Salesforce's New AI Harness Bets the Differentiator Isn't the Model — It's Who Controls the Context Around It

Salesforce unveiled the Trusted Enterprise AI Harness, a composable architecture bundling six capabilities — Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models — under a single AI Control Plane, built on existing Salesforce infrastructure including Data 360, MuleSoft, Agent Fabric, and Salesforce Guardian. President and Chief Platform Officer Rohan Kumar’s stated bet: “what will differentiate an enterprise is the trusted, proprietary context it brings to that intelligence, starting with the customer.” General availability lands in early fiscal year 2028, with the announcement explicitly stressing openness to Claude, Slack, and Microsoft Teams rather than a Salesforce-only stack.

That’s a notably different pitch than a bigger or smarter agent. Salesforce is betting the moat sits in data and governance, not in whichever model happens to be plugged in underneath — a claim that lines up with what it’s already shipping. Salesforce’s job-ready agent portfolio, announced days earlier, put concrete numbers behind that harness the same week: Salesforce says it’s delivered 7 billion Agentic Work Units across Agentforce and Slack in two years, including 3.2 billion in a single quarter, with Hunter — a new sales-prospecting agent — already building 60% of a customer’s pipeline in pilot. That’s the “Trusted Agency” and “Trusted Action” pieces of the harness with adoption data already attached, not a roadmap slide.

The “any model” framing is also already load-bearing in a live deal. Salesforce and Anthropic’s Claudeforce partnership shows Salesforce committing roughly $300 million to Anthropic tokens this year, on top of an existing $5 billion stake, to ship 37 prebuilt sales skills inside Claude — sellers reasoning over live revenue context and taking governed action without opening the Salesforce app at all. The harness is the infrastructure that made a deal like that defensible in the first place: hand a third-party model access, but keep governance, security, and context ownership on the CRM side.

For a rep or RevOps leader picking a platform, the harness reframes the question worth asking a vendor: not which model powers your agent, but what happens to your customer data and audit trail when that model gets swapped for a better one next year.