What makes it different
Most AI coding setups give you a fast way to write code. Gemba Flow gives you a supervised production line — the affordances of a factory manager, not an extra-fast typist.
| Differentiator | What it means in practice |
|---|---|
| Production-line paradigm | The affordances are those of a factory manager — a work queue that computes itself, WIP limits, gemba walks — not those of a code-review tool. Covers ideation through last-mile delivery, not just the coding step. |
| Encode intent, not prompts | Each bootstrap phase writes durable artifacts (PRD, architecture, agent configs) that downstream phases read. Your intent isn’t a prompt; it’s the DNA the whole system inherits from. |
| Defense-in-depth controls | Platform-level branch protection, account separation, Claude Code permission denies, and workflow conventions stack so no single layer is a single point of failure. |
| No two agents grab the same work | Claiming a work item either succeeds or fails — there is no race to lose. Validated across autonomous multi-worker runs with zero duplicate pickups, because assignment is a database operation, not a card someone remembers to move. |
| The queue can’t contain blocked work | Work items live in a dependency-aware tracker that ships inside the repo (ours is called beads), and “ready” is computed from the dependency graph — not a column anyone curates. When a blocker closes, the work behind it surfaces by itself. See the lifecycle. |
| The board can’t drift | The kanban you look at is a read-only rendering of tracker truth, regenerated automatically in under a second. Nothing writable exists to fall out of sync — why we designed it that way. |
| The audit trail writes itself | Work items close with structured proof — the pull request, the merge commit, the timestamp — recorded mechanically at the moment of completion, not reliant on humans filling in fields. |
Why this matters when agents are typing
A single agent session can produce hundreds of lines of code in minutes. Without structure, that speed becomes a liability: changes are too large to review, history is too noisy to audit, and the human’s attention becomes the single point of failure.
Gemba Flow distributes that load across the system — trunk-based branches keep diffs small, structured commits keep history legible, preview environments let you see the change before you spend attention reviewing it, and the agent roster keeps authority explicit.
Speed is only valuable when you can verify what was built. Gemba Flow keeps verification affordable.
Where to go next
- If you want the concept behind each differentiator: Encode Intent, Structured Handoffs, and Layered Controls cover the three in depth.
- If you want to see when it’s the wrong choice: Honest Limits names the cases where a different tool is right.
- If you’re ready to install: Quickstart.