Flow without quality moves defects faster. Jidoka builds quality into the process by detecting an abnormality and stopping work before the defect travels downstream. The stop creates a chance for people to investigate and fix the cause, rather than asking the next stage to absorb the damage. section 11.7
In software, the “line” might be a build, a message queue, or a call to another service. The stopping point should sit close to where the problem can be detected. section 11.7 section 12.1.1
Choose a stop that matches the failure:
- Reject invalid input before processing it. Route a corrupted message to a dead-letter queue instead of retrying it indefinitely and blocking other messages. section 11.7
- Open a circuit breaker when calls to a dependency exceed a failure threshold. Fail immediately rather than adding load to a struggling service. section 11.7
- Run formatting, lint, type checks, and tests before defects reach later stages. A failing test stops the build; a regression caught on each push need not wait until production. section 13.4.3 section 12.1.1
Constrain behavior before generating code
For a shipping-cost change, first write a unit test asserting the expected cost for a domestic package under 5 kg. Implementation—including AI-generated implementation—must then satisfy that behavior. A fast, stable test stops a defect near the smallest unit of change; a test written afterward may merely confirm the choices already made. section 5.2
Automation supplies the stop, not the judgment. Everyone who changes a component owns the quality of what they leave behind; the developer who merges AI-generated code also owns its edge cases and integration assumptions. Ask where an abnormality first becomes visible, what should stop there, and who will investigate it. section 13.4.3 section 15.5
