Databricks is sharing how its engineering teams use Unity AI Gateway to manage coding-agent spend across tools and models. By routing agent traffic through one control point, teams can combine centralized budgets, unified usage visibility, and flexible policies without managing each coding agent separately.
Whatโs new
- One gateway for every coding agent: Route tools such as Claude Code, Codex, Cursor, and others through Unity AI Gateway to apply consistent spend and access policies across models and tools.
- Separate daily and monthly controls: Use a smaller daily limit to catch sudden or accidental spend, and a higher monthly limit for planned project work. These limits address different kinds of waste and should not be treated as one budget.
- Self-service for normal usage: When engineers reach a daily threshold, they can acknowledge that the spend is intentional and continue without waiting for a ticket or approval.
- Time-limited exceptions: Higher monthly limits can be approved for specific projects and set to expire, so temporary needs do not become permanent entitlements.
- Unified observability: Because usage is routed through Unity AI Gateway, teams can monitor spend and activity across coding tools in one place and use the data to refine policies and model choices.
The post also explains why a single monthly limit created unnecessary friction: it had to catch both short-term runaway spend and legitimate, sustained project usage. Splitting those jobs into separate daily and monthly controls gives engineers more room to use AI while keeping unusual spend visible and bounded.
๐ Read the full post here
