.NET Rocks! Building Software using Squad with Brady Gaster
Mar 19, 2026
Brady Gaster, a software engineer and product lead behind GitHub Copilot and developer tools, created Squad to coordinate specialist AI agents for app development. He discusses building focused agents that scan repos, debate and reach consensus, manage memory and skills, and speed up demo app generation. The conversation highlights token costs, parallel agents, reuse across projects, and practical limits like latency.
AI Snips
Chapters
Transcript
Episode notes
Specialist Agents Reduce Hallucination And Improve Quality
- Squad composes many specialist agents to keep each agent's context small and focused, reducing hallucination and improving code quality.
- Brady built agents that scour GitHub, blogs, and Reddit per product so each agent becomes an expert on its domain.
Rapid Product Analysis Generated 200 Markdown Files
- Brady tested squads by asking coordinated agents to analyze multiple products and they produced structured reports in minutes.
- He spun up agents across product repos and got 200 markdown files within 20 minutes during early experiments.
Lock Failed Attempts And Force Cross‑Agent Learning
- Avoid letting one agent retry endlessly; lock failed attempts and require another agent to try so knowledge is shared.
- Squad agents write to each other's history so repo-specific memory improves over time.