Databricks says Unity Gateway cut AI coding agent costs up to 90%

Databricks says it cut unit costs for its AI coding agents by as much as 90% after building Unity Gateway, a control layer between developers and models.

The company says moving traffic to cheaper models produced the largest gain, at 50% or more, and that spend warnings, approvals and model downshifts also contributed. Databricks says token spend per head is roughly flat while total usage keeps rising.

Databricks also proposes shared infrastructure for managing growing numbers of AI agents, naming Agent Bricks for building and governing agent fleets, Omnigent for composing agents across harnesses, and Unity Gateway for centralized access, cost controls and observability. These are company claims, and the reports include no independent measurement.

Yuchen Jin, who says he joined Databricks, separately states that Databricks AI Gateway cut the company's internal coding costs by 35% through Smart Routing, and that the gateway offers open-source and proprietary models on Day 0 and lets teams track spending and set budgets. His figure differs from the 90% maximum in the Databricks Blog report, and the reports do not reconcile the two.