The Agent Governance Stack Is Finally Here

Five enterprise AI governance products shipped in twelve days. SAP, Collibra, Dataiku, Island, and Broadcom, all inside a thirteen-day window in late September and early October. I've been in enough client conversations about governance to know what that signals: the market has been waiting for tooling, not philosophy. The tooling just showed up.

We've spent two years talking about why agent governance matters. We haven't had the software to act on it unless you build it yourself. Most of those conversations stalled because that's also no small feat, especially given the limited talent pool that can do so. That just changed in the first part (though the second part remains a challenge).

The pattern across every product that launched is the same: inventory first, then control. You have to know what agents you have running before you can govern them. I've written about this specific problem before, and most enterprises are still stuck at step zero. Dataiku calls it "agent sprawl": employees creating agents faster than central teams can observe or manage them. Their Agent Management platform connects to AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio, Salesforce Agentforce, and Snowflake Cortex. That list isn't accidental. It's a map of where enterprise agents are actually proliferating.

Broadcom's AgentMinder takes an identity-layer approach: it verifies each agent's identity and intended mission before authorizing action. You can't undo what you can't see. That's the principle at the infrastructure level now.

For most organizations, the question isn't whether to build a governance practice. It's which layer to own first: inventory, policy, monitoring, or identity. Each is separating into its own product category. That means integration work, and someone has to own the architecture. That someone now has a leg up.