Salesforce Unveils AIforce at Dreamforce 2026, Touting an “Interface Revolution”
Salesforce took the wraps off AIforce at Dreamforce 2026 on September 15-16, framing the platform as a live interface layer that exposes the full Salesforce stack — data, workflows, business logic, semantics, permissions, security and governance — to any AI surface. The formal press push landed September 16, putting a name on a strategy Salesforce has been hinting at for months as agentic AI moved from pilot projects to production rollouts. In a keynote quote that anchored the announcement, CEO Marc Benioff said “AI is creating an interface revolution,” arguing that enterprises no longer have to choose between their CRM and the large language models their employees increasingly reach for first.
Three Launch Products Anchor the AIforce Rollout
The debut is not a single product but a bundle of three entry points designed to meet companies where their users already work. Claudeforce, the first, ships Salesforce inside Anthropic’s Claude via a Model Context Protocol server and ships with 37 prebuilt sales skills; a beta is available immediately. Slackforce extends the CRM into Slack with an upgraded Slackbot, deeper Slack CRM integration, and a new Slack Code surface for developer collaboration. Agentforce Coworker, billed as a Lightning AI teammate, has already crossed 100,000 users in the 35 days since its soft launch, a figure Salesforce used to underscore the appetite for embedded agents inside everyday workflows.
Headless Toolkit and AgentExchange Open the Architecture
Underpinning the products is the Headless Toolkit, Salesforce’s open architecture for connecting agents to enterprise systems through MCPs, APIs, plug-ins and reusable skills. The company is positioning AgentExchange as the surrounding ecosystem, with launch partners including Anthropic, AWS, Google, Microsoft and OpenAI — a roster designed to blunt the argument that Salesforce is locking customers into a single model provider. By exposing business logic and permissions as first-class primitives, Salesforce is betting that the next generation of enterprise AI will be defined less by model choice and more by which platform can safely mediate what an agent is allowed to see and do.
Zero Data Retention and a Security-First Pitch
Data governance is central to the AIforce pitch. Salesforce is leaning on a Zero Data Retention policy under which agents only see what the authenticated user can see, with no model-side caching of customer records. The framing is a direct response to enterprise security teams that have stalled AI rollouts over fears of training on confidential data or leaking Personally Identifiable Information through third-party tools. Combined with native respect for Salesforce’s existing role hierarchy and sharing rules, the company argues that AIforce agents inherit the same guardrails a human sales rep would, without requiring parallel governance programs.
Early Pilots: Deloitte, GitLab and Legora Test Salesforce in Claude
Salesforce in Claude, the consumer-facing expression of Claudeforce, has been piloted with Deloitte, GitLab and Legora, giving Salesforce three reference customers across consulting, developer tooling and legal tech. Each firm brings a different pressure test: Deloitte’s global sales org, GitLab’s distributed commercial team, and Legora’s document workflows. While Salesforce did not disclose outcome metrics, the diversity of the pilot cohort is meant to signal that the integration works across industries where regulated data, multi-stakeholder deal cycles and complex approval chains are the norm rather than the exception.
Koa Reasoning Model Debuts on NVIDIA Nemotron 3 Super
Separately on September 15, Salesforce unveiled Koa, its first CRM reasoning model built on NVIDIA Nemotron 3 Super and tuned for Agentforce. The model is positioned as a specialized alternative for customers who want to keep inference inside the Salesforce stack rather than route every call to a frontier general-purpose LLM. Paired with AIforce’s headless architecture, Koa gives Salesforce a way to compete on cost, data sovereignty and latency while still letting customers swap in external models when the workload demands it. The combined message from Dreamforce is that Salesforce intends to be both the system of record and the system of reasoning for enterprise AI.
Trailblazers Mark 20 Years as Salesforce Charts the Next Decade
The Dreamforce floor also doubled as a homecoming for the Trailblazer community, which celebrates its 20th anniversary this year. Two decades after the first Dreamforce, the audience has shifted from administrators learning click-not-code customization to enterprise architects wiring agents into mission-critical processes. Salesforce used the milestone to position AIforce as the next chapter in that lineage — the same platform, now rearchitected for an era where the user interface may be a chat thread, a code editor, or a model that never renders a screen at all. AIforce is generally available now, with Salesforce in Claude in beta.
AIforce lands at a moment when enterprise AI buyers are explicitly shopping for platforms that can carry identity, permissions and audit trails into the agent era, and the inclusion of Anthropic, AWS, Google, Microsoft and OpenAI in AgentExchange suggests Salesforce is willing to share the model layer in exchange for owning the connective tissue. Whether that trade pays off will depend on how quickly Claudeforce, Slackforce and Agentforce Coworker move from novelty to default inside the world’s sales and service organizations. AIforce is Salesforce’s bet that the answer is already underway.
Source: Salesforce press release, Sep 16 2026 (https://www.salesforce.com/news/stories/aiforce-announcement/)