Data-Native AI Agents: Why Agents Must Move to Your Data

Summary

Enterprise AI pilots often move data into separate vector databases, SaaS LLMs, or serving layers, creating governance gaps, compounded latency, fragmented costs and observability, and duplicated lifecycle work. The post advocates data-native agents: models, agents, tools, retrieval, and memory run inside the governed data platform, with policy enforced during query planning and computation rather than after responses are produced. It argues that post-hoc controls cannot undo sensitive information encoded in aggregations and can trigger token-burning retry loops. For state and memory, it presents Lakebase, managed PostgreSQL within Databricks, as transactional storage and a shared source of truth for multi-agent swarms. The described platform pattern combines Unity Catalog, Unity AI Gateway, Model Serving, MLflow 3, AI Search, Lakebase, and business-context services, and recommends inventorying workloads already outside the perimeter before closing seams incrementally.

Context

The post identifies problems caused by running agents in a separate AI stack and moving data outside governed systems: weakened access control and lineage, compounded network latency, fragmented costs and observability, vendor coordination, and loss of business context. It also explains that post-hoc governance cannot undo sensitive information encoded in computations and can cause agents to retry, increasing token use.

Approach / What changed

The proposed approach is to run models, agents, tools, retrieval, state, and memory inside the data platform. It describes enforcing policies during query planning and computation, using Unity AI Gateway for deterministic guardrails, Lakebase for governed transactional state and shared multi-agent memory, and platform services including Unity Catalog, Model Serving, MLflow 3, AI Search, Unity Catalog Semantics, and Genie Ontology.

Takeaways

• Governance must be applied before and during query execution because aggregations and other derived values can already encode rows a user was not permitted to influence.
• Lakebase provides transactional agent state and memory inside Databricks, giving multi-agent swarms a shared source of truth with atomic updates and traceable writes and reads.
• The post recommends inventorying agents and AI workloads already outside the governed perimeter, then closing the lakehouse-to-AI-stack seam incrementally rather than through one re-platforming project.

Related reading