Twelve months ago Salesforce sold you a CRM with AI features attached. Today it sells an agent platform that happens to include a CRM. That is not marketing spin… it is an architectural decision that changes where your users work, what your admins configure, and eventually what you pay for.

Here is what actually moved between October 2025 and October 2026, and what it means if you run a Salesforce org.

October 2025: The Agentic Enterprise becomes the story

Dreamforce 2025 introduced Agentforce 360, a packaging of four things that had previously been separate products: the Agentforce agent layer, Data 360 (the renamed Data Cloud), the Customer 360 applications, and Slack. It added a conversational builder, hybrid reasoning, voice support, and better agent observability. If that is your entry point, we covered the practical first steps in how to get started with Salesforce Agentforce in minutes.

The framing mattered more than the feature list. Salesforce stopped describing Agentforce as a feature inside Sales Cloud and started describing the whole company as an “Agentic Enterprise” platform. The pitch: humans and agents working together across data, logic, and surfaces, rather than users clicking through traditional Salesforce screens.

That last clause is the thread that runs through the entire year.

The naming experiment, and the partial retreat

Between Dreamforce 2025 and the following spring, Salesforce applied the Agentforce prefix aggressively. Sales Cloud became Agentforce Sales. Service Cloud became Agentforce Service. Marketing Cloud, Commerce, and even the core platform picked up the same treatment. Data Cloud became Data 360.

By September 2026, ahead of Dreamforce 2026, most of that reversed. The clouds reverted to their classic names: Sales Cloud, Service Cloud, Marketing Cloud. Agentforce stayed as the AI and agent brand. Data 360 kept its new name. Headless 360 concepts migrated under the Salesforce Platform and Headless Toolkit naming.

What to take from this if you are an admin: product names are not stable right now. If your documentation, training material, or internal runbooks reference “Agentforce Sales,” they are already partly stale. Reference capabilities by what they do, not by the marketing label attached that quarter.

Dreamforce 2026: the interface inverts

The headline announcement was a new interface layer. Initially announced as AIforce, Benioff renamed it SIforce on X in early October 2026. The concept behind either name is the same: surface Salesforce data, workflows, business logic, permissions, and governance into any surface a user already works in.

It is powered by the Headless Toolkit and an experience layer built on MCP servers, APIs, and prebuilt skills. The stated goal is “Salesforce on any surface” instead of requiring everyone to live in Lightning.

Concretely, that includes Claudeforce (beta), Slackforce, and native experiences outside the Salesforce UI. This is the real shift of the year. For two decades Salesforce’s model was: we host the system of record, and you come to it. The headless model says: the system of record stays, but the interface is wherever the work happens.

Claudeforce and the Anthropic partnership

Announced around August 2026, Claudeforce is the Anthropic partnership. Salesforce data, workflows, and governance become available inside Claude through a secure MCP connection and prebuilt skills. The initial plugin shipped with 37 sales skills, and Claude models also power reasoning inside Agentforce and Slack.

Two details are worth noting for architects:

1. Zero data retention and governance inheritance. The claim is that Salesforce permissions carry into the Claude surface rather than being reimplemented there. If that holds, it solves the hardest part of any external-surface rollout.
2. It requires a direct Anthropic subscription for the Claudeforce path. Access through Bedrock or Vertex is not sufficient. That is a procurement consideration, not just a technical one.

Agentforce Coworker and the move to job-ready agents

Agentforce Coworker is an autonomous teammate built into Lightning. It answers natural-language questions, researches across the CRM and Slack, plans, and orchestrates within the user’s existing permissions. Early adoption was reported at roughly 100,000 users in the first 35 days, with GA around or shortly after Dreamforce 2026.

More strategically, Salesforce shifted from selling a DIY agent builder toward selling pre-packaged, role-specific agents. Named examples include Piper for inbound pipeline and SDR work, Hunter for outbound, Casey for service, Paige for operations, Carter for commerce and shopper interactions, and Marshall for supply chain and HR.

That is a meaningful change in what you are buying. A generic builder requires you to define the job, the skills, the context, and the guardrails. A role-specific agent arrives with those decisions largely made. For most mid-market orgs, the second option is the one that actually ships.

Koa: a reasoning model built for CRM

Salesforce announced Koa, its first purpose-built CRM reasoning model, developed with NVIDIA and based on Nemotron. It is trained for multi-step CRM workflows, tool use, and enterprise tasks rather than general conversation. Pilot at Dreamforce 2026, broader GA later.

The bet here is that general-purpose models are the wrong shape for CRM work. A model that understands opportunity stages, quoting logic, and case escalation paths natively could be more reliable at orchestration than a much larger general model prompted to behave like one. That is unproven, but it is the clearest statement yet that Salesforce intends to own the reasoning layer rather than rent it.

CPQ reaches end of sale

Salesforce CPQ went end of sale in March 2025. This is end of sale, not end of life. Existing customers can renew, add licenses, and continue receiving support and critical fixes. New customers cannot buy it.

New investment goes to Revenue Cloud Advanced, briefly marketed under an Agentforce Revenue name before reverting. A migration assistant is targeted for around December 2026, and Winter ‘27 adds quoting capabilities on the Revenue Cloud side. Salesforce has acknowledged that some orgs can run both in coexistence.

If you are on Salesforce CPQ, this is your planning horizon. It is not an emergency, but “end of sale” is the beginning of a clock. The Winter ‘27 ramp-deal pricing work, including compound price uplifts for multiyear ramp deals, is landing on the Revenue Cloud side, which tells you where the roadmap points.

Marketing Cloud goes agentic, then buys its way forward

Winter ‘27 brought the most substantive marketing changes:

• Account Discovery Agent for a buying-committee view rather than a contact list
• Campaign Agent for agentic content generation and adaptive campaigns
• Personalized Paths in Flow for branching orchestration
• Stronger multi-channel and data-driven orchestration overall

Marketing Cloud Next is positioned as the modern flagship built on the platform plus Data 360, and the umbrella name reverted from the temporary Agentforce Marketing label. If you are still deciding between Pardot and Marketing Cloud, note that the decision is now also a decision about which generation of the platform you are betting on.

Then Salesforce bought the problem outright. In October 2026 it agreed to acquire Braze for approximately $5.66 billion, at $47.50 per share in cash, a 58.1% premium. Braze brings roughly 2,770 customers including Canva, Instacart, and Venmo, and real-time cross-channel engagement. The deal is expected to close in Salesforce’s fiscal 2028 and does not change its fiscal 2027 forecast.

Commerce: agentic shoppers and open protocols

Commerce Cloud gained Storefront Next at GA, the Carter Shopper Agent, agentic commerce search, support for open protocols such as the Universal Commerce Protocol, and merchandising and consent improvements. Salesforce also acquired Cimulate for AI-driven product discovery, and was named a Gartner Magic Quadrant Leader for Digital Commerce for the eleventh consecutive year.

The pattern across Commerce and Marketing is identical: ship the agent layer, then acquire the data and engagement depth the agent needs to be useful.

The acquisition spree, in one place

Target	Approx. value	What it adds
Informatica	~$8B	Data management and governance
Fin (formerly Intercom)	~$3.6B	Proven high-resolution service agents, reported ~76% autonomous resolution
Contentful	undisclosed	Content infrastructure
Cimulate	undisclosed	Commerce AI and product discovery
Braze	~$5.66B	Real-time multi-channel engagement

Fin is the notable one for service leaders. It sits alongside customizable Agentforce as a fast-deploy option built on an agent that already resolves a large share of contacts autonomously. Buying proven resolution rates is a shortcut past the long tuning project most orgs underestimate.

Platform maturity: the unglamorous work

Underneath the announcements, the platform work that determines whether agents actually survive production:

• Observability and Optimizer to analyze agent sessions and suggest fixes
• Long-horizon agents with goals spanning days or weeks rather than single turns
• Multi-agent orchestration, including Superagent and A2A patterns
• Expanded voice, with more languages plus Field Voice and employee voice
• Custom scorers for evaluating agent output against your own criteria
• A/B experimentation and conversation testing
• An upgrade path from Einstein Bots to Agentforce

That list is where most implementations succeed or stall. An agent that works in a demo and cannot be measured, tested, or handed a multi-day goal is not a deployed agent. For the service-side starting point, see begin your journey with an Agentforce service agent.

Data 360 and governance as the foundation

Data 360 remained the stated foundation: unified, governed data with Intelligent Context, Tableau Semantics, and tight coupling to both agents and headless access.

Governance got equal billing: Security Mesh, headless identity, opt-in controls on MCP servers, permission inheritance, zero-data-retention claims for partner model paths, and auditability. If you are pitching an agent program to a risk committee, this is the section that gets you approved.

Salesforce also expanded its OpenAI partnership from Dreamforce 2025, making GPT models available in the platform and surfacing Agentforce data and insights inside ChatGPT.

The results, with the caveats attached

Salesforce reported Agentforce ARR of roughly $1.5 billion, with growth in the region of 240% year over year for the relevant period, and AI plus Data ARR approaching $3.9 billion. It raised fiscal 2027 guidance after strong results, and the stock moved sharply positive on earnings, including one notably large single-day move.

The mixed note: premiums paid for acquisitions, Braze in particular, drew skeptical analyst commentary. Salesforce also reported mixed reaction to whether the agentic story is converting into durable expansion revenue rather than initial land deals.

Treat all forward-looking figures as reported claims rather than settled fact. Revenue recognition for agent products, seat-based versus consumption-based pricing, and how the acquisitions consolidate are all still moving.

What to do about it

Four concrete steps for the next two quarters:

1. Audit your naming and documentation. The responsibilities of a Salesforce administrator have shifted toward agent configuration and evaluation, so this is a role conversation as much as a documentation one. If your runbooks, training decks, or internal wiki reference Agentforce Sales or Agentforce Marketing, they are already inconsistent with current names. Fix labels before you onboard anyone to the new surfaces.
2. Decide your CPQ horizon. End of sale in March 2025 means the migration conversation is coming. Start the assessment now, while coexistence is still an option and you are not forced into a date.
3. Pick one headless surface and pilot it. The headless and MCP direction is the largest architectural change of the year, and it touches permissions. Test governance inheritance in a sandbox before anyone claims it works in your org.
4. Buy observability with the agent. An agent without session analysis, testing, and scorers is a demo. Budget for the measurement layer in the same conversation as the agent, not the quarter after.

The through-line of the past twelve months is not that Salesforce added AI. It is that Salesforce decided the interface is negotiable. Everything else, the acquisitions, the rebrands, the reasoning model, the headless toolkit, follows from that one bet.

Whether it rebounds depends less on the announcements and more on whether governance inheritance and orchestration hold up under real enterprise permissions. That is the part to test first.