A kanban board for AI agents is a task board where a card can be handed straight to a coding agent, and the agent’s answer comes back on the same card. In AgentVera the board is shared with your organization: you write the task on a card, send it to Claude Code, Codex or another CLI agent running on your computer, and the card moves to In progress. When the agent’s turn ends, its last reply is posted as a comment and the card moves on to a Test or Review column for a human to check.
This post explains why a board fits agent work, how to set one up, how sending a card to an agent works step by step and what the rest of your team sees.
Why use a kanban board for AI coding agents
Once you run more than one agent, the hard part is no longer typing prompts. It is remembering which agent is doing what, what is finished and what still needs a review. A chat window does not answer that; a board does.
| Without a board | With a board |
|---|---|
| Tasks live in terminal history | Every task is a card with a title, description and checklist |
| You copy a prompt into each agent by hand | You pick an agent on the card and the text is sent for you |
| Results scroll away | The agent’s summary is saved as a card comment |
| Team does not know what agents touched | Teammates see the card move and read the comment |
The board also keeps humans and agents on one list. A card can be done by a person, by an agent, or by an agent first and a person second.
Set up the board in AgentVera
- Sign in to your AgentVera account. The board is shared with your organization, so it needs an account. The Kanban button in the top bar opens it.
- Link the local project to a cloud project. Pick an existing project in your organization or create a new one. Only the link is stored; the server never knows your local folder.
- Shape the columns. A new board starts with To do, In progress, Test and Done. You can rename lists and add new ones.
- Write cards. A card has a rich-text description, labels, members, a due date, checklists, an activity log and attachments up to 25 MB, which you can drag in or paste.
Changes show up live for everyone in the organization, and the “Assigned to me” filter narrows the board to your own cards.
Send a card to an agent, step by step
Open a card and find its Agent section. The hint there sums up the whole flow: the card goes as a task to an agent on this computer, moves to “In progress” and a summary is posted as a comment when done.
- Choose the agent. You see the project’s running agents, grouped by agent group, with their status: ready, busy or awaiting approval. Agents waiting for an approval are disabled.
- Or start a new one. Choose “Start a new agent”, pick the agent kind from the CLIs installed on your computer (Claude and Codex come first) and press Start. The new agent is named after the card, so you can spot it in the grid.
- The task is sent. The card title, description, checklist and comments go to the agent as its task. Files attached to the card are downloaded next to the project so the agent can read them. An agent on an SSH host gets only the file names.
- The card moves. A comment notes which agent took it, and the card moves to In progress.
- The turn ends. AgentVera posts a comment saying the agent finished its turn, with its last reply (up to 4,000 characters). If the board has a column named Test, Review or Code review, the card moves there. If the agent stopped with an error, the comment says so and the card stays put.
If the agent you picked is busy, the card waits. A busy agent picks it up after finishing its current work, so you can queue several cards on one agent.
What makes a good agent card
An agent reads the card the way it reads a prompt, so write it like one:
- One outcome per card. “Add rate limiting to the login endpoint” beats “improve auth”.
- Use the checklist for acceptance criteria. The agent gets the checklist too, and a reviewer can tick it off.
- Attach the evidence. A screenshot, a log file or a sample input on the card is downloaded for the agent.
- Keep internal notes as internal notes. Comments can be marked as internal notes for your team.
For longer work, split the card and give each piece to its own agent. Running each agent in its own git worktree keeps them from editing the same files; see git worktrees for parallel agents.
Review the result on the board
Moving the finished card to Test instead of Done is deliberate: an agent finishing its turn is not the same as the work being right. The comment tells the reviewer what the agent says it did, and the agent itself is still open in AgentVera with its diff. Read our AI code review guide for how automatic review checks each turn.
When the reviewer is happy, they drag the card to Done. If not, they add a comment and send the card to an agent again.
Kanban for the whole team
The board is for people as much as for agents.
- Live sharing. Everyone in your organization sees cards move and comments arrive.
- Public boards. A board can be shared through a public link, and guests can sign in with an e-mail link.
- Client boards. A client board starts with Requests, Planned, In progress and Done, and “Transfer to development” sends a card to the development board, with its status reflected back on the client card.
- Phone. The AgentVera mobile app has a Boards screen on Pro and Team, where you can read cards and send one to an agent. The apps are not in the app stores yet.
- Webhooks. An outside service such as GitHub, Sentry or Jira can create a card and an agent task by calling a webhook, and the result is written back to the card. See webhooks.
On the Team plan, team sync shares which repo, branch and changed file paths each teammate is on and warns when two people work on the same file.
Kanban, tasks and the task line
AgentVera has three ways to organize agent work. Pick by who needs to see it.
| Tool | Best for | Shared |
|---|---|---|
| Kanban | Team work, cards for people and agents | Your organization |
| Tasks and reminders | Your own to-do list next to the project | Only you |
| Task line | One job through stages such as analysis, plan, build and review | Only you |
The board is where work is agreed and tracked. The task line is how one big card can be pushed through several agents in order. For bigger splits, a lead agent can plan a team of agents for you.
Plans and limits
| Plan | Kanban |
|---|---|
| Free | No team board |
| Pro | Up to 10 boards |
| Team | Unlimited boards |
Running agents side by side, the terminal grid and many other core features stay free. See pricing for the full list.
Get started
Download AgentVera, sign in, open a project and press Kanban in the top bar. Link the project, write your first card and send it to an agent. When you want to run several agents on cards at once, our guide to running multiple Claude Code agents covers the setup.