Fragments: May 27

According to Martin Fowler's latest reflection, a powerful trend is emerging in how engineers work with AI coding agents. Ian Johnson documented restructuring a legacy codebase over three months with Claude Code, starting with tight oversight of every file edit. As he added tests, static analysis, and architectural patterns, his working relationship with the agent changed. He shifted from micromanaging to what he calls curation: defining patterns, reviewing test specs, making strategic decisions. The key insight? Agents can't be trusted until there's something forcing them to do the right thing. But there's a hidden cost. Adam Tornhill observes that agentic coding is mentally exhausting. The high decision density—making more choices in less time—maxes out cognitive capacity after about two hours, and he suspects many developers share this limit. Finally, Fowler notes recent controversy in the U-K, where the National Health Service closed open source repositories supposedly to block LLM security threats. The government's GDS team pushed back, arguing that hiding code reduces scrutiny and doesn't address underlying weaknesses.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton