Product··4 min read

Why your AI agent needs its own feedback board

Since shipping the MCP server, the teams getting the most out of it aren't the ones handing their agent the public board. They're the ones giving it a separate one.

When we shipped the MCP server, the obvious first move was pointing an agent at the same board your users see and letting it triage away. A few weeks in, that's not what the teams getting the most out of it are actually doing.

One board, two very different kinds of work

A user-facing board carries requests you're accountable for: things people asked for, voted on, and are watching. An agent working through MCP generates a different kind of item entirely, a dependency bump it noticed was overdue, a flaky test it wants tracked, a refactor it's three steps into. None of that is dishonest to log, but none of it belongs next to a feature request with forty votes either. Mixing them doesn't make the agent's work more visible, it makes the user's board harder to trust.

Give it a board of its own

The create_board tool exists for exactly this: an agent can spin up a workspace that's entirely its own, separate columns, separate everything, with create_feedback defaulting to Dev-only so anything it files stays off the public widget unless you say otherwise. Your users still see their bug reports and feature requests, sorted by votes, same as always. Your agent gets a backlog that's actually its own to work through.

  • A refactor backlog the agent maintains and works down between other tasks
  • Test flakiness or CI failures it noticed and wants to come back to
  • Dependency or migration work it's tracking across multiple sessions
  • Notes to itself mid-task that would otherwise just be scrollback

Visibility, not autopilot

None of this means the agent's board runs unsupervised. The value is that you can open it and see what it's been doing, the same way you'd check in on a teammate, instead of either trusting it blindly or making it justify every action in chat. Separate boards make that check-in cheap: you look at the agent's board when you want to, and the public board stays exactly what your users think it is.

Set it up

If you've already connected an agent through the MCP tab, ask it to create a board for its own work the next time it triages the inbox. If you haven't yet, the same command from the MCP server announcement gets you there in one line.

Start collecting feedback today

Free plan. 2-line integration. Your API key is ready the moment you sign up.

Get started free