Databricks engineers use coding agents daily, mixing between Claude Code, Codex, Cursor, and others, often several at once, making coding agent spend one of the fastest growing line items in R&D. A single runaway automation loop can burn through a month of budget in an afternoon. Databricks solved this internally using Unity AI Gateway Budgets, routing all coding agent traffic through the gateway to enforce one spend policy across the entire fleet. Key lessons included separating runaway spend protection from monthly spend limits, noting most limit hits are normal engineers doing normal work requiring self-service unblocking, and reserving approvals for truly exceptional cases with time-limited entitlements. Initially, every engineer got a default monthly spend limit (e.g., $500), with increases in the same $500 steps and anything past $2500 needing manual review; every increase was permanent. Between 500 and 1,000 engineers were hitting the limit every month, generating hundreds of tickets and interrupted work sessions.