Software Engineering Radio - the podcast for professional software developers SE Radio 711: Scott Hanselman on AI-Assisted Development Tools
80 snips
Mar 11, 2026 
Scott Hanselman, Microsoft VP of Developer Community and long-time dev advocate, blogger, and podcaster. He traces AI tools from autocomplete to agentic loops. He warns against vague prompts and shows why fundamentals let you steer models. He covers verification: tests, harnesses, sandboxing, and when to treat models as junior or senior collaborators.
AI Snips
Chapters
Transcript
Episode notes
Test Until You're Not Afraid To Ship
- Verify LLM-generated code with rigorous tests and CI; write enough tests until you're not afraid to ship.
- Hanselman used ~270 tests and 70% coverage for a personal blood sugar system to ensure safety.
Own And Review Agent Produced Code
- Always review and own code produced by agents; run the same security, provenance, and testing gates as for open source contributions.
- If code enters your repo, the human engineer remains ultimately accountable regardless of agent authorship.
Revitalizing Windows Live Writer With Iterative Specs
- Scott describes modernizing 20-year-old Windows Live Writer by dictating a 2.5 page spec, looping ~20 times, and getting a working port to .NET 10 with WebView2.
- He emphasized specifying DPI, COM-to-JS bridge, and plugin stubs so the agent knew success criteria.