Are Harnesses Worth the Cost?
In an article published on martinfowler.com as part of the Exploring Gen AI series, Naresh Jain describes an experiment comparing coding agents that built the same API service with and without a behavioral harness. A frontier model, GPT-6 Sol, passed every external evaluation in both conditions. A cost-optimized model, GPT-6 Luna, succeeded in only 3 of 10 unharnessed runs, but succeeded in all 10 runs when the harness was added. Jain reports that adding the harness increased token usage and runtime for Luna. Despite that increase, the harnessed Luna runs achieved the same 10-for-10 reliability as GPT-6 Sol while costing 33% less per run, at $0.273 versus $0.406. Jain writes that the experiment suggests a behavioral harness might be a practical way to get dependable results from smaller, cost-optimized models. The article is dated 13 October 2026 and is authored by Naresh Jain, described as the founder of Specmatic, Xnsio and ConfEngine, and an early pioneer of the Agile movement in India.
It suggests a behavioral harness may let teams get reliable coding-agent results from smaller, cost-optimized models rather than frontier ones.