
Most code review tools find what they were trained to find. Bob Council finds what five different experts would each find — in parallel, in seconds. Traditional review is single-threaded. A security engineer catches injection vulnerabilities. A perf engineer catches N+1s. An accessibility lead catches missing ARIA. You rarely get all three on the same PR. Issues slip through the gaps between specialties. Bob Council closes those gaps. Built entirely on IBM Bob, it uses Bob's Orchestrator mode to dispatch every PR diff to seven Custom Modes in parallel: five specialist reviewers (Security, Performance, Accessibility, Cost, Junior-Dev empathy), a Synthesizer that reconciles their findings into a verdict with a contentiousness score, and a Director that scripts a narrated walkthrough. The walkthrough plays in a static HTML viewer with text-to-speech narration synced to severity-coded code, an actionable Suggested Fix for every issue, and a printable Council Report exportable to PDF. Two end-to-end reviews ship in the demo — a JWT auth endpoint with SQL injection, hardcoded secrets, and plaintext password comparison; and a performance regression with N+1 queries and per-route connection pools. The proof of Bob use lives in the repo, not the pitch deck: 99.5% Bob Factor, 14 documented sessions, 7 custom modes, 1 orchestrator playbook, "Made with Bob" markers throughout. Receipts in docs/how-we-used-bob.md. Bob Council hits all three pillars of meaningful Bob application: contextual reasoning (every reviewer sees the full PR), repetitive automation (orchestration + narration are zero-touch), and knowledge transfer (every walkthrough becomes onboarding material). This isn't a chatbot wrapped around git diff. It's how code review should have worked all along.
17 May 2026