Type: url
Original: https://martinfowler.com/articles/harness-engineering.html
Status: partial
Summary
In this April 2026 article, which supersedes her February 2026 memo on the subject, Birgitta Böckeler proposes a mental model of harness engineering for users of coding agents: an outer harness of guides (feedforward controls) and sensors (feedback controls), each either computational (deterministic tools such as tests and linters) or inferential (LLM-based review), which humans steer by iterating on it whenever an issue recurs. She discusses distributing controls across the change lifecycle to keep quality left, distinguishes maintainability, architecture fitness and behaviour harnesses (calling the last the least solved), introduces harnessability and speculative harness templates tied to service topologies, and argues that a harness should direct human input to where it matters most rather than eliminate it.