The Engineering Room with Dave Farley Using AI Agents To TRULY INCREASE Software Engineering Productivity | David Yanacek In The Engineering Room Ep. 45
18 snips
May 3, 2026 David Yanacek, Senior Principal Engineer at AWS and lead advisor on Agentic AI, helped build DynamoDB, CloudWatch and the Kira IDE. He describes spec‑driven development that turned months into weeks. He explores testing agents with property tests and formal methods, using agents for proactive ops, setting safe agent boundaries, and balancing architecture with self‑verifying tests.
AI Snips
Chapters
Books
Transcript
Episode notes
AWS Rebuild Done By Six People In 76 Days
- David Yanacek describes an internal AWS re-architecture that was estimated at 18 months with 30 people but was completed in 76 days with six people using the Kiro IDE and spec-driven development.
- The example illustrates agentic AI plus spec-driven workflow turning a massive project into a small-team rapid delivery success.
Spec Drives Agent Focus And Quality
- Spec-driven development tames wandering agentic loops by producing a requirements doc, detailed design, and task list before coding begins.
- Kiro uses that spec to keep agents on a planned implementation path and generate scaffolding and mocks for early testing.
Generate Property Tests To Verify Agent Outputs
- Use property-based tests and simulators so agents can exhaustively check invariants across many inputs and states to keep non-deterministic outputs correct.
- Yanacek cites a traffic-light invariant example and says Kiro generates property tests to verify it under random state sequences.
