are agents. The other is the old problem with experimental products: compatibility, whether it's open source, and how to connect non-Vercel environments all still require trying it yourself.

Vercel says Board is harness agnostic; Claude Code, Codex, Pi, OpenCode, eve, fx, Mastra, and LangChain can collaborate as if they were in the same harness. It provides two templates: one for an agent factory in Slack, and one for an agent cluster. The getting-started command is npx skills add vercel/board, and the viewer is npx @vercel/board inspect.

If you're building multi-agent workflows, Board is worth opening an experimental channel for. Don't throw out Slack just yet; put tasks, PRs, permissions, and logs into it and run it once to see whether it can get agents to ask you fewer questions.

Reference links:

发布时间: 2026-11-19 17:29