In five days, Salesforce closes a $3.6 billion deal for Fin, launches seven named Agentforce agents, and previews a governance layer built to manage its rivals' agents too

Salesforce closed its acquisition of Fin, the AI customer-service company formerly known as Intercom, on September 10, 2026 -- a deal first announced June 15 at roughly $3.6 billion, about three times Agentforce's own current annual recurring revenue by outside estimates. The next day, timed to land just ahead of its Dreamforce conference (September 15-17), Salesforce named 7 role-specific AI agents built into Agentforce, previewed a cross-vendor governance layer, and folded its newly-acquired company into the lineup as one of the seven -- under its own name.

Six of the seven agents are generally available today: Casey handles customer service across voice, SMS, WhatsApp and web chat; Paige routes internal IT and HR requests; Carter runs e-commerce discovery and checkout; Marshall automates supply-chain back-office work; and Piper works inbound sales leads. Hunter, built for outbound sales pipeline development, is still in pilot, with general availability targeted for November. The seventh, Fin, handles customer-experience workflows -- and is also the name of the company Salesforce had just finished acquiring, whose own model suite and 30,000-company customer base now sit inside Agentforce's seventh slot rather than as a separate product line.

Fin's own reported average resolution rate -- 76% across chat, email, WhatsApp, SMS, voice and Slack -- is the strongest single number in Salesforce's press materials, and it belongs to the company Salesforce just finished buying rather than to anything Salesforce built in-house. Fin will keep serving its existing 30,000-company customer base rather than disappearing entirely into Agentforce's release cycle, according to both companies' own announcements.

“The big strategic benefit of our decision to join Salesforce is magnitudes more businesses and their customers that we can bring it to.” — Eoghan McCabe, co-founder and CEO, Fin (formerly Intercom)

The price is the number worth sitting with. $3.6 billion for a company whose own revenue Salesforce hasn't disclosed works out, by outside estimates, to roughly three times Agentforce's own current annual recurring revenue -- a rich multiple for a product line that, by Salesforce's own count, only about 12% of its customer base has adopted so far. That isn't necessarily an overpay; a 76% resolution rate and 30,000 existing customers are real assets, and the deal buys Salesforce a second, faster-to-deploy agent product it didn't have to build in-house. But it does mean the company is paying a growth-stage price for a business it is simultaneously routing straight into its own product line rather than growing independently.

The same week added a third piece: a preview of what Salesforce calls the Trusted Enterprise AI Harness, a six-part governance framework -- Trusted Context, Agency, Action, Governance, Security and Models -- plus an “AI Control Plane” meant to manage agents at the account level across an entire organization. What makes it notable is what it explicitly does not restrict itself to: Salesforce says the same governance tools reach agents running on Microsoft's Copilot, Google's or OpenAI's platforms, Slack and Teams, not just Agentforce. Broad availability is targeted for early 2027; what's live now is a preview.

Salesforce is not the first vendor to pitch this. Microsoft's Agent 365 has been generally available since May 2026 as a control plane for governing agents across and beyond Microsoft's own ecosystem; Amazon Web Services' Bedrock AgentCore Harness followed in June, built around letting customers swap the underlying model without rebuilding agent logic. The market reason all three exist at once is concrete: an independent VentureBeat Intelligence survey of 107 enterprises this July found 85% already run two or more agent orchestration platforms, averaging 3.1 apiece, with more than half expecting a hybrid control plane by the end of the year. A single-vendor governance pitch is selling into a problem most of the market already has.

The harder question a press preview doesn't answer is whether a vendor that also sells its own agents can credibly govern its rivals'. VentureBeat's own reporting notes a more mundane technical wrinkle underneath the positioning fight, too: a harness tuned around one model's planning and execution behavior may need to change whenever that underlying model does -- a maintenance cost that applies to Salesforce's multi-vendor ambitions as much as to any competitor's.

None of the three vendors has published independent, audited evidence that its harness actually reduces incidents or governance failures across a multi-vendor deployment -- what exists so far, on all sides, is a preview, a control-plane diagram, and a survey establishing that the problem it's meant to solve is real. Fin's 76% resolution rate and Agentforce's seven named agents are shipped, measurable products; the harness meant to govern all of it, and its rivals' agents besides, is not one yet.