How Databricks manages its own coding agent spend with Unity AI Gateway Budgets

Summary

Databricks describes how it manages coding-agent costs as thousands of engineers use Claude Code, Codex, Cursor, and other tools, creating exposure to runaway automation and growing R&D spend. The company routes all agent traffic through Unity AI Gateway Budgets and separates short-term runaway-spend protection from long-term monthly spend governance. A daily budget triggers self-service acknowledgement through Slack, an internal portal, or the CLI, while a high monthly limit uses manager-approved, project-scoped tiers that expire. Both budgets apply simultaneously, so effective usage is capped by the lower of the month-to-date total plus one runaway increment and the monthly maximum. Centralizing metering also gives managers and finance shared usage data, and the company reports that approval queues disappeared, monthly requests became rare, and engineers stopped rationing usage.

Context

Databricks found that a single monthly spend limit created repeated approval requests, permanent limit increases, and no frictionless way to handle urgent work or accidental runaway automation. The company distinguishes short-term waste from long-term waste and notes that no single limit can address both without disrupting normal usage.

Approach / What changed

All coding-agent traffic is routed through Unity AI Gateway Budgets, where daily and monthly limits apply together. Daily tiers can rise through self-service acknowledgements and automation, while monthly tiers require manager or skip-level approval, are scoped to projects, and expire. Fixed tier groups, Slack notifications, portal and CLI workflows, and Unity Catalog usage data support enforcement and observability.

Takeaways

• Daily limits are deliberately small and target rapid accidental expenditure; users can acknowledge intentional spend through Slack, the budget portal, or the CLI without an approval queue.
• Monthly increases use a few coarse tiers, require manager or skip-level approval, and typically last one, three, or six months before reverting.
• Gateway-wide attribution and metering enforce budgets across Claude, GPT, Gemini, open source models, and multiple coding-agent tools under one policy.

Related reading