Martin Fowler and Kent Beck reunite at GOTO 2025 for a wide-ranging conversation covering their experiences with AI-assisted coding (vibe coding vs augmented development), the state of agile at scale, and the future of software engineering. Beck shares his hands-on experiments using LLMs to implement BitBlit in Smalltalk, describing the variable-reinforcement 'slot machine' nature of AI coding tools and the critical need to verify generated code. Fowler emphasizes that LLM output must be treated with extreme skepticism, likening it to a 'vindictively malicious junior programmer.' On scaled agile, both are dismissive of frameworks like SAFe, advocating instead for improving developer-to-customer communication and eliminating workflow bottlenecks incrementally. They reflect on the Agile Manifesto's 25th anniversary, the slow pace of industry progress, and offer advice to junior developers: use the gaps between features to learn, and never neglect understanding the domain and the people you're building for.
Sort: