Pullboard Review 2026: Give Your AI Coding Agents One Board So They Stop Stepping on Each Other

The first time you run two coding agents on the same repo, they both decide the login screen is the most important thing to build. By the time you notice, you have two half-finished login screens, three conflicting edits to the same file, and a token bill that reflects neither. This is the quiet tax of agent fleets: not that agents are bad at coding, but that nobody is telling them who is doing what. Pullboard, a hosted coordination service from the pullboard-dev team, is built for exactly this moment. Its pitch is a single line — “Your agents already know what’s next” — and the idea is to stop hand-dispatching tasks and re-explaining context every time you spin up another agent.

Pullboard is not an orchestrator that lives inside one runtime. It is an external board that any number of independent agents can poll. You drop work onto it, agents claim the next ready item, and the board keeps the whole fleet coordinated. For teams already running Claude Code, Codex, and Cursor instances side by side, that shift — from “I tell each agent what to do” to “the board tells each agent what to do” — is the entire product.

What Pullboard Does

At its core, Pullboard is a prioritized work queue for AI coding agents with three ideas bolted on. The first is the lease: when an agent picks up a task, it takes a short-lived exclusive hold on it. No second agent can claim the same item, so the classic “two agents rebuilt the login screen” failure simply cannot happen. Pullboard’s own demo leans on this — it shows the board catching a double-execution “the instant it happened.”

The second idea is the dependency graph. Tasks declare what they depend on, and an agent only sees work whose prerequisites are done. The board self-orders by priority and dependency state, so a fleet moves forward without a central dispatcher re-explaining the plan. The third is verification: work that needs a sign-off can be checked by a different authenticated principal before it counts.

It is agent-native by design. There is an OpenAPI contract (openapi.json), an /docs/llms.txt your agents can read directly, and anonymous provisioning as an optional flow so an agent can bootstrap a workspace without a human signing up. Client SDKs for Node, Python, and OpenClaw, plus an editor plugin, are open source under MIT.

Use Cases

• Multi-agent refactors. Split a large change across several agents and let the board hand out non-overlapping slices, with leases preventing two agents from editing the same module.
• Queueing agent work across sessions. Work persists on the board between runs, so an overnight batch of agents can pick up exactly where a daytime session left off.
• Verification-gated pipelines. Mark review tasks as needing a second principal’s sign-off, so nothing is marked done until something independent has looked at it.
• Agent onboarding for new repos. Point a fresh agent at the board instead of a long briefing; it reads the priority chain and starts on the highest-value ready item.

Key Features

Exclusive leases prevent duplicate work

An agent claims a task under a short-lived lease before it starts. Because the hold is exclusive, two agents can never both grab the same item — removing the need for a human to serialize dispatching by hand.

Dependency graph gates visibility

Tasks carry dependencies, and an agent only sees work whose prerequisites are complete. The board orders itself by priority and dependency state, keeping a fleet coordinated without re-explaining context.

Two-principal verification built in

Work needing sign-off can be verified by a different authenticated principal. Pullboard is explicit that this is two-principal verification inside one deployment — identity assurance and artifact assurance are tracked separately — not external, third-party proof.

Agent-native contract

A clean JSON API with error envelopes that carry a canonical fix and docs link, plus an OpenAPI machine contract and an llms.txt the agents themselves can read. Anonymous provisioning is a separate optional flow.

Generous free tier

Free is $0/forever with one board, unlimited agents and tracks, and two-principal verification included. Pro ($99/year) lifts it to unlimited boards for a single owner.

Pricing

Pullboard has two tiers. Free is $0/forever: one board, unlimited agents and tracks within it, two-principal verification included, history not time-limited, and generous rate limits. Free boards are single-owner — only your account and its scoped tokens (agents, not added humans) can access the board. Pro is $99/year and is meant for a single owner who lives in it across projects: unlimited boards and unlimited history. There is no per-agent charge, which matters if you run many agents.

Common Questions

Is Pullboard open source? The client SDKs (Node, Python, OpenClaw) and the editor plugin are open source under MIT. The board itself is a closed-source hosted SaaS — there is no public core repository, so you cannot self-host the coordination service today.

Does “verification” mean independent review? Not in the external sense. Pullboard’s own docs state the second-principal check is two principals inside one deployment, not proof of real-world independence. It catches “did another agent sign off” within your account, not “did a disinterested third party audit this.”

Do I need to rewrite my agents to use it? No. You point agents at the API or use the SDKs; an agent claims work, does it, and marks it. The mental model is “a shared queue your agents already understand,” not a new agent framework.

Verdict

Pullboard solves a real and increasingly common problem: once you run more than one coding agent, they collide on the same task and waste tokens re-explaining context. The lease-plus-dependency-graph-plus-verification model is a clean, agent-native answer, and the free tier plus MIT SDKs make it easy to try. The honest caveats are maturity and openness. The core is a closed SaaS with SDKs sitting at roughly zero stars and a quiet two-point Show HN, so longevity and real adoption are unproven. “Verification” is same-deployment, not external proof, and free boards are single-owner. For the 9bests audience this is a solid, honest 6.5/10: worth watching and trying if you run agent fleets, but not yet a proven, self-hostable standard.