Type: url
Original: https://martinfowler.com/articles/exploring-gen-ai/humans-and-agents.html
Status: generated
Summary
In this March 2026 “Exploring Gen AI” article, Kief Morris distinguishes a “why loop”, in which humans turn ideas into working software, from a nested “how loop” of intermediate artefacts such as specs, code and tests, and compares three positions for humans: outside the loop (vibe coding), in the loop (inspecting what agents produce, which makes humans a bottleneck), and on the loop, where humans build and improve the harness that controls the how loop instead of fixing its artefacts. He argues that internal code quality still matters because it affects the time and cost of agent work, and describes an “agentic flywheel” in which agents evaluate the loop’s performance and recommend, and eventually apply, improvements to their own harness.