| status | active | ||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| updated | 2026-11-05 | ||||||||||||||||||||||||||||||||||||||||||
| section | Trust and Govern | ||||||||||||||||||||||||||||||||||||||||||
| audience |
|
||||||||||||||||||||||||||||||||||||||||||
| level | advanced | ||||||||||||||||||||||||||||||||||||||||||
| duration | 90 | ||||||||||||||||||||||||||||||||||||||||||
| format | core-talk | ||||||||||||||||||||||||||||||||||||||||||
| decision | What repository and CI infrastructure creates trustworthy agent throughput? | ||||||||||||||||||||||||||||||||||||||||||
| prerequisites |
|
||||||||||||||||||||||||||||||||||||||||||
| related |
|
||||||||||||||||||||||||||||||||||||||||||
| references |
|
The Question This Talk Answers: "How do I ship more with agents and still trust every change?"
Thesis: L3 is the launchpad: reliable context, repeatable checks, and habitual use create the conditions for outcome-focused governance and measured autonomy.
Duration: 90 minutes | Target Audience: Architects / Platform Engineers / Engineering Leaders
Deck: Open the native deck · Notes and handout: Read the speaker notes
Assess platform readiness beside adoption practice. L3 is where repeatable infrastructure, quality controls, and habitual use first compound. The notes carry the five-level criteria, caveats, and instruction hierarchy.
Give Copilot a bounded intent; let CI produce evidence; keep approval with named human owners. The section covers evidence bundles, policy gates, contextual triage, and recovery when a check or review disagrees.
Build fast feedback without sacrificing trustworthy checks. The notes distinguish scenario estimates from observed results and cover affected analysis, flaky tests, hermetic inputs, caching, and provenance.
Progress from the L3 foundation through L4 optimization to measured L5 autonomy. Each phase pairs platform work with adoption and governance signals; promotion follows evidence.
- Today: Rate platform and adoption separately; measure PR-check time and flake rate; count recent changes crossing repository boundaries.
- This week: Review repository instructions; require CI and CODEOWNERS for sensitive paths; choose one bounded agent task and define acceptance evidence.
- This month: Pilot affected analysis and cache invalidation; assign an owner to a quarantined flake; revisit the maturity assessment and choose the next constraint.
Apply the decision to one repository and record what the evidence says.
- FanHub walkthrough — a fork-based exercise for maturity assessment, fast-feedback CI, CODEOWNERS routing, evidence-oriented PRs, and baseline measures. It is an instructional artifact, not a reported run.
- The native deck includes repository-boundary diagrams, an illustrative intent contract, evidence-bundle contents, a staged CI model, and a three-phase checklist. Numeric scenarios inherited from the long README are identified in the slide notes as estimates or illustrations, not observed outcomes.
- GitHub Agentic Workflows — evidence-gated issue-to-PR automation.
- Local Agentic SDLC — specialist agents, worktree isolation, and integration records.
- Copilot Hooks — governance mechanisms and agent behavior controls.
- Enterprise Patterns — organization-wide scaling patterns.
- Unattended Agents — delegated work and oversight boundaries.
- The Agentic Operating Model — executive framing for a transformation plan.