Elise Gonzales

Elise Gonzales is a Staff Product Manager on Databricks’ Agent Bricks team, focused on building agents. Her product work and coauthored writing address concrete problems in enterprise AI: carrying a user’s permissions into tool calls, coordinating specialized agents, and making search follow instructions about which information to retrieve.

From classroom software to enterprise data

Elise studied computer science at Carnegie Mellon University. In 2011, she worked in Montevideo with TechBridgeWorld’s iSTEP program, leading development and testing of an English-literacy application for One Laptop Per Child computers. The collaborative project worked with Uruguay’s public education administration and a local secondary school. Students could complete assignments, receive grades, and review their work; a companion authoring tool let teachers supply the content. Elise’s responsibility was the student-facing OLPC application within that larger team effort.

Her subsequent product career included program management on Windows Calendar and HoloLens at Microsoft, work on Alexa Accessibility at Amazon, and product management at Curalate. She later led the product team at Amperity, a customer data platform, before joining Databricks. These roles span classroom software, consumer interfaces, accessibility, and enterprise data products.

Connecting tools, permissions, and specialist agents

Elise’s Databricks writing explains how shared infrastructure can help teams build agents around their existing data and tools. Her contributions are part of collaborative product and research efforts.

• Permissions that follow the user. In the managed MCP launch, coauthored with Sid Murching in June 2025, Elise connected the Model Context Protocol—a standard for exposing tools to models—to enterprise access controls. The support-agent example combines account information, knowledge-base articles, and deterministic billing functions. Managed servers act on the user’s behalf and respect Unity Catalog permissions, allowing an agent to use those resources under the existing permission system.

• Tools that other teams can find and reuse. Her November 2025 article with Nikhil Gaekwad and Harish Gaur extends the problem from connecting one tool to managing an organization’s collection of integrations. The MCP Catalog makes servers discoverable while governing access and recording activity. Teams can find existing integrations and build on them rather than duplicate work they did not know another team had completed.

• One entry point for specialized agents. Elise’s Supervisor Agent announcement addresses the burden of knowing which bot holds which information. The supervisor coordinates Genie Spaces for structured data, Knowledge Assistant agents for documents, and MCP tools. Teams can improve their specialists independently while users ask questions through a common interface. Subject-matter experts supply questions and guidelines to improve answers and routing, with interactions tracked through MLflow.

Instructions must reach the search itself

Elise also coauthored the Databricks research team’s Instructed Retriever article. It addresses a limitation of retrieval-based agents: an answer generator cannot reliably honor a request for recent, authoritative information if search has already selected unsuitable documents.

The architecture carries instructions, examples, and descriptions of available metadata into retrieval and answer generation. A request about product battery life can become separate searches for recent reviews and official specifications, with date and document-type filters applied before the answer is written. Ranking can account for those instructions as well as text similarity. Instruction-aware retrieval therefore changes which evidence reaches the model, instead of asking the final generation step to compensate for poor search results.

Databricks’ Knowledge Assistant puts this research into a managed product: source-aware queries can prioritize recent content, use metadata, and favor authoritative sources, while page-level citations let users inspect the material behind an answer. The product connects the retrieval problem to the broader Agent Bricks work Elise helps develop—giving enterprise agents access to information in a form they can use and people can examine.