Hallway Chat The two types of Agentic Engineering, and their teams
12 snips
Mar 26, 2026 They debate two engineering modes: long-running autonomous agents versus rapid, iterative prompting cycles. They explore multiplayer AI-assisted knowledge work through Nabeel’s Camp experiment. They argue team structure is shrinking from pods of six to pairs and discuss leaders needing to use AI tools themselves. They examine recommendation biases, Gemini’s strengths for webpages, and why data moats are fragile.
AI Snips
Chapters
Transcript
Episode notes
Start From Good Open Source Then Swap Infrastructure
- Use tasteful open-source starting points for good defaults, then replace infrastructure where needed.
- Fraser started from Conductor's Chorus, kept model routing, then replaced the backend with Convex for realtime and type safety.
Models Default To Safe Common Choices
- Models default to safe, commonly referenced choices and struggle to recommend niche best-in-class providers.
- Both hosts note models often lack deep domain knowledge and will pick the default unless guided by expert prompts.
Gemini Became A Bespoke Tailor For Clothing Choices
- Fraser became a heavy Gemini user because it handles URLs and web pages far better, acting like a bespoke tailor for clothing fit.
- He used Gemini to pick shirts, supplied photos, and iterated on size and washing instructions to get perfect results.