Source record

Published by martinfowler.com (Exploring Gen AI series, incl. Böckeler) (T1 source). The original is at https://martinfowler.com/articles/are-harnesses-worth-the-cost.html.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

A Martin Fowler series experiment tests whether a behavioral harness improves reliability of a cost-optimized coding model. The harnessed smaller model matched the frontier model's 10/10 success rate at 33% lower cost per run ($0.273 vs $0.406).

Concrete cost-reliability data from a controlled experiment, useful for teams weighing frontier model spend against harness engineering for cheaper models.

Cited by

No citations on record.