Salesforce AIforce Brings CRM Data to Every Agent Surface

On September 15, 2026, Salesforce stood up in front of 53,000 attendees at Dreamforce in San Francisco — with another 10 million watching on Salesforce+ — and told the room that the CRM was no longer a place employees go. Instead, Salesforce unveiled AIforce, described by the company as a live interface layer that carries the full power of the platform to wherever people and agents already work. The announcement marks the clearest break yet from the menu-driven enterprise software model: data, workflows, permissions, logic and agents that leave the Salesforce screen and show up inside Claude, Slack, Lightning and custom-built surfaces.

Salesforce AIforce arrives at a moment when the enterprise AI race is shifting from models to interfaces. Between September 8 and October 2 alone, five major platforms launched personal agents, including OpenAI Dots, Meta Muse and Google Gemini 4 Argon. Salesforce's answer is not another agent launch. It is a bet that the winner will be whoever can put trusted, governed business context inside every surface where work already happens — a vision CEO Marc Benioff framed as an interface revolution.

What Salesforce AIforce actually is

The core idea behind Salesforce AIforce is decoupling the enterprise platform from its traditional interface. Instead of employees navigating menus to find a function, they describe what they need in plain language; the system interprets the request, checks available data and permissions, and carries out authorized actions. According to the announcement, agents can reason and take action across all the data, workflows and logic inside Salesforce from whichever surface the user happens to be in.

Salesforce AIforce launches with five elements. Claudeforce puts Salesforce inside Claude through an MCP server with 37 prebuilt sales skills, plus a development plug-in for Claude Code; the skills have already been trialled by Deloitte, GitLab and Legora, according to reporting from cxm.world. Slackforce brings Slackbot, Slack CRM and Slack Code together with Slackforce Surfaces, which generate live dashboards and working views from a plain-language request — with live data available from October. Agentforce Coworker sits inside the existing Lightning interface as an AI teammate that acts on a person's behalf, and Salesforce reported that 100,000 people switched it on within the tool's first 35 days. The Headless Toolkit exposes the platform through MCP servers, APIs, plug-ins and skills, while AgentExchange brings partners including Anthropic, AWS, Google and Microsoft into the ecosystem.

Two partner moves widened the reach at Dreamforce. Salesforce capabilities are coming to Gemini Enterprise through a federated connector, now in private preview with general availability planned for late October, while Gemini models now run in the Agentforce Reasoning Engine on Google Cloud. Salesforce context also appears in Amazon Quick, and the company reported 7 billion AWUs across Agentforce and Slack — a usage figure that suggests agentic traffic inside the platform is already enormous.

Koa: a reasoning model trained on CRM, not the web

The most technically distinctive piece of the Dreamforce announcements is Koa, Salesforce's first purpose-built reasoning model for CRM work. Built on NVIDIA Nemotron, Koa embeds decades of enterprise business knowledge directly into the model, enabling it to reason through complex workflows while operating securely within a company's trust boundary, according to Salesforce's Dreamforce day-one recap.

The training approach is unusual. Rather than learning from web-scale text, Koa was trained on synthetic scenarios drawn from what the company describes as 27 years of customer deployments. The theory: CRM decisions follow repeatable patterns — approval chains, escalation rules, discount thresholds, case routing — and a model trained explicitly on those patterns should outperform a general-purpose model on CRM reasoning tasks. Salesforce is targeting enterprise general availability in the United States in winter 2026, close enough that organizations evaluating CRM AI roadmaps should be factoring it in now, according to analysis by Dotsquares.

Koa also signals a deeper shift in how enterprise AI gets built. The past two years of agent development have been dominated by general models wrapped in tools and prompts. Salesforce is arguing that vertical reasoning — a model whose world is approvals, pipelines and service-level agreements — is the next competitive edge. It is the same logic behind named, job-ready agents: Salesforce introduced ready-to-deploy Agentforce agents such as Casey, Hunter, Piper and Carter for specific functions, reducing the configuration overhead of the build-your-own-agent model that still ships alongside them.

The governance question: easy button, hard decisions

Analysts were quick to separate the demo from the deployment reality. Liz Miller, VP and Principal Analyst at Constellation Research, told CX Today that the Dreamforce story is "an easy button" — not that Salesforce makes agentic AI simple, but that it reduces the time and expertise needed to turn an AI idea into an enterprise capability. Her practical read: organizations should be able to stand up agents and skills in days rather than years, provided they start from the work itself rather than the interface.

Salesforce answered the governance concern directly in the launch framing. The company said AIforce agents preserve the CRM's existing rules and permissions, and Benioff described the system as built with Zero Data Retention — defined as business data that is not retained by the model provider. For enterprises that watched the summer of rogue agents — from self-modifying malware swarms to compromised open-source components — the permissions question is the whole ballgame. An interface revolution that cannot answer which data crossed which boundary will stall in every regulated industry.

Real deployments offered some evidence. Siemens confirmed it has connected Agentforce to its Teamcenter engineering software, letting field technicians and sales staff answer questions that previously required an engineer on the phone. The collaboration embeds the industrial digital twin into the commercial workflow — a virtual engineer in the hands of service technicians, as described by Siemens CEO Roland Busch in the announcement.

Still unanswered: pricing. No prices were published at Dreamforce, and the release states that pricing and packaging are subject to change — a notable gap for CIOs being asked to bet on the agentic enterprise this budget cycle.

Salesforce co-founder Parker Harris summed up the ambition: customers may never need to log in to Salesforce again. That line will either read as prophecy or hubris within a year. But the direction is unmistakable. The CRM wars are over; the interface wars have begun, and Salesforce just declared that its data will be on every battlefield. For the growing crowd of enterprise agent builders — the same audience flocking to marketplace agent platforms and open-source agent decision models — AIforce is the clearest signal yet that the platform game now belongs to whoever governs the context agents act on. Sources: Salesforce Dreamforce 2026 live blog, Dotsquares, cloudnews.tech, cxm.world.