AIforce explained: what Salesforce announced and what it means for enterprise teams

Salesforce introduced AIforce at Dreamforce on September 15, 2026. It is an interface layer that makes Salesforce data, workflows, business logic, permissions, and governance available inside other AI tools, so work involving Salesforce records no longer has to begin in the Salesforce application.

The announcement packages three things customers can use now or will be able to shortly: Claudeforce, Slackforce and Agentforce Coworker. Beneath them sits the Headless Toolkit, an architecture that exposes Salesforce capabilities through open APIs, Model Context Protocol (MCP) servers, plugins and skills.

Marc Benioff, Salesforce chair and CEO, framed the release as an interface shift, describing it as “combining model intelligence with all the context that customers have built into Salesforce to create an intelligent, dynamic, composable system that is securely governed, built with Zero Data Retention, and designed to work with the core systems that already run your business.”

Patrick Stokes, Salesforce president of applications and marketing, was more direct in keynote coverage, saying the traditional interface often gets in users’ way and slows them down, and that AIforce “disaggregates the UI and brings AI in to kind of replace it.”

What AIforce actually is

AIforce sits between the Salesforce platform and whatever interface a person or agent is working in. Salesforce describes it as a layer rather than a new application, and it does not replace Customer 360 or Data 360.

Salesforce positions it alongside the rest of its agentic architecture:

• Data 360 provides harmonized and federated data, with metadata and memory.
• Customer 360 holds business logic, processes, permissions, and actions.
• Agentforce supplies the digital workforce of ready-to-deploy and custom agents.
• AIforce makes all of that reachable from outside the Salesforce user interface.

Four characteristics define how Salesforce describes the layer. Agents reason across data and logic in Salesforce at the same time rather than querying one object at a time. Requests run inside a user’s existing permissions, so a person sees only what their profile already allows. Interfaces can be composed by describing them in natural language. And business data is not retained by the model providers, which Salesforce refers to as Zero Data Retention.

The compatibility claim matters as much as the capability claim. Salesforce states that AIforce requires no new permissions model and no migration, and works with the security and sharing configuration an organization already operates.

The components

Claudeforce

Claudeforce connects Salesforce to Claude through a prebuilt MCP server. Salesforce and Anthropic first announced the partnership on August 26, 2026, and Dreamforce expanded it.

The Salesforce in Claude plugin ships with 37 prebuilt sales skills covering tasks such as meeting preparation, deal health review and pipeline review. Actions are routed back through Salesforce so business rules are enforced, and authentication is handled once at admin level rather than per user. A separate development plugin offers more than 40 skills along with access to a GitHub library. Skills for Tableau analytics, service, marketing and commerce are planned.

Claudeforce is in beta. Salesforce names Deloitte, GitLab and Legora among the pilot customers.

Under the broader Anthropic agreement, Claude serves as the reasoning model for Salesforce’s Atlas Reasoning Engine, powers Agentforce Vibes and Agentforce Coworker by default, and is available in Agent Builder. For regulated industries, Claude is accessible via Amazon Bedrock within the Salesforce Trust Boundary. Dario Amodei, Anthropic CEO and co-founder, said the aim is for Claude to bring “frontier intelligence into the systems where much of the world’s commercial activity happens.”

Slackforce

Slackforce brings the same capability into Slack, and has four named parts:

• Slackforce Surfaces creates interactive live interfaces inside Slack.
• Slackbot acts as a personal assistant reasoning across both Slack and Salesforce context.
• Slack CRM allows accounts and records to be managed through prompts without leaving the conversation.
• Slack Code supports team-based AI development that produces working code.

Elia Wallen, founder and CEO of Engine, described the practical effect for a company handling more than 800,000 customer inquiries a year: “instead of waiting on someone else’s time and expertise to turn data into a dashboard or report, anyone on our team can just describe what they need and get a real, working interface back, built from data we already have, without ever leaving the conversation.”

Agentforce Coworker

Agentforce Coworker is an AI teammate that operates inside the Lightning interface and, according to Salesforce, across Slack, Microsoft Teams, Claude, ChatGPT and mobile. It inherits existing Salesforce configuration, custom workflows and governance rules rather than requiring separate setup, and it can call specialized Agentforce agents.

Salesforce reports 100,000 users activated in the first 35 days. Brandon Bernard, senior vice president and sales technology strategist at Fulton Bank, said the bank went “from zero to more than 20 use cases live in production, supporting approximately 3,000 users across the organization” within weeks.

The Headless Toolkit and the partner ecosystem

The Headless Toolkit is the underlying architecture, exposing platform elements through open APIs, MCPs, plugins and skills. Salesforce lists AgentExchange partners including Anthropic, Amazon Web Services, Google, Microsoft, Lovable, Vercel, Docusign, Gamma, Jasper and Rippling. Reporting from the keynote also referenced native integrations with Amazon Quick and Google Gemini Enterprise.

What this changes for enterprise teams

Most of the operational consequences of this announcement fall outside the technology itself.

When Salesforce work occurs within the Salesforce application, the application itself enforces a high level of process discipline. Required fields, validation rules, page layouts, and record types all guide users to enter data correctly. Training programs, adoption metrics, and super-user networks are built around that interface. Most Salesforce adoption programs are designed on the assumption that the interface is where the work happens.

Once the same work can begin in Slack, in Claude, or in a Teams thread, several assumptions need re-examining:

• Data quality. Salesforce states that business rules are enforced because actions route back through the platform. Whether that holds for every path, and how validation errors surface to a user working in a chat window, are things each organization will need to test against its own configuration.
• Auditability. Compliance teams will want to know that audit trails remain complete when a record is updated from a conversation in another product.
• Access sprawl. Permissions do not change, but the number of doors into the data increases. Reviewing profile and permission set design before opening those doors is a reasonable precaution.
• Prompting as a job skill. Describing an interface in natural language is a different competency from clicking through one. It rewards people who understand the underlying data model and exposes those who do not.
• Role redesign. If an agent handles pipeline hygiene and meeting preparation, the value a sales operations analyst or a Salesforce administrator adds moves toward design, governance, and exception handling.

The launch raises questions of governance: spreading CRM access across several interfaces creates a genuine control problem, and the announcement focused on technical capability rather than organizational readiness.

What organizations should do next

There is time to prepare. Claudeforce is in beta, and its pricing, packaging, and regional availability are subject to change according to Salesforce. A measured sequence works better than a rush to enable everything.

1. Audit the permission model first. AIforce inherits what already exists. Configuration debt that was tolerable behind a familiar interface becomes more visible when agents reason across the entire dataset at once.
2. Decide which interfaces are sanctioned. Slack, Claude, and Teams do not all have to be enabled on the same timeline, and a phased approach gives change and compliance teams something to work with.
3. Pick a bounded pilot. A single team, a small set of skills, and a clear measure of success will teach an organization more than a broad rollout.
4. Rewrite the enablement; don’t retire it. Training focused on data literacy, prompt construction, verification of agent output, and knowing when to escalate is becoming increasingly valuable.
5. Set the verification standard early. Teams need an agreed position on which agent-initiated actions require human review and which do not, documented before usage scales rather than after an incident.

The pattern will be familiar to anyone who has run an ERP or CRM transformation. Technology arrives ahead of capability, and the gap between the two is where value is lost. K2U’s experience across SAP and Salesforce programs is that the organizations getting returns from new platforms are the ones treating capability building as part of the implementation rather than as a follow-up activity.

AIforce moves Salesforce into the tools where people already work. Whether it leads to faster decisions or faster mistakes depends on how well the surrounding workforce around it has been prepared.