REVENUE SYSTEMS ARCHITECTURE

The Claudeforce Roadmap: Headless 360, AIforce, and the Next Few Quarters

A dated timeline of Headless 360, the AIforce harness, and Claudeforce, with what each piece ships and what your Salesforce org has to be able to support.

Shannon Maguire•September 8, 2026

Claudeforce did not arrive on its own. It sits on top of two Salesforce platform moves that were announced earlier in 2026 and that are still shipping, and reading the three together tells you what your org has to be able to support over the next few quarters. The timeline below uses only what Salesforce and Anthropic have published, with dates.

April 15, 2026: Headless 360 is announced

Salesforce announced Headless 360 with a line that reads as a product spec and a warning at the same time: everything on Salesforce is now an API, an MCP tool, or a CLI command, and agents can use all of it. Parker Harris framed the direction as a question about why anyone should log into Salesforce again. The premise is that the browser session stops being the place work happens and becomes one client among several.

The consequence for your org is mechanical. Every validation rule, sharing rule, required field, and automation that used to be enforced by a page layout now has to be enforced by the data model and the permission model, because an agent never sees the page layout. Configuration that depended on a human reading a screen and knowing what to type stops holding.

Summer 2026: the platform surface grows

Through the Summer 26 release cycle Salesforce added the Headless Experience Layer to the platform roadmap, covering orchestration and runtime for agent driven interfaces. The public Headless 360 page now counts more than 60 MCP tools and more than 30 prebuilt skills, with governance inherited by every agent and interface that connects.

August 19, 2026: Headless 360 expands to the enterprise clouds

A week before the Claudeforce news, Salesforce announced that Headless 360 was expanding across the platform so that every cloud becomes a reusable enterprise capability an authorized agent can discover and call through open standards. Discovery is the operative word. Objects, fields, and automations that were only ever meant for one team become callable by anything holding the right permissions, so scope that was previously hidden by obscurity is now published.

This is the phase most companies underestimate. A duplicate custom field that two departments filled differently for four years was survivable while humans knew which one to trust. Once an agent discovers both and has to choose, the duplicate becomes a wrong answer delivered with confidence.

August 26, 2026: AIforce and Claudeforce

On the Q2 FY27 earnings call Marc Benioff named AIforce as the company's enterprise AI harness, describing it as something that has evolved over the past three years and that Dreamforce will lean on heavily. In his post announcing Claudeforce he put the pieces in order: Claude gets governed access to Data 360, Tableau, Slack, and Salesforce workflow through the AIforce harness and Headless 360. AIforce is the layer that decides what a model is allowed to touch. Headless 360 is what makes the touching possible. Claudeforce is the first branded product on top of both.

Salesforce reported $11.3 billion in Q2 FY27 revenue, up 11 percent year over year, on the same call, and its investment in Anthropic was valued near five billion dollars as of mid 2026. The commercial weight behind this roadmap is not speculative.

September 2026: open beta

Salesforce in Claude, the first Claudeforce product, ships as a Claude plugin with 37 prebuilt sales skills covering meeting preparation, deal health review, pipeline review, and account planning. It is with select pilot customers now, and open beta is planned for September 2026. An administrator connects the org once and the plugin runs under the sharing and permission model already in place. No standalone pricing has been published.

The rest of 2026 and into 2027

Salesforce has said more skills are planned for later this year, extending past sales into service, marketing, and commerce, and that Claude becomes the default model inside Slack. Dreamforce is where the AIforce story gets its full telling. Beyond those statements the sequencing is not public, so treat anything more specific as guessing.

What is public is enough to plan against. The permission model becomes the product boundary. Field definitions become the answer quality ceiling. Pipeline stage rules become the difference between a forecast an agent can compute and a forecast it fabricates from whatever the stage names imply.

What to do before the beta reaches you

Inventory the fields your reporting depends on and retire the duplicates that two teams populate differently.
Write down what each pipeline stage requires as entry criteria, then check whether the records in that stage actually meet it.
Audit profiles, permission sets, and sharing rules for access nobody intended to grant, because an agent inherits all of it.
Move validation out of page layouts and into rules that hold when a record is written through an API.

If you want a scored read on where your org sits across those four areas, the Claudeforce readiness assessment takes twelve questions and returns a per area result with the weakest area named first.