TL;DR: on September 15, 2026 at Dreamforce, Salesforce announced AIforce, an interface layer that makes CRM data, workflows and business logic accessible from Claude, Slack or any AI interface, without ever opening the Salesforce application.
Three weeks after the Claudeforce partnership, Salesforce is following the logic through. Patrick Stokes, the company's president of applications, put it bluntly: the interface is not the product. When the vendor of the world's best-selling CRM says its screens are no longer the point, any business owner paying for licenses their team never opens has a good reason to read on.
Key takeaways
- Salesforce announced AIforce on September 15, 2026, on day one of Dreamforce (Moscone Center, San Francisco, September 15 to 17): a "live" interface layer sitting above Agentforce, Data 360 and Customer 360.
- AIforce launches with three surfaces: Claudeforce (Salesforce inside Claude, 37 prebuilt sales skills), Slackforce (the CRM inside Slack) and Agentforce Coworker, activated by 100,000 users in 35 days according to Salesforce.
- Every request runs on the user's existing permissions; Salesforce promises zero data retention on the model provider's side.
- The strategic message: a CRM's value no longer lives in its screens, but in the data, business logic and permissions it structures.
What Salesforce announced on September 15
AIforce is presented in Salesforce's official announcement of September 15, 2026 as a "live interface layer": it carries the data, workflows, business logic, semantics and security rules stored in Salesforce to whatever interface people and AI agents work in. In practice, a sales rep can query their pipeline, update a record or trigger a workflow from Claude or Slack without opening a single Salesforce screen. They can also compose an interface on demand, simply by describing it in natural language.
The table below sums up the components announced at Dreamforce.
| Component | What it is | Announced availability |
|---|---|---|
| Claudeforce | Salesforce inside Claude through a prebuilt MCP (Model Context Protocol) server, 37 sales skills | Open beta; pilots at Deloitte, GitLab and Legora |
| Slackforce | The CRM inside Slack: live data surfaces, Slackbot assistant, account management | Announced September 15, 2026 |
| Agentforce Coworker | AI teammate embedded in the Lightning interface | 100,000 users activated in 35 days |
| Headless Toolkit | Open architecture: MCP servers, APIs, plug-ins and skills to connect other interfaces | Announced; over 40 integration partners |
According to IT Pro, the stated ambition is to let enterprises "break free of a shared user interface": every employee would get their own dynamic interface, generated for the need of the moment. Marc Benioff, Salesforce's CEO, calls it an "interface revolution" driven by AI, combining model intelligence with the context customers have built into Salesforce.
Why "the UI is not our product" is the key line of Dreamforce
The most important statement of the conference was not about a feature. It came from Patrick Stokes, Salesforce's president of applications, quoted by IT Pro:
"The UI is not actually our product, the Salesforce product is the platform where our customers store their data." Patrick Stokes, IT Pro, September 2026
For twenty-five years, Salesforce sold the opposite: screens, tabs, dashboards, and training courses to learn them. If the vendor is flipping its value proposition, it is because the screen layer has become bypassable. Since AI assistants learned to read and write into third-party systems through MCP (Model Context Protocol), an open standard published by Anthropic in November 2024, any well-structured piece of software can be driven from a conversation. Salesforce would rather organize that bypass than suffer it.
The repositioning shows where the value actually lives. Three layers survive the disappearance of screens: the data model (accounts, contacts, deals, history), the business logic (validation rules, workflows, calculations) and the permissions (who sees what, who edits what). Salesforce leans hard on that last point: every AIforce request runs on the user's existing rights, and the vendor promises zero data retention on the model provider's side. The screens, meanwhile, become a by-product, generated on demand.
What does it change for a small or mid-sized business?
Directly, little in the short term: AIforce targets large Salesforce accounts first, with pilots at Deloitte, GitLab and US-based Fulton Bank, which reports more than 20 use cases in production within weeks. Indirectly, the announcement concerns every company, including those with 10 to 100 employees, for three reasons.
First, it addresses an old small-business problem: the CRM nobody opens. If your sales team lives in email and chat, a CRM reachable from those tools has a far better chance of staying up to date than one that demands a login and six clicks. That principle does not depend on Salesforce: it holds for any tool properly exposed to an AI assistant.
Second, it changes the buying question. The right question is no longer "which software has the nicest interface?" but "which tool best structures my data and my business rules?". A company choosing a CRM in 2026 should look at the quality of the data model, the granularity of permissions and the existence of an API or MCP server before comparing screenshots.
Third, it validates an approach that does not require an enterprise budget. MCP is an open standard: a custom business tool, built around your data and your rules, can be connected to Claude or another assistant in exactly the same way. What Salesforce sells as a revolution is, at small-business scale, a few weeks of work, provided the data is clean and the rules are explicit.
What AIforce does not fix
The announcement deserves its caveats. An interface generated on demand fixes neither dirty data nor fuzzy processes: if your pipeline is not maintained, the same neglected pipeline will simply show up in Claude instead of Lightning. Zero data retention and permission guarantees are vendor commitments, to be verified contractually, especially for customer data covered by GDPR (General Data Protection Regulation); Salesforce's own announcement notes that availability and pricing may vary by region. And for a small team that fits in one room, a disciplined spreadsheet or a well-adopted lightweight CRM sometimes remains a better investment than an AI layer on top of oversized software.
How this connects to my own work
I worked as a Salesforce administrator for e-Enfance, the French nonprofit that operates 3018, the national hotline against cyberbullying: Apex development, LWC components, 3CX telephony integration. What Dreamforce confirms, I saw in the field: the platform's value was in the data model and the business rules, never in the screens, which we spent our time adapting to how people actually worked. It is also the logic behind my recent projects: structure the data and the logic first, then expose them where teams already work. I covered the Claudeforce side in an article on August 28; AIforce is its generalization, announced three weeks later.
Salesforce just said out loud what the market had been whispering: screens can be generated, data has to be built. The question to ask is no longer "which CRM should we pick?" but "are our data and business rules clean enough for an AI to work on them?". Whoever can answer yes is ahead, whatever software they run.
Frequently asked questions
What is AIforce, announced by Salesforce at Dreamforce 2026?
AIforce is an interface layer announced by Salesforce on September 15, 2026 at Dreamforce. It makes CRM data, workflows and business logic accessible from any AI interface, including Claude and Slack, without opening the Salesforce application. It launches with three surfaces: Claudeforce, Slackforce and Agentforce Coworker.
Do you still pay for Salesforce licenses if nobody opens the interface?
Yes. AIforce moves CRM access into other interfaces, but every request still runs through the Salesforce platform, with its permissions and its billing. The value being billed shifts from screens to the data and business logic hosted by the vendor.
Should a small business wait for AIforce to connect its CRM to an AI assistant?
No. MCP (Model Context Protocol), the standard AIforce builds on, is open. Both off-the-shelf CRMs and custom tools can already expose their data to an AI assistant through an MCP server, provided the data and permissions are properly structured.