Rollout Matrix - Capability Rules by Phase

This note explains how the capability model should be applied across the lifecycle phases.

Phase 1. Alignment and boundaries

• define E1, E2, and E3 usage expectations
• define R3 work that requires stronger review regardless of oversight-readiness band
• communicate that title alone does not determine safe AI usage

Phase 2. Baseline and workflow discovery

• identify where E1 engineers are currently overusing AI on unfamiliar work
• identify where E3 engineers can safely pilot stronger acceleration
• classify pilot workflows by familiarity, risk, and verification difficulty

Phase 3. Paired-engineering pilot

• E1: explanation-first, attempt-first, no default full-solution generation on new work
• E2: guided acceleration with explanation and modification requirements
• E3: stronger leverage on bounded work, but still accountable for review quality

Phase 4. Capability-aware expansion

• formalize different workflow rules by capability band
• add stronger review rules for R3 work
• treat low-verifiability work as elevated risk even when the surface task looks lightweight
• prevent seniors from normalizing unsafe habits that juniors will imitate without equivalent judgment

Phase 5. Standards, self-service, and internal platform support

• encode capability-aware rules into templates, standards, and examples
• encode verification checks and escalation paths into the templates, not just prompt patterns
• make learning-mode and delivery-mode workflows clearly distinct
• use Verification Standards by Artifact and Work Type so verification requirements vary by artifact rather than collapsing into a generic review rule

Phase 6. Measure, adjust, and decide whether to scale

• check whether capability-aware rules were actually followed
• review whether lower-oversight-readiness engineers are growing toward independent judgment
• review whether higher-leverage usage is creating hidden review debt
• review where verification difficulty was underestimated during the pilot