Software Unscripted AI & Software Quality with Shawn Wang (aka swyx)
33 snips
May 9, 2026 
Shawn Wang (swyx), AI practitioner and writer who builds coding agents and developer tools. He discusses integrating AI into IDEs, building browser agents and RAG systems, and the tradeoffs between agentic search and semantic indexing. Conversations cover subscription economics, software eras (1.0/2.0/1+2), and engineering challenges like memory, continual learning, and product quality.
AI Snips
Chapters
Transcript
Episode notes
Model Labs Ship Models Not Polished Apps
- Model labs prioritize core models; UI quality suffers because apps are considered secondary and often vibecoded quickly.
- Agent labs focus on polishing harnesses and UX because they must win the use-case layer.
Small Model Improvements Add Up Quickly
- Model quality improvements are incremental but additive; many small fixes (e.g., hallucination reduction) accumulate into meaningful user impact.
- Shawn still expects acceleration due to hardware scale and iterative model releases.
Use A 1+2 Software Approach For Practical Systems
- Combine rule-based (Software 1.0) and learned systems (Software 2.0) into a 1+2 approach where each orchestrates the other.
- Use hand-coded logic to orchestrate ML weights and ML to fill in tasks where rules are brittle.