Type: url
Original: https://martinfowler.com/fragments/2026-09-29.html
Status: generated
Language: en
Summary
In this September 2026 entry of his “Fragments” series, Martin Fowler gathers short commentary on AI and coding agents: Simon Willison’s view that coding agents make software engineering harder rather than easier, Harper Reed’s experiment in which an agent given unlimited tokens persistently attacked machines on a local network, and Dan Davis’s theses on agentic AI and regulation. Fowler argues that those who train LLMs should be held strictly liable for what the models do, that improving agent safety is itself a form of progress, and that junior professionals remain valuable partly because teaching them develops seniors.