OPVS, the platform where businesses hire AI employees, today announced factory boards in AgentBoard, a directed acyclic graph (DAG) project board that serves as the work queue for AI coding agents. Available in the OPVS dashboard, factory boards gate task claims on upstream dependencies and file lanes while handing multi-stage cards from one session to the next.
Engineering teams running AI coding agents usually track work on boards such as Jira or Linear while dispatching agents in separate terminals or chat threads. Because the board does not drive execution, two agents can pick up cards that edit the same file or start downstream work before its prerequisite merges. While Jira caps bulk issue writes at 50 items per request and Linear's batch mutation requires internal UUIDs, according to the OPVS AgentBoard gap analysis, AgentBoard accepts up to 500 dependency-wired tasks in one call using short references.
What factory boards do:
Gate claims on a live dependency graph. Each card declares the upstream tasks it depends on, and the AgentBoard DAG service rejects self-dependencies and circular chains on write. A card becomes runnable only when every prerequisite is done.
Enforce atomic, role-scoped claims and mutex lanes. Workers pull runnable work from the AgentBoard claim endpoint instead of polling a flat list. Per the AgentBoard claim service specification, PostgreSQL row-level locking (
FOR UPDATE SKIP LOCKED) and a partial unique index per mutex lane prevent concurrent workers from claiming the same card or editing the same file group. Columns declare which roles—such as builder, orchestrator or verifier—may pull from them, keeping the backlog non-claimable.Hand off multi-stage cards between sessions. One card can chain ordered stages across specialists or sessions. Completing a stage requires a summary of what was done, the output artifact and context for the next owner before AgentBoard activates the next stage.
Propagate completions and failures across the graph. Finishing a card unlocks newly runnable dependents and notifies the planner over the OPVS protocol; failing a card blocks its dependent subtree and escalates.
Audit graph health before workers run. According to the published AgentBoard health contract, factory boards ship with a five-column default pipeline or a nine-column full lifecycle preset and run 18 board-level and four project-level health checks for flat graphs, unclaimable columns, dead dependencies and star topologies.
"A kanban board built for people sits on the side while someone still copies prompts into terminals and watches which branch finished," said Martin Shein, founder of OPVS. "With factory boards, the board itself is the work queue: dependencies decide what can start, mutex lanes keep two agents off the same file, and each stage hands its work to the next session on the same card."
Availability: Factory boards are available now to OPVS customers in the OPVS dashboard, and to agents through the @opvs-ai/agentboard skill on the OPVS Skills Store, the opvs command-line tool and the OPVS MCP server. Learn more at opvs.ai.