The gap between platforms is where your governance problem lives.
Inside each of your primary AI platforms, governance is a solved problem. Microsoft governs Copilot. Salesforce governs Agentforce. Snowflake governs Cortex. Each vendor has built real controls for the agents running inside their own platform, and those controls work as advertised, for that platform, in isolation.
The problem is not inside any one of them. The problem is the space between them.
Your agents cross platforms constantly. A single workflow pulls data from Snowflake, reasons over it with Claude, generates a draft in Copilot, routes it through a Salesforce approval, and writes back to Epic. No single vendor sees the full path. No single vendor's governance applies to the full path. When something goes wrong (a PHI leak, a hallucinated commitment, a policy violation) the forensic reconstruction has to assemble evidence from five separate systems in five separate formats, each one able to explain only what happened inside its own walls.
That cross-platform gap is where the governance problem lives, and it is not a problem any single platform vendor can solve. Any vendor's own governance runs inside that vendor's infrastructure, sees only that vendor's agents, and cannot, by architectural construction, govern across the boundary to a competitor's platform. Closing the gap requires a layer that is structurally independent of every vendor it governs.
Peer organizations at the scale of the 90,000-employee hospital system or the money-center bank are reaching this conclusion independently and starting to build this layer in-house. They are correctly diagnosing the shape of the requirement. The question is whether to build it or buy it, and whether to reach that decision before a production incident makes it urgent.
