Software Engineering Radio - the podcast for professional software developers SE Radio 710: Marc Brooker on Spec-Driven AI Dev
Mar 4, 2026
Marc Brooker, VP and Distinguished Engineer at AWS with decades in software engineering, discusses specification-driven development for AI-assisted coding. He covers how crisp specs guide code generation, testing, and validation. Topics include Kiro (an agentic IDE), modularity and APIs to reduce context, agent roles for building and review, and practical rollout, metrics, and cultural readiness for adoption.
AI Snips
Chapters
Transcript
Episode notes
Aurora Driver Built From Spec With Property Tests
- Aurora D-SQL driver work used a spec listing API expectations, then generated the driver and property tests for invariants like every connection having an auth token.
- Property-based tests exercised thousands to millions of API permutations to check those properties.
Use AI For Drafting But Humans Pick Design
- Use AI to draft specs and designs but keep humans in the loop to choose architecture, APIs, and implementation tradeoffs.
- Marc describes a flow: define customer/problem, draft spec, pick service/library/UI design, choose language/frameworks, then generate code and tests iteratively.
Separate Builder And Reviewer Agents
- Use separate agents with distinct goals (builder vs reviewer) rather than self‑supervised single agents.
- Marc found a dedicated code-review agent focusing on maintainability and API quality improves long-term outcomes.