Salesforce spent the last 12 months tearing down the wall between its CRM UI and the rest of the world. The result is not a new product line. It is a different thesis about where work happens: not inside Lightning, not inside a console, but wherever a human or an AI agent needs to reach into Salesforce data, logic, and governance without opening a browser tab.

Here is what changed, what rebranded, what reversed, and what it means for anyone running Salesforce today.

Agentforce 360 and the Agentic Enterprise (Dreamforce 2025)

At Dreamforce 2025, Salesforce launched Agentforce 360 — a platform layer that combined the Agentforce agent runtime, Data 360 (the evolved Data Cloud), Customer 360 apps, and Slack into one stack (Salesforce). The pitch: the CRM is not a system of record that humans log into. It is an agentic platform where humans and AI agents cooperate across data, logic, and surfaces.

Agentforce 360 shipped with a conversational agent builder, hybrid reasoning, voice input, and improved context and observability tools. It went GA around October 2025 and set the framing for everything that followed: Salesforce is no longer selling a CRM. It is selling the Agentic Enterprise.

The rebrand that partially reversed

At and after Dreamforce 2025, Salesforce renamed its core products with an Agentforce prefix. Sales Cloud became Agentforce Sales. Service Cloud became Agentforce Service. Marketing Cloud became Agentforce Marketing. The platform itself became Agentforce 360 Platform. Data Cloud became Data 360 (Salesforce Ben).

By September 2026, ahead of Dreamforce 2026, most of those names reverted. Sales Cloud, Service Cloud, Marketing Cloud, and Commerce Cloud all returned to their classic names. Agentforce survived as the AI and agent brand. Data 360 kept its new name. The Headless 360 concept shifted toward the Salesforce Platform and what would become the Headless Toolkit.

The lesson: Salesforce’s core product names have too much equity to swap out, even for a flagship launch. The AI layer gets its own brand. The products keep theirs.

SIforce and the headless architecture (Dreamforce 2026)

Dreamforce 2026 pushed the interface story further. Salesforce announced a new surface layer — initially called AIforce, then renamed SIforce when Marc Benioff posted the change on X in early October 2026 (Marc Benioff on X) — that surfaces Salesforce data, workflows, business logic, permissions, and governance into any application, not just Lightning.

SIforce is powered by the Headless Toolkit and Experience Layer, which includes MCP servers, APIs, and prebuilt skills. The goal is explicit: Salesforce on any surface. If you want to reach into Salesforce from Slack, from Claude, from a custom internal tool, or from a partner’s app, the Headless Toolkit exposes the same data, logic, and permission model without forcing anyone into the Lightning console.

This is the architectural shift that underpins everything else. Headless is not a developer feature. It is the position that Salesforce’s value is its data and governance layer, not its UI.

Claudeforce: Salesforce inside Claude

In August 2026, Salesforce and Anthropic announced a strategic partnership: Salesforce data, workflows, and governance available inside Claude via a secure MCP connection (Anthropic). The initial plugin shipped with 37 prebuilt sales skills. Claude models also power reasoning inside Agentforce and Slack.

The partnership emphasized zero data retention and governance inheritance — meaning permission and security rules defined in Salesforce travel with the data when it surfaces in Claude. Claudeforce entered open beta around Dreamforce 2026 and requires a direct Anthropic subscription for the Claude-side integration path, not only Bedrock or Vertex.

This is the practical expression of SIforce. If the headless toolkit is the plumbing, Claudeforce is the first proof that a major AI platform can reach into Salesforce without compromising the governance model.

Agentforce Coworker and job-ready agents

Agentforce Coworker is the autonomous AI teammate built into Lightning. It answers natural-language questions, researches across CRM and Slack, plans, and orchestrates within the permissions of the logged-in user. Salesforce reported roughly 100,000 users in its first 35 days, with GA around Dreamforce 2026 (Salesforce).

Alongside Coworker, Salesforce shifted from pure DIY agent building toward job-ready, role-specific agents. Instead of building an agent from scratch, admins deploy agents pre-packaged for a role: Piper for inbound pipeline and SDR work, Hunter for outbound, Casey for service, Paige for operations, Carter for commerce, Marshall for supply chain and HR (Salesforce). Skills, context, and tooling come pre-configured rather than fully generic. The message to admins: stop building agents from zero. Deploy one tuned for the job, then customize.

Koa: Salesforce’s first CRM reasoning model

Salesforce and NVIDIA built Koa, Salesforce’s first purpose-built CRM reasoning model, based on NVIDIA’s Nemotron architecture (NVIDIA). Koa is trained for multi-step CRM workflows, tool use, and enterprise tasks rather than general language tasks. It went into pilot at Dreamforce 2026 with broader GA planned for later.

The signal: Salesforce is no longer relying solely on third-party frontier models for its AI layer. It is training its own model on CRM-specific patterns, which matters for accuracy in complex deal desks, service routing, and multi-step pricing logic where general LLMs still struggle.

CPQ end-of-sale and Revenue Cloud Advanced

In March 2025, Salesforce CPQ reached end-of-sale — not end-of-life. New customers can no longer buy it. Existing customers can renew, add licenses, and receive support and critical fixes (Salesforce). New feature investment goes to Revenue Cloud Advanced, which was briefly marketed under an Agentforce Revenue name before reverting.

A migration assistant is targeted for roughly December 2026. Winter ‘27 adds quoting capabilities on the Revenue Cloud side, closing the gap between CPQ and its successor (The Cloud Update). Coexistence is possible in some orgs, but the direction is clear: if you are on CPQ, start planning the move to Revenue Cloud Advanced.

Marketing Cloud: agentic campaigns and a reversion

Marketing Cloud gained agentic features through the Winter ‘27 release. The Account Discovery Agent builds a buying-committee view. The Campaign Agent generates content and runs adaptive campaigns. Personalized Paths in Flow add dynamic multi-channel orchestration (Salesforce).

Marketing Cloud Next is positioned as the modern flagship, built on the Salesforce Platform and Data 360. The umbrella name reverted from the temporary Agentforce Marketing back to Marketing Cloud, consistent with the broader reversal.

Commerce Cloud: agentic shopping and acquisitions

Commerce Cloud moved toward agentic commerce with Storefront Next (GA), the Carter Shopper Agent, and Agentic Commerce Search (Salesforce). Salesforce added support for open protocols like the Universal Commerce Protocol, improved merchandising and consent tools, and was named a Gartner Magic Quadrant Leader for Digital Commerce for the 11th consecutive year.

The Cimulate acquisition added AI-powered product discovery. Combined with the Carter agent, Commerce Cloud is building toward a shopping experience where an AI agent helps a buyer find, compare, and purchase without navigating a catalog manually.

Major acquisitions: Fin, Contentful, Cimulate, Braze

Salesforce made several acquisitions that shore up the agentic stack:

• Fin (formerly Intercom, ~$3.6B, closed September 2026) — proven high-resolution customer service agents, claiming roughly 76% autonomous resolution, deployable alongside customizable Agentforce (Intercom).
• Contentful — headless content management for the experience layer.
• Cimulate — AI-powered commerce discovery.
• Braze (~$5.66B, announced October 2026) — real-time multi-channel customer engagement, strengthening Marketing Cloud, Data 360, Agentforce, and SIforce (Braze).

The pattern: Salesforce is buying the gaps. Fin adds a proven agent for service resolution. Contentful adds the content layer for headless experiences. Cimulate adds commerce AI. Braze adds real-time engagement across channels that Data 360 alone could not cover.

Platform maturity: observability, long-horizon agents, and voice

Agentforce added features that move it from a launch to a production platform:

• Observability and Optimizer — analyze agent sessions and suggest fixes.
• Long-horizon agents — goals that span days or weeks, not single conversations.
• Multi-agent orchestration — Superagent and A2A patterns for agents that coordinate.
• Expanded voice — more languages, Field Voice, and employee voice.
• Custom scorers and A/B experimentation — test agent quality systematically.
• Conversation testing — validate agent behavior before deployment.
• Einstein Bots to Agentforce upgrade path — migrate legacy bots to the new agent runtime (Salesforce).

Data 360 as the foundation

Data 360 remains the foundation. Unified, governed data with Intelligent Context, Tableau Semantics, and strong ties to agents and headless access. The thesis: agents are only as good as the data they reach into, and Data 360 is how Salesforce ensures that reach is consistent, permissioned, and governed (Salesforce).

OpenAI partnership and builder tools

At Dreamforce 2025, Salesforce and OpenAI announced that GPT models are available in the platform, and Agentforce data and insights surface inside ChatGPT (OpenAI). Builder and admin tools gained conversational and “vibe” building: Quick Pages, Setup powered by Agentforce, and a conversational Agent Builder. Multi-framework support (React and Angular, micro-frontends) and a Headless Experience Layer for consistent UI widgets across surfaces round out the toolkit.

Governance and security

The headless shift raised the stakes on governance. Salesforce responded with Security Mesh, headless identity, MCP server controls (opt-in), permission inheritance, and zero-data-retention claims for partner model paths (Salesforce). Auditability runs through every layer: if an agent acts in Salesforce, the action is traceable to a permission, a user, and a policy.

Business results

Agentforce ARR reached approximately $1.5 billion, up roughly 240% year over year. AI and Data ARR approached $3.9 billion. Salesforce raised FY27 guidance after strong results, and the stock reacted positively to earnings, including a large single-day move. Some acquisition premiums, particularly Braze, drew mixed analyst commentary (Salesforce Investor Relations).

What this means for Salesforce admins and architects

Three takeaways for anyone running Salesforce today:

1. The UI is no longer the product. SIforce, Claudeforce, and the Headless Toolkit mean Salesforce expects work to happen in Claude, in Slack, in custom apps, and in agents — not only in Lightning. Start thinking about which surfaces your users actually need, and whether a headless integration serves them better than another console license.

2. CPQ is on a timeline. End-of-sale means no new customers. The migration assistant is coming. Revenue Cloud Advanced is where the investment goes. If you run CPQ, plan the migration now rather than when the assistant ships and everyone rushes.

3. Agents are becoming job-specific, not DIY. The shift from generic agent building to role-specific agents (Piper, Hunter, Casey, Carter) means your fastest path to value is deploying a pre-built agent and customizing it, not architecting one from scratch.