The eight dimensions
Each dimension is a place where speed can stall. Each has three statements, rated from 0 (not true) to 4 (fully true), and its score is their average.
D1Flow and constraintsThe path from request to customer use, including stages outside engineering.Red flag: Approval or release stages that wait days and were never touched by AID2Verification and evidenceTests, independent checks, AI review, and proof that a change is safe.Red flag: One model writes the code, the tests and the review; unreviewed merges risingD3Risk-tiered governanceWhich changes need which level of human involvement.Red flag: Either every change gets the same heavy review, or green CI alone means mergeD4Intent and specificationClarity of requirements, acceptance criteria and recorded decisions.Red flag: Vague tickets; tests written after the codeD5Environment and platformCI speed, repository scale, previews, rollback and observability.Red flag: CI wait grows with agent volume; no one-step rollbackD6Accountability and ownershipNamed humans for services and for release decisions.Red flag: Nobody can say who approves production for a serviceD7MeasurementOutcome metrics rather than activity metrics.Red flag: Dashboards show PR counts and AI-code share onlyD8People and changeSkills, roles, junior pipeline, incentives and training.Red flag: Reviewers overloaded; AI mandates without enablement; no path for juniors
How the scores combine
The binding constraint matters more than the average, so weights are not fixed and a strong dimension does not offset a weak one:
- The readiness level is capped at one above your weakest dimension.
- The autonomy floor is the lowest of D2, D3, D5 and D6: verification, governance, platform and ownership. It decides which autonomy tiers your evidence supports.