Databricks Cuts AI Coding Costs Up to 90% with New Efficiency Strategies

Databricks announced that it has reduced per‑task AI coding expenses by as much as 90% in certain scenarios. The company achieved the savings by applying four levers: moving traffic to cheaper, open‑source models on the "efficiency frontier," automating model selection through smart routing, providing developers with progressive spend visibility instead of hard caps, and trimming token overhead via context pruning and cache tuning. The approach was detailed in a post on August 7 2026 by co‑founder and VP of engineering Patrick Wendell together with Akshat Bhatia, Vinay Gaba, Erich Elsen and Ivan Zhou. Databricks, a San Francisco‑based data and AI platform, runs thousands of coding agents—including Claude Code, Codex and Cursor—for its engineers, and AI token costs have become one of the fastest‑growing line items in R&D, ranking among the top three expenses for its customers after salaries and IT.

The new methods aim to curb the recent surge in AI coding costs that many companies face, offering a model‑efficiency focus, smarter request routing, clearer budget insights for developers, and reduced token consumption to keep AI usage affordable while maintaining broad access.