From chat workflow to contract workflow
Teams often pair an IDE agent with Jira or Linear in the cloud. Work Graph is the tracker that lives in the repo: same board and statuses, plus intent in .bvc and proof per work.id. It complements the IDE; external PM sync is not the core product.
What changes compared to a plain AI workflow
Plain AI workflow
intent lives in heads or chats
done = agent words
Jira / Linear
plans work and statuses in the cloud
contract and evidence live outside the repo
CI / tests
checks commands
does not know why the task exists
Work Graph
links intent, execution and memory
done = evidence + verified gate
Competitor matrix
Work Graph sits at the intent-graph layer: goals, decisions, work items, evidence and verified memory in one local graph.
| Competitor | Layer | Evidence per task | WG stance |
|---|---|---|---|
| Cursor / Claude Code | Writes code and runs tools | partial | Execution layer; WG is the evidence ledger above it |
| GitHub Copilot | IDE completions and agent sessions | partial | Execution layer; WG scopes work.id and proof per change |
| Windsurf / Cline | Open IDE agents and edits | partial | Same as Cursor; WG supplies backlog and gates in git |
| Linear / Jira | Plans and tracks issues | no | PM layer; WG keeps local BVC contracts and proof |
| Asana / ClickUp | Team tasks and project tracking | no | PM layer; no machine-readable contract per task |
| Notion / Confluence | Docs and informal task lists | no | Knowledge layer; WG links decisions to executable work items |
| Langfuse / LangSmith | LLM traces and experiment logs | partial | Observability; WG ties traces to BVC checks and done |
| Mem0 | Stores agent memory | no | Memory layer; WG stores accountable work records |
| Devin | Autonomous ticket-to-PR sessions | partial | Cloud execution; WG defines acceptance and evidence |
When Work Graph fits
You want audit, not vibes
Tasks must cite files, commands and gates before done. Regulated, platform or long-lived codebases benefit most.
You already live in git
Backlog and evidence as files match PR review, forks and offline work. No vendor lock-in for the intent graph.
You run IDE agents
Cursor or Claude Code execute; WG supplies get_work_contract, evidence tools and PolicyViolation when proof is missing.
What Work Graph is not
Not enterprise cloud PM
WG is a local tracker in git, not a multi-tenant SaaS like Jira Cloud. Optional sync with external PM tools is outside the core scope.
Not an IDE
WG does not edit files or run models. It governs contracts around the edits your agent already makes.
Not fuzzy memory
Mem0-style recall is complementary. WG memory is closed work with evidence — accountable, not conversational.