Enterprises are increasingly driving agent development for financial analysis co-pilots, customer service assistants, and internal knowledge retrieval. But this rapid growth brings a new challenge: how to find and manage them all. Teams end up playing agent roulette, toggling between dozens of specialized bots and trying to remember whether the “travel policy” resides in the HR agent or the finance agent. This cognitive load is slowing down productivity, leaving teams searching around aimlessly, creating agents that have already been created, or referencing outdated information. Enterprises need a single entry point that can reason about intent, coordinate specialized agents, and securely act on the user’s behalf.
The Agent Bricks Supervisor Agent, now generally available (GA), is a managed orchestration layer that lets you tie agents and devices together, completely governed by the Unity Catalog. It uses a dynamic observer pattern to analyze the user’s question and organize between genie space for structured data, knowledge assistant agents for unstructured data, and MCP servers for tools to answer complex questions and provide deep analysis. This gives teams the flexibility to independently own and iterate on the quality of their agents, and gives users a single place to get their work done.
Governance-by-design: Secure by Unity Catalog
For IT and security teams, agentic AI often works outside of enterprise security. Most tools require duplicating permissions or using blanket service accounts, creating a compliance gap where the agent can access data that the end user is unauthorized to view.
Agent Bricks uses the Unity Catalog as a control and governance layer for your models, data, and tools, as well as agents. Supervisor Agent natively supports Certification from (OBO)Acting as a transparent proxy for the human user. Each data fetch or tool execution is validated against the user’s existing permissions in the Unity Catalog: : Can they query a table, or do they have access to a specific tool through the MCP catalog. This ensures that agents stay in sync with your governance policies without any extra work.
For Franklin Templeton, scaling AI means making regulated fund documentation useful without compromising compliance. Using Agent Bricks, with governance built through the Unity Catalog, the team combined public fund documents with performance data to power a governed fund analysis agent based on approved enterprise sources.
“Agent Brix gives us the scale of reliable, compliant fund analysis. What used to take days now takes seconds, and we have confidence that each insight is based on our data and business logic.” -Colin Zimmerman, CFA, Lead Data Scientist, Franklin Templeton
Continuous improvement through research-supported education
A production-grade agent is never “finished”; It should evolve based on real-world performance. You need to evaluate its feedback, incorporate new information, and continually improve the agent to remain useful.
The supervisory agent has a built-in quality loop Agent Learning on Human Feedback (ALHF). Add questions and guidelines that the supervisor can incorporate to improve its answers, how it routes between sub-agents, and provide context to the system. It also makes it easier to collaborate with subject matter experts (SMEs): for example, your marketing team can provide guidelines about brand and style for agent responses, and the supervisor can learn directly from it. with built-in MLflow usage and integrationEvery interaction is tracked and measurable, allowing you to quickly see and address deficiencies.
Customers like Zapier have used agent learning based on human feedback to quickly iterate and improve their agents. zapier is using the Observer agent to democratize access to data, and has leveraged ALHF to improve Observer’s orchestration between different Genie locations and devices.
“The Agent Bricks Supervisor agent gives us a structured way to coordinate multiple data intelligence endpoints in a single system. Instead of hard-coding routing logic, we can guide how the agent prioritizes Genie and controlled data in the Unity Catalog through explicit instructions. This makes it much easier to create an internal ‘Ask Data’ experience that is flexible and reliable as it evolves.” – Alvaro Martin, Senior Data Engineer, Zapier
Use Supervisor Agent today
With general availability, the Supervisor Agent provides a managed foundation for orchestrating AI agents at enterprise scale. Teams can now route intent from a single control plane, control access through the Unity Catalog, and continuously improve agent quality.
Get started with Supervisor Agent today by creating your first agent and connecting it to your existing agents and tools. Explore the documentation to see how the Supervisor Agent fits into your production workflow.
