People are (once again) talking about AI in Salesforce, and in this video I zoom in on the bit that’s getting missed: what happens when you move from a couple of AI agents in a pilot to dozens running across teams, clouds, and business processes. Capability is one thing, but scale changes everything. I’ve previously debated whether “93% accuracy” is good enough in isolation—but when you multiply a 7% error rate across high-volume workflows and multiple departments, it stops feeling like a feature and starts behaving like enterprise infrastructure. I walk through what shifts at scale: duplication, drift, and “agent sprawl” when different teams deploy agents independently with slightly different prompts, data, and assumptions. I also share a useful reference point from Snowflake’s rollout of an internal GTM AI assistant to ~6,000 users—what stood out was the deliberate approach: MVP first, limited personas, phased rollout, and a focus on protecting trust. From there, I unpack the governance reality: agents add intelligence, but they don’t remove the need for ownership, data hygiene, and accountability. MuleSoft Agent Fabric can standardize data access, context, orchestration, and policy controls—but it doesn’t decide who owns each agent, who monitors performance, or who’s accountable when things go wrong.