Salesforce has unveiled AIforce at Dreamforce 2026, a new interface layer designed to bring Salesforce’s data, processes, permissions, business logic, and agents into different work environments without requiring users to open the CRM. The launch debuts with Claudeforce, Slackforce, and Agentforce Coworker, turning access to enterprise information into an experience that can be built directly from a conversation.
AIforce in 20 Seconds
- AIforce brings Salesforce’s data, workflows, permissions, and logic to other interfaces.
- The launch includes integrations with Claude and Slack, along with Agentforce Coworker.
- Salesforce says the agents preserve the CRM’s existing rules and permissions.
- Salesforce in Claude ships with 37 prebuilt sales skills.
- The Headless Toolkit architecture allows MCP, APIs, plugins, and skills to be used to build new experiences.
The core idea behind AIforce isn’t to add another chatbot to the CRM. Salesforce is trying to decouple the enterprise platform from its traditional interface so that data and actions can show up wherever employees and agents are already working.
The company presented the initiative on September 15 at Dreamforce 2026, held in San Francisco. According to Salesforce, AIforce can expose the knowledge that already resides in its platform — including data, workflows, semantics, permissions, security, and governance — to different AI interfaces.
The approach fits a shift affecting much of enterprise software: instead of navigating menus to find a specific function, users start describing what they need. The system interprets the request, checks the available data, and, when authorized, carries out actions.
Salesforce Wants the CRM to Stop Being a Window
AIforce is built on the architecture Salesforce calls the Headless Toolkit. Rather than requiring every interaction to go through Salesforce’s conventional interface, this architecture exposes platform components through application programming interfaces (APIs), the Model Context Protocol (MCP), plugins, skills, and developer tools.
This makes it possible to build interfaces tailored to specific jobs.
An employee could ask for information about an account from within a Slack conversation, request a summary of sales activity, and then ask for a record to be updated. Under Salesforce’s approach, the agent can check the available context and carry out the action using the same business rules and permissions that exist in the CRM.
The difference compared with a traditional integration is that Salesforce is trying to make the interface adapt to the request, rather than forcing the user to adapt to a fixed set of screens.
The company describes these experiences as dynamic, composable interfaces. Users can request a particular view or piece of information, and the system can build a specific experience using the available data and logic.
This doesn’t mean any external model gets automatic access to all of Salesforce. One of the central points of the proposal is precisely to preserve existing authorizations.
Salesforce says every request respects current permissions and business rules, so the agent only accesses information the user is authorized to see, and actions are carried out back within Salesforce’s controlled environment. The company also says the enterprise data used to generate responses isn’t retained by the model provider under its Zero Data Retention approach.
These are features described by Salesforce, and they will need to be evaluated based on each specific configuration, contract, and service.
Claudeforce Connects Claude with CRM Data
One of the first examples is Claudeforce, the integration Salesforce and Anthropic announced ahead of Dreamforce.
Salesforce in Claude embeds a prebuilt MCP server directly into Claude and, according to the company, avoids much of the manual configuration, authentication, and capability-assignment work these kinds of integrations usually require. The first version includes 37 sales skills for tasks ranging from prospecting to keeping pipeline information up to date.
The company is also preparing additional capabilities for analytics with Tableau, as well as for service, marketing, commerce, and industry-specific use cases.
For developers, there’s also a Salesforce plugin for Claude Code with more than 40 skills and access to a broader library of Salesforce capabilities.
The integration isn’t entirely new from an architectural standpoint. Salesforce and Anthropic had already announced in August that Claude would be used within Agentforce and would be available through Amazon Bedrock inside the Salesforce Trust Boundary for certain scenarios.
Salesforce in Claude was initially being tested with select customers, and the company announced it would reach open beta during September 2026.
The technical implication is interesting: Anthropic’s model can provide reasoning and task execution while Salesforce retains the business context, data, and business rules.
Slackforce Turns Conversations into a CRM Interface
The second component is Slackforce, which brings Salesforce directly into Slack.
Here the ambition goes beyond answering questions. Salesforce envisions Slack becoming an interface from which CRM information can be viewed and updated without opening a separate application.
With Slackforce Surfaces, users can create interactive interfaces using data from Salesforce, Slack, and other tools. These surfaces can be viewed, filtered, commented on, and used by a team.
The example Salesforce gives shows Slackbot analyzing conversations and Salesforce data to spot accounts with declining activity, check related support cases and conversations, reassign an owner, create a task, and draft a win-back email — all from within Slack.
Slack CRM is also arriving, letting users create accounts, log call notes, or update records through instructions issued from Slack.
The approach has an important consequence for the architecture of enterprise applications: the interface is no longer necessarily the main product. Data, permissions, processes, and actions can become services that different agents and interfaces consume.
Salesforce had already unveiled its Headless 360 strategy in August, built around this same separation between the platform’s capabilities and its traditional interface.
Agentforce Coworker Brings the Agent into Salesforce Itself
The third component, Agentforce Coworker, works in the opposite direction. While Claudeforce and Slackforce bring Salesforce to other environments, Coworker embeds the agent directly into the Lightning interface.
The agent can reason over accounts, activity, and history, and carry out actions within existing permissions. It can also call on other specialized agents previously built with Agentforce.
Salesforce says 100,000 users activated Coworker within its first 35 days. The company has also presented enterprise cases in which customers have put dozens of use cases into production. These are figures and examples provided by Salesforce, not an independent measure of adoption.
The company is also expanding Agentforce with specialized agents capable of working over extended periods, retaining context, and coordinating with one another. In parallel, it has announced multi-agent orchestration capabilities and tools for testing and improving agents.
MCP, APIs, and Skills as Building Blocks of an Open Platform
AIforce’s architecture rests on an idea gaining traction in enterprise software development: AI models don’t need a monolithic application, but rather structured access to tools and data.
Salesforce’s Headless Toolkit provides exactly those building blocks. MCP makes it possible to expose tools and context to compatible models, while APIs and plugins enable more traditional integrations. Skills encapsulate specific capabilities that an agent can use.
On top of that foundation, Salesforce is developing AgentExchange, an ecosystem where third parties can build interfaces, agents, applications, integrations, and workflows.
The company lists Anthropic, Amazon Web Services, and Google among its partners, along with developers such as Lovable and Vercel and agent and tool providers such as DocuSign, Gamma, Jasper, and Rippling.
This approach also explains why Salesforce insists AIforce isn’t just a new interface. Its intent is to turn the elements that make up the CRM — data, semantics, permissions, processes, and actions — into components other agents can consume.
For a company, this can reduce the need to build a different integration every time a new AI assistant appears. But it also shifts part of the complexity toward governance: the more agents and interfaces that have access to business processes, the more important it becomes to control what each one can view and which actions it’s authorized to carry out.
Salesforce presents AIforce as a layer that lets people and agents work on the same information without being tied to a single application. The launch also comes alongside a broader Agentforce, Claudeforce, and Slackforce strategy, so it remains to be seen how these integrations will perform in large-scale enterprise deployments.
The company also notes that availability may vary by region and contract, and that commercial terms may change. Some of the announced capabilities are still under development.
For more on how Salesforce and Anthropic’s collaboration began, see how Salesforce and Anthropic first connected Claude to CRM data, and for related integration efforts, see how AWS and Salesforce connect their AI agents, data, and models without moving data.
