The Changelog: Software Development, Open Source Agent, take the wheel (Interview)
170 snips
Jul 2, 2025 
In this enlightening discussion, Thorsten Ball, a software engineer at Sourcegraph, shares his insights on AI-assisted coding and the revolutionary Amp tool he’s working on. He explains how coding agents can transform software development and discusses the divide between AI believers and skeptics. Thorsten highlights recent advancements in AI, the unique features of Amp, and the necessity for adaptability in engineering skills. He delves into the impact of large language models, questioning traditional coding methods, and envisioning a future where AI enhances coding efficiency.
AI Snips
Chapters
Transcript
Episode notes
Build Light AI Scaffolding
- Build AI products with light scaffolding around models to easily adapt as they evolve. - Pick the best model automatically and avoid heavy user configuration to provide seamless experiences.
Agent Finishes Keystroke Recording
- Thorsten recounted using AMP to record his keystrokes and have the agent continue work on tests. - The agent understood 160 naive keystrokes and accurately completed refactoring, proving powerful human-AI collaboration.
Skepticism Rooted in Early Disappointments
- Skepticism arises where users try AI once, find it lacking, then dismiss it completely. - A major barrier is the learning curve; models need thoughtful prompting and user perseverance to be effective.