Three Levels of Autonomous AI: Review, Coding Agent, Pipeline

Three Autonomy Levels

• **Level 1 — Autonomous PR Review** — ADO webhook triggers Lambda, AI analyzes the diff and posts structured comments with confidence scoring (>=90% MUST-FIX, >=80% SUGGESTION, <80% skip). Includes fix patches the author can accept with one click. Cost ~$0.15-0.50 per review.
• **Level 2 — Copilot Coding Agent** — Assign a GitHub issue to Copilot, it spins up a GitHub Actions environment, creates a branch, writes code, runs tests, and opens a draft PR. Reads both AGENTS.md and CLAUDE.md for context. Supports custom .agent.md agents and MCP servers.
• **Level 3 — Full Automation Pipeline** — ADO webhook triggers work item router, selects the right agent, runs complete workflow: DoR validation, dev implementation, 6-phase verification, PR creation, AI review. Runs 24/7 on AWS Lambda with token budgets and dead letter queues.

Choosing Your Level

1. 1 **Start with Level 1** — It's read-only (low risk) and high-value (instant feedback on every PR)
   • › No code changes, just structured review comments
   • › Confidence scoring filters noise automatically
   • › Cheapest to run, easiest to trust
2. 2 **Match level to team trust**
   • › Level 1 → read-only review, no code changes
   • › Level 2 → code changes but human merges the PR
   • › Level 3 → full autonomy with guardrails
3. 3 **Safety controls for Levels 2-3**
   • › Token budgets per run prevent runaway costs
   • › Capability gating limits what each project allows
   • › Dead letter queue catches failed jobs for human review
   • › Set up alerting before enabling any autonomous code generation