Enterprise data platforms are shifting toward lakehouses because they marry a data lake’s low-cost, varied data storage with a warehouse’s structure and governance. Vendors such as Snowflake, Databricks and Microsoft Fabric now offer built-in vector indexing to handle embeddings for RAG pipelines, while some are adding MCP connectivity so AI agents can query data directly. Gartner reports 65% adoption of lakehouses among its clients, and analysts say vector database support and agent interoperability are driving the push. At DocuSign, data from Salesforce feeds into Snowflake to train internal sales agents and customer-facing ML models. Every dataset goes through strict security reviews—first on ingestion, then on egress. Sensitive customer records stay locked down; only low-risk info like product specs or web content is exposed to agents. “We’re proceeding very cautiously,” says Shivi Verma, who oversees the process. Lemongrass took its AWS S3 lake, layered in custom governance four years ago, and now plans a Q3 proof of concept on an MCP server for incident and change-management data from ServiceNow. They’ll swap their homegrown add-ons for a standard lakehouse solution. AWS’s native integration with Anthropic’s Claude AI—and zero egress fees when models and data live in the same cloud—makes that move attractive. Security and governance remain central as enterprises open lakehouses to agentic AI. Early RAG setups managed permissions per use case, but autonomous agents introduce new risk. Companies need audit trails, role-based access and semantic context enforcement before letting AI pull data on its own. Without those controls, costs and compliance liabilities can spiral out of control.