Changelog Interviews Flowing with agents
24 snips
Sep 17, 2025 
Beyang Liu, CTO at Sourcegraph and co-creator of AMP, dives into the fascinating world of agentic coding tools. He explains how AMP is engineered to leverage cutting-edge models for autonomous reasoning and complex task execution. The conversation touches on the architecture behind AMP, practical suggestions for workflows, and the importance of context windows in efficient prompting. Beyang also shares insights into the community's role in shaping the tool and upcoming features aimed at improving user experience. Perfect for anyone interested in the future of coding agents!
AI Snips
Chapters
Transcript
Episode notes
Multi-Model, Single Experience
- AMP is a coding agent that uses multiple models under the hood but hides model selection from the user to simplify the experience.
- The product chooses the best model per subtask to optimize latency and capability rather than exposing model toggles.
Cody vs AMP Origin Story
- Adam recounts earlier confusion about signing up for Cody and difficulties accessing it outside enterprise.
- Beyang clarifies Cody remains used for enterprise, while AMP was built for agentic workflows outside of enterprise constraints.
Always Review Generated Code
- Do always review generated code; code review remains the key bottleneck when using coding agents.
- Read emitted code to catch subtle bugs and ensure the agent matched your high-level intent.