Databricks smart routing cuts agent cost 30%

Databricks says its AI Gateway Smart Router cut average coding-agent task cost by more than 30% while roughly matching the priciest model’s quality. Its coding-cost playbook also reports that harness and cache tuning reduced generated tokens and associated cost almost 50% without observed quality loss, although both results are internal measurements rather than independent benchmarks. For teams weighing whether an AI gateway can clear its break-even point, the operator lesson is to measure cost per completed task before buying more seats: model routing and context hygiene may reclaim more budget than a blunt usage cap.