Birgitta Böckeler

Birgitta Böckeler works at Thoughtworks as a Distinguished Engineer specializing in AI-assisted delivery, with more than twenty years in software development, architecture and technical leadership. On 2 April 2026 she published «Harness engineering for coding agent users», which proposed the distinction the field had been missing: the internal harness, built by whoever makes the agent, versus the external harness that users assemble for their own context.

Her framework rests on two control mechanisms. Guides act feedforward, steering the agent before it acts through coding conventions, specifications, skills and «MCP» servers. Sensors act feedback, observing after the fact through linters, tests, static analysis and AI code review. Each can be computational (deterministic, fast, reliable) or inferential (semantic, costlier, non-deterministic). Both are required: without feedback the agent repeats its errors, without feedforward it encodes rules with no evidence they work.

Böckeler also names the limits. Maintainability harnesses are the mature category because the tooling already exists. Behavioural harnesses remain the hard problem, since AI-generated test suites are not yet reliable enough to answer whether the application does what it should. Her notion of harnessability adds that codebases are not equally amenable to any of this: typed languages, abstract frameworks and modular architectures offer more purchase.

When Addy Osmani consolidated the field into doctrine on 19 April 2026, Böckeler was one of the four sources he credited with having published the ideas separately since early 2026. She herself concludes that human experience remains an irreplaceable implicit harness.