The two biggest names in enterprise data spent the last six weeks making the same announcement in different words: the data platform is now the agent platform. At its Data + AI Summit on June 16, Databricks reported more than 100,000 agents built on Agent Bricks since launch, processing over one quadrillion tokens per year. Two weeks earlier, Snowflake announced its intent to acquire Natoma, an enterprise MCP platform, to govern what agents can access and do across business systems.
Strip the branding and one strategy remains: whoever holds the governed data intends to hold the agents that act on it. For data and AI leaders, that reframes the platform decision — and quietly restates why the data estate underneath is the real asset.
Two moves, one thesis
Databricks expanded Agent Bricks from a quality-focused agent builder into what it calls a comprehensive developer platform: support for LangGraph, CrewAI, Agno, the Claude Code SDK, and OpenAI Agent SDKs; managed agent memory via Lakebase; and MCP-based tool access — all running on governed lakehouse data with Unity Catalog supplying access controls, lineage, and auditing. The same summit introduced the Genie suite of agentic coworkers for business teams, per ChatForest's summit coverage.
Snowflake attacked the same problem from the governance side. Natoma is a centralized MCP gateway that enforces identity, policy, and audit at the tool-call level — who requested an action, what permissions they hold, whether the action is allowed. The Register notes it is Snowflake's sixth acquisition announcement since June 2025, and that it landed the same day as a five-year, $6 billion AWS infrastructure agreement. Post-close, Snowflake says its Cortex Agents will connect to over a hundred business systems out of the box, per Technology Magazine.
The telling admission
The most useful line in either announcement came from Databricks' own framing of Agent Bricks: the hard part was never the core agent loop — it was the surrounding 99%. Governing the data, the quality, and the context the agent consumes, as Atlan's platform analysis puts it. Both vendors are selling the same conclusion the market keeps rediscovering: agent capability is abundant; governed, trustworthy data is scarce.
That is also the honest caveat on the headline numbers. A hundred thousand agents built is not a hundred thousand agents in production earning trust. Token volume measures activity, not accuracy. Agents running on a fragmented estate process quadrillions of tokens of fiction just as fast.
What data and AI leaders should take from this
- The platform bet is a governance bet. Evaluate Agent Bricks, Cortex, and their peers on how they enforce identity, policy, and audit at the action level — not on demo quality.
- Locality matters in regulated industries. Agents running where the data lives inherit its access controls and lineage; agents bolted on from outside re-create the integration problem.
- Consolidation is accelerating. Six Snowflake acquisitions in a year and a $6B infrastructure commitment signal that the control-plane race will be capital-intensive — expect more M&A around agent identity, MCP gateways, and observability.
- Your leverage is the estate, not the vendor. A unified, processed, governed data foundation makes any of these platforms work — and keeps you portable between them.
The bottom line
Databricks and Snowflake just told the market where they think the value sits: not in the model, not even in the agent, but in the governed data layer agents depend on. They are right — and that layer is not something you buy at a summit. It is engineered, unified, and governed inside your own estate first. The platforms amplify whatever they find there.
Sources: Databricks — Agent Bricks at DAIS 2026 (June 16, 2026) · Snowflake — Natoma press release (May 27, 2026) · The Register (May 28, 2026) · Technology Magazine (June 2, 2026) · ChatForest DAIS 2026 builder guide (June 2026) · Atlan — Agent Bricks analysis (2026)