Dreamforce 2026: The Agentic Enterprise

Salesforce users may never have to log in again.

At least, that was the premise Salesforce co-founder and CTO Parker Harris put forward at Dreamforce 2026. This year’s conference introduced a series of major announcements spanning AI models, purpose-built agents, new interfaces, and expanded partnerships. Taken together, they demonstrate Salesforce’s rapid migration from its traditional application interface toward an agentic enterprise that sits underneath the work, wherever that work happens.

Introducing AIforce

AIforce puts that vision into practice. The new interface layer builds on Headless 360 and exposes Salesforce capabilities to external AI tools, allowing users to access and act on Salesforce data without opening Salesforce itself. Users can work in platforms such as Claude, Slack, Amazon Quick, or Gemini Enterprise, then carry that work into Salesforce within their existing permissions and governance. Salesforce says business data is not retained by the originating model provider.

AIforce launches with Claudeforce and Slackforce. Claudeforce brings Salesforce data and 37 prebuilt sales skills into Claude, allowing a user to ask for an update across deals and take action in Sales Cloud without opening the application. Slackforce brings similar capabilities into Slack, while Slackforce Surfaces allows users to ask Slackbot in plain language to build live dashboards and reports from Salesforce data that remain connected to the source and update as the underlying data changes.

Claudeforce is currently in beta, while AIforce and Slackforce do not yet have announced general availability dates.

Koa gives agents CRM-native reasoning

A general-purpose model can reason through a sales or service task, but it does not inherently understand the data structures, business processes, and actions underneath a Salesforce deployment. Koa is designed specifically for that environment.

Koa is Salesforce’s first CRM reasoning model, developed with NVIDIA by post-training NVIDIA Nemotron 3 Super. Salesforce trained it on synthetic data modeled on nearly three decades of CRM deployments, with scenarios designed around the complex, multistep workflows Agentforce agents perform across the customer lifecycle.

On Salesforce’s CRM benchmark, the company says Koa matches or exceeds leading models on CRM actions while producing three times fewer errors. It runs within Salesforce’s trust boundary, giving Agentforce a reasoning model purpose-built for the environment in which it operates.

Koa is currently available to select pilot customers, with general availability expected in winter 2026 in U.S. regions.

Agents built for the job, not the prompt

Salesforce also expanded Agentforce with seven prebuilt agents designed around specific business functions:

• Casey for customer service

• Paige for IT and HR service

• Carter for commerce and shopping

• Hunter for outbound sales

• Marshall for supply chain and operations

• Piper for inbound pipeline generation

• Fin for customer experience

Hunter, currently in pilot, is the clearest example of Salesforce moving beyond the traditional prompt-and-response model. Its long-horizon runtime allows the agent to maintain a goal over weeks or months, supporting complex sales processes without requiring a seller to restart the interaction each time.

The other agents are generally available today.

Agentforce expands further

Salesforce also announced:

• Agentforce Coworker: An AI teammate that works within Salesforce, reasoning across accounts, activity, and history and taking action within the permissions and business rules of the user.

• AgentExchange: A marketplace for agents, actions, and other components that organizations can use to extend Agentforce without building everything from scratch.

• Cloud partnerships: Expanded relationships with AWS and Google Cloud that give customers more options for connecting Salesforce data and AI capabilities with the broader technology environment.

The pattern is clear. Salesforce is making its platform accessible through other applications and AI systems while keeping its data, business logic, permissions, and governance underneath.

What does this mean for Salesforce users?

Salesforce has historically sold its platform around users. Organizations buy user licenses, and those licenses determine the access and entitlements attached to the platform. API access has generally been part of those agreements, subject to the limits tied to a customer’s edition and licenses.

That model still holds for now. But Salesforce’s move toward a headless, agentic architecture raises a fundamental question about what a “user” means when the person interacting with Salesforce may never open Salesforce at all.

If a user works with Salesforce through Claude, Slack, or another AI interface, the value being consumed is no longer tied as neatly to time spent inside the application. The work may still depend on Salesforce data, workflows, permissions, and business logic, but the interface through which that value is delivered has changed.

That creates a natural pressure point for Salesforce’s commercial model. Will API calls become more distinct from user licenses? Will consumption play a larger role? Could customers eventually allocate spend across Salesforce capabilities based on usage rather than simply the number of people provisioned to use the platform?

The answer will influence how organizations think about their Salesforce footprint itself. As the platform becomes infrastructure for agents and other applications, the unit of value shifts from access to activity. Salesforce has spent decades monetizing the people who use its software. The next phase may depend increasingly on the work its platform enables, whether a human ever logs in or not.