When every developer or AI tool chooses a different way to handle the same routine task, small variations become review effort, debugging work, and maintenance burden. Standard work gives the team a shared baseline: the documented best-known way to perform a task under current conditions. It makes unexpected choices visible before they spread. section 3.4 section 4.3.3
A standard is not a prescribed method to apply regardless of context. It records what currently works, links to real examples, and leaves room for a reasoned exception. Repeated exceptions are evidence that the baseline may need to change. section 4.4.2 section 4.4.3
Put the baseline where the work happens:
- For AI-assisted changes, specify acceptable tasks, required contracts, prompt constraints, and where a human must decide behavior. Require a human-authored statement of intent and working tests before review. section 3.4 section 3.4.2
- For routine coding, document error handling, logging, validation, file structure, testing, and naming; link each rule to code that demonstrates it. section 4.4.2
- For operations, keep runbooks and deployment checks aligned with the system that actually runs. Automate repeatable checks where possible; retain human judgment where needed. section 12.4
One error-handling pattern, two features
Suppose a team standard says to use AppError types, fail fast at boundaries, and log context. Referencing that standard in prompts for both a payment change and a refund change narrows the AI’s choices. Reviewers can then check the changes against a known pattern instead of debating two plausible approaches. If a worker thread cannot use AppError, document why it differs rather than hiding the exception. section 4.3.3 section 4.4.2
Listing 4.10 Standard work document
## Error Handling Standard Use AppError base class with specific error types. Example: src/api/users/createUser.ts:45-52 Pattern: Fail fast at boundaries, throw with context Don't: Return null/undefined, use generic Error, swallow errors silently
A short error-handling standard links the rule to an existing implementation and names patterns to avoid. section 4.4.2
Standard work protects end-to-end flow, not consistency for its own sake. Clear intent and constrained changes keep avoidable uncertainty out of pull requests. Review checks compliance, but also supplies feedback: recurring, well-justified deviations call for investigation and possibly a revised standard. The same principle applies after incidents. Fixing one migration helps once; updating its template can make the safer step the default for future migrations. section 3.4.3 section 4.4.3 section 8.5
