Skip to content

Metrics for an AI-native SDLC

AI-native SDLC metrics pair a fast leading signal for each play with a lagging quality or reliability outcome. Engineering leaders can instrument the loop from Git, pull requests, CI, deployments, and incident records without buying a new analytics system or replacing the DORA measures they already use.

Read timestamps from git and pull-request metadata. Do not invent a new analytics product first.

PlayLeading indicatorLagging indicatorWhere it lives
Capture intent.mdTime from first conversation to a committed intent.mdSurvival rate (accepted into Design vs closed); edits to intent.md after the first spec.mdGit history on the intent home
Write spec.mdTime from intent.md commit to spec.md commitspec.md commits after the first plan.md for the same changeGit log
Plan modeShare of changes that merge from the first implementation pass; time from plan approval to merged PRRework cycles per change; how often the merged diff still matches plan.mdPR metadata
CLAUDE.md / rules / AGENTS.mdHow often the agent repeats a mistake the file should have caughtTime to first merged PR for a new teammateGit history of the instruction file; PR history
SkillsTime from policy-owner approval to the updated skill mergingPR review findings that cite the policy (should fall toward zero)PRs on the skill folder
Parallel sessionsConcurrent sessions per engineer while review quality holdsChanges merged per engineer per week, read with rework rateOpenTelemetry / vendor export; PR history
Feedback loopFirst-pass CI success rate for agent-written changesReview time per PR; change failure rateCI; PR metadata; incident tracker
Continuous evalsEval pass rate over time; time for a production incident to become a permanent evalRegressions caught in CI vs productionEval suite; incident tracker
PR reviewTime to first review; share of comments resolved without a human touching the branchDefects and vulnerabilities caught before merge vs productionGitHub/GitLab; incident tracker
Approval-gate hooksTime spent waiting on each gateGate violations reaching production, before vs after hooksHook telemetry; incident tracker
CI/CDShare of pipeline failures triaged without paging a humanDORA (deployment frequency, lead time, change fail rate, time to restore)CI and deploy tooling
Close the loopTime from band breach to an intent.md in triageShare of findings that become merged fixes; repeat incidents of the same classDetection script log; PR history
Scheduled scansShare of connected repositories on a schedule; time from finding to patch entering reviewVulnerabilities found by the scan vs production or external report; findings per scan over timeScan history; incident tracker