Each step builds on the last. By the end, every AI-assisted action in your org is tracked, validated, and visible to leadership.
An automated scanner analyzes your repos for existing AI tooling, git hooks, CI workflows, and eval infrastructure. A structured SA interview captures organizational context. Together, they map you to a PRISM maturity level (L1–L5) and route you to the right onboarding track.
Set up Claude Code backed by Amazon Bedrock — every AI interaction stays inside your AWS account with CloudTrail audit logging. Configure CLAUDE.md project rules so the AI follows your team's conventions. Install Kiro IDE for spec authoring.
Write specs before code. Kiro-format specs define requirements, acceptance criteria, and API contracts in markdown. Claude Code implements against the spec. The result: reproducible AI output, reviewable artifacts in git, and no more prompt-iterate-revert cycles.
Git hooks auto-tag every commit with AI-origin metadata (tool, model, confidence, session ID). GitHub Actions extract trailers on push and emit structured events to Amazon EventBridge. You get traditional DORA metrics plus 6 AI-native dimensions from day one.
An LLM-as-Judge pattern using Bedrock Evaluations validates AI output against your specs before merge. Structured rubrics catch hallucinated APIs, spec mismatches, missing error handling, and logic errors — automatically, in every PR.
Two dashboards for two audiences. Team velocity dashboards in CloudWatch give engineers real-time, per-repo drill-downs. Executive readout dashboards in QuickSight give leadership weekly trends, cost efficiency, and AI ROI — real metrics, not vibes.
Teams inherit the PRISM bootstrapper: CLAUDE.md templates, spec templates, eval harness, GitHub workflows, and git hooks. Plug it into any repo. Weekly DORA assessments track maturity progression automatically as practices spread across your org.