MLOps.community Building Coding Agents: Design Decisions, Prompting Tricks, GUI Anti-patterns
68 snips
Aug 22, 2025 
In this engaging conversation, Beyang Liu, CTO of Sourcegraph, dives into how AI coding agents are transforming software development. He discusses the innovative AMP tool, which enhances code generation and streamlines user interactions. Beyang shares strategies top users employ to maximize agent efficiency, along with prompting techniques that elevate performance. The future of software interfaces is examined, moving from clicks to conversations, emphasizing natural language processing for a seamless user experience. The balancing act of user experience design is also explored, showcasing the complexities and innovations in coding tools.
AI Snips
Chapters
Transcript
Episode notes
Agentic Models Dramatically Increase Code Generation
- Tool-use plus reasoning models unlock agentic coding that generates most of a user's code.
- Beyang reports AMP users generate 80–95% of their code with these agents.
Agents Replace Manual Context Micromanagement
- Agents reduce manual context management because they fetch and use tools themselves.
- Beyang compares the shift to moving from Yahoo-style point-and-click to Google-style natural queries.
Give High-Level Context, Not Stepwise Orders
- Give agents high-level context and define the feedback loop rather than micromanaging every step.
- Let the agent discover and iterate on the right searches and checks itself.