Governing enterprise agents with Agent Bricks and Unity AI Gateway

AI agents are moving from demos to production. The hard questions are no longer about what agents can do. They are about cost, security and accountability.

Agent Bricks

Agent Bricks is Databricks' platform for building enterprise agents. It is now generally available, with managed memory, sandboxes and support for more models. Agents run close to governed data instead of pulling it into separate tools.

Unity AI Gateway

Unity AI Gateway adds controls while agents run. Teams can set spend limits, route requests to different models, apply security policies and govern MCP servers. In short, it gives one place to see and control what agents do.

A simple governance checklist

• Owner: every agent has a named business owner and a technical owner.
• Data: the agent only reads data its users are allowed to see, through Unity Catalog.
• Budget: each agent has a monthly spend limit and an alert before it is reached.
• Approval: actions that change records or reach customers need human approval at first.
• Evaluation: a test set of real questions is run before every release.

Start with an internal agent

We usually start with an agent for an internal team, such as finance or operations. The users are close, mistakes are cheap, and the rules you set become the template for customer-facing agents later.