Writer research finds agentic harness efficiency beats model choice for cutting AI costs

Writer's research paper (arXiv:2607.06906) found that improving agentic harness efficiency cut costs an average of 40% across testing, outperforming model choice as a cost-reduction lever. The finding challenges the benchmark-driven release narrative dominating AI vendors. CEO May Habib said enterprise customers are "absolutely sick of chasing the next benchmark" and want flattening cost. The shipped harness upgrades launched August 13, 2026.