Databricks uses Unity AI Gateway for internal AI coding agent spend control

The system employs a daily limit to catch runaway AI spend, which engineers can acknowledge via Slack to self-increase the limit without manual approval. A separate, higher monthly limit governs extraordinary spend and requires manager approval, with increases being time-limited to specific projects.

Engineers receive Slack notifications when approaching their daily AI spend limit and can acknowledge intentional usage with a single click to raise the limit. This self-service mechanism is also available via an internal budget portal and CLI, aiming to unblock engineers quickly.

Databricks shares its internal strategy for managing AI coding agent spend, using Unity AI Gateway Budgets to enforce budgets, visibility, and policies across all models and tools. The approach balances innovation with cost control through separate daily and monthly budgets and self-service budget increases.

The strategy is based on principles that allow engineers to spend unimpeded for AI leverage while intervening only for short-term accidental waste (caught by daily limits) and long-term excessive techniques (governed by monthly limits).

Enforcing a single spend policy across all coding agents via the gateway ensures scalability without needing to manage individual agent consoles. This centralized control point is crucial for the effectiveness of the unified governance model.