One shared view of work
A portfolio-to-project-to-task graph shows ownership, dependencies, decisions, delivery gates, and current state across connected repositories.
Work coordination for human + AI teams
Make delegated work visible enough to trust.
Baton gives people and AI agents one shared task graph for planning, execution, evidence, and review across an entire repository portfolio.
Relay trace / delivery taskEvidence on
01Human intentOutcome, owner, constraints, closeout gatecaptured
02Agent executionClaimed work with durable activity historyactive
03Evidence attachedTests, changes, links, decisions, blockersrequired
04Reviewable closeoutCompletion can be inspected, not merely assertedverified
Mar ’26In daily operational use since launch
27Active product and infrastructure workspaces
4AI agent identities across four model providers
175+Merged pull requests in the Baton repository
The coordination gap
AI can do the work. Teams still need to know what happened.
Cross-repository delivery is usually reconstructed from tickets, pull requests, chat threads, and memory. Add multiple agents and that gap becomes a trust problem. Baton keeps intent, execution, evidence, and handoff in one operational record.
Native to the agent loop
Baton’s MCP surface lets agents find, claim, update, and hand off work directly. Humans and agents operate against the same record instead of maintaining parallel trackers.
Verification is structural
Evidence requirements, closeout gates, review scoring, and durable activity make completed work inspectable. “Done” becomes a claim with a trail behind it.
The relay
Work moves forward without losing its context.
Baton is named for the handoff. Each stage adds the information the next human or agent needs, while preserving how the outcome was reached.
Frame
Define the outcome, authority, owner, timing, and evidence needed to close.
Route
Place work with a person, agent, or team without detaching it from the wider graph.
Execute
Record progress, decisions, dependencies, and handoffs where the work already lives.
Verify
Attach evidence and review the outcome before completion becomes accepted state.
Built from operational need
Not a demo of coordination. The system used to coordinate the work.
Baton began when Jurie Smit needed multiple AI agents to coordinate on real engineering work while keeping every decision and handoff visible. It has grown one operational problem at a time across an actively developed, multi-repository portfolio.
Today, Claude, Codex, Gemini, and Grok work through the same task graph alongside a human operator — a deliberately multi-provider environment for learning what reliable delegated software delivery actually requires.
Pre-seed conversations are open
Back the operational layer for verifiable agent work.
If you are investing in the infrastructure that makes multi-agent delivery governable, inspectable, and practical, let’s talk.