The Agent Board: Kanban for Machines; Durable Coordination for Agents That Outlive the Conversation
Organizations: Corvic AI Research
Abstract
Harnesses for large language models have converged on a single-threaded shape: one agent, one history, one machine, one sitting. That shape fails work that spans many agents, outlives any one model call, or waits on systems that are not agents. We describe the Agent Board, the coordination substrate in the Corvic harness. The board is a durable, shared table of semi-semantic tasks: a few exact fields the system routes on, and a free-form body that only language models read. Agents post tasks, watch tasks, and block on them, ending their turn and holding no process until a matching post wakes them. A system agent, the Board Keeper, acts as the interrupt handler for the board: it carries events from non-agent parts of the harness onto the board, matches them to waiting tasks by reading their prose, and allocates new agents for tasks that name a kind of agent rather than an instance. Loadable coordination policy kernels give statuses and roles meaning for a given pattern of work, without the board enforcing them. The design mirrors the kanban board a human team puts on a wall. We argue it is what persistent agent products such as Meta Muse, OpenAI dots and SpaceXAI Grok Bot need once an agent must wait, hand off work, or work with agents from another vendor. We view the board as an operating-system scheduler for agents and state where the analogy breaks. We separate what the mechanism guarantees from what it leaves to convention, and report a trace from a live board and three small simulated pilots. In the pilots, the wait cost of the board was flat in wait length while naive polling grew about 165-fold, but a plain external watcher was cheaper than the board. The Board Keeper matched prose waiters with no false wakes but missed 3.6 percent of real wakes.
Figures & tables
| Operating system | Agent Board |
|---|---|
| Thread | Agent |
| Time slice | Turn |
| Run queue / process table | The board |
| Block on a wait queue | Watch a task and end the turn |
| Wakeup | Notification |
| Condition variable | Watch on a task |
| Sub-call | Scratchpad | Wire proto. | DAG engine | Board | |
| Free waiting | – | – | ✓ | ✓ | |
| Durable across restart | – | ✓ | – | ✓ | ✓ |
| Peer-to-peer, not just parent | – | ✓ | ✓ | ✓ | ✓ |
| Receiver may not exist yet | – | – | ✓ | ||
| Non-agents can participate | – | – | ✓ | ✓ | |
| Plan may change mid-run | ✓ | ✓ | ✓ | – | ✓ |
| Field | Meaning |
|---|---|
| Identifier | The system-generated name of the task. |
| Topic | What the task is about; groups related tasks. |
| Priority | Rank within the topic. |
| Status | Planned, waiting, in progress, done, or failed. |
| Content | Free-form text, read only by LLMs. |
| Assigned agents | The agents the task is for. |
| Planned Build the accounts dashboard unassigned Check that the dashboard renders unassigned | Ready Score accounts by fit pipeline agent | In progress Enrich companies with industry and headcount pipeline agent | Review Merge duplicate accounts pipeline agent | Done Connect the CRM export pipeline agent |
| Mean tokens | 10 min | 1 h | 4 h | 12 h | 24 h |
|---|---|---|---|---|---|
| Naive poll | 273 | 925 | 6,169 | 41,273 | 152,025 |
| Stateless poll | 198 | 626 | 2,089 | 5,993 | 11,881 |
| Hold the call † | 288 | 893 | 874 | 884 | 888 |
| External watcher | 203 | 204 | 202 | 205 | 212 |
| Board | 915 | 922 | 947 | 897 | 923 |
| Precision | Recall | False wakes (of 50 negatives) | |
|---|---|---|---|
| Board Keeper (model) | 1.00 | 0.96 | 0 |
| Keyword baseline | 0.93 | 0.47 | 3 |
| Success (of 20) | |||
| Full history | 20 | 19 | 20 |
| Note, agent’s own choice | 20 | 15 | 1 |
| Note, with fact checklist | 20 | 20 | 20 |
| Vague note | 0 | 0 | 0 |
| Empty record | 0 | 0 | 0 |
| Resuming prompt (tokens), history | 635 | 2,081 | 8,308 |
| Kernel | Roles | Waits on |
|---|---|---|
| Pipeline wait | agent | the Board Keeper |
| Delegation | asker, worker | one task |
| Swarm | coordinator, workers | a topic |
| Escalation | agent, person | one task |
| Kanban | planner, worker, reviewer | a topic |
| Benchmark | What it measures |
|---|---|
| METR time horizon [ 31 ] | Task length, as human-expert time, at which a model succeeds with a given reliability. |
| Long-Horizon-Terminal-Bench [ 26 ] | 46 terminal tasks with partial-credit grading; reported runs average 239 episodes, 9.8M tokens and 88.9 minutes. |
| Vending-Bench [ 6 ] | Long-term coherence of one agent running a simulated business over runs above 20M tokens. |
| ParaCook [ 44 ] | Time efficiency of scheduling and coordinating agents in a parallel cooking simulation. |
| Gaia2 [ 17 ] | Agents in asynchronous environments with an event clock, temporal reasoning and multi-agent collaboration; a wait tool lets hours of simulated time run in minutes. |
| Handoff Debt [ 24 ] | Rediscovery cost when a coding agent takes over an interrupted task from a predecessor. |
| ID | Family | What the driver injects | Primary measures |
|---|---|---|---|
| S1 | Long wait | A job that finishes after of simulated time | Model calls, tokens and resident seconds as a function of |
| S2 | Race | The event fires before, during, and after the watch is set | Missed-wake rate |
| S3 | Crowd | concurrent waiters with prose conditions, events, and decoy events that answer no one | Wake precision and recall; duplicate wakes |
| S4 | Handoff | A fresh agent resumes with the task record only, with full history, and with nothing | Success, turns and tokens to resume |
| S5 | Approval | A person with a latency distribution who rejects the first draft with probability | Rework rounds; wakes per task; end-to-end time |
| S6 | Failure | A worker dies, a pipeline fails, a model refuses, or an awaited event never arrives | Silent-stall rate; time to surface |
Multi-Agent Coordination
Multi-agent coordination studies how autonomous agents align actions, share information, and divide work to achieve individual or collective goals.
See all 550 papers