Salesforce Dreamforce 2026: AIforce, Headless 360, and What It Means for Your Stack
Dreamforce 2026 had one central message: Salesforce is making its own interface optional. The company that built a $200 billion business on the premise that humans log into its platform every day just spent its biggest event of the year building infrastructure for agents to access Salesforce without ever opening the UI.
AIforce, Headless 360, Koa, and AgentExchange — here is what each announcement means and what it changes for enterprise teams building automation workflows.
The 4-Layer Stack Salesforce Presented
Salesforce unveiled a new architectural vision organized in four layers:
1. Data 360 — the foundation: data, metadata, and agent memory 2. Customer 360 — application logic and semantic intelligence (the existing Salesforce core) 3. Agentforce — the agent execution layer (announced last year, now with new capabilities) 4. AIforce — the interface layer on top: Claude, Slack, Lightning, or any surface
The implication of this stack: the UI is now one option among many, not the required path. Agents sit between the data and the user, executing actions without the human ever navigating to a Salesforce page.
Headless 360: The Foundation
Announced August 19, 2026, Headless 360 is the infrastructure that makes everything else possible. It exposes every Salesforce cloud as a reusable enterprise capability accessible via:
– API — standard programmatic access – MCP (Model Context Protocol) — the emerging standard for agent-to-tool communication – CLI — for developer workflows
Salesforce Headless 360, announced August 19, 2026, exposes every Salesforce cloud — Sales Cloud, Service Cloud, Marketing Cloud, and others — as reusable enterprise capabilities that any authorized AI agent can discover and use through open standards. The Headless 360 MCP server offers more than 60 tools covering data access, workflow execution, approval routing, and record management, all accessible without opening the Salesforce interface. This architecture means AI agents including OpenAI’s GPT-6 Astra, Anthropic’s Claude, or custom-built agents can execute Salesforce workflows — updating records, routing approvals, managing cases — through the MCP protocol without requiring a licensed human user to navigate the UI. Headless 360 forms the foundation of AIforce, Salesforce’s composable interface layer announced at Dreamforce 2026.
The 60+ tools in the Headless 360 MCP server cover: – Data access (read/write across all Salesforce clouds) – Workflow execution (trigger flows, process automation rules) – Approval routing (submit, approve, reject approval processes) – Record management (create, update, delete, relate)
What this changes: Previously, automating Salesforce workflows required either a human navigating the UI or a developer building custom API integrations. Now, any MCP-compatible agent can discover and use Salesforce capabilities through a standard protocol — with no custom integration work required.
AIforce: The Interface Layer
AIforce is the composable interface layer built on Headless 360. It is not an agent itself — it is the infrastructure that lets agents from any vendor interact with Salesforce data and workflows.
Key components:
Claudeforce — Salesforce integration with Anthropic’s Claude. Claude can directly query Salesforce data, trigger workflows, and update records through a dedicated integration layer.
Slackforce — Salesforce capability surfaced natively in Slack. Agents can execute Salesforce actions from within Slack conversations without navigating to salesforce.com.
Agentforce Coworker — Human-facing agent that works alongside employees in any surface (Slack, Lightning, or Claude) — answering CRM questions, generating pipeline reports, updating records on command.
Headless Toolkit — Developer primitives for building custom agent integrations on top of Headless 360.
Koa: The CRM-Specific Reasoning Model
Perhaps the most underreported announcement at Dreamforce: Koa, a specialized reasoning model trained specifically for CRM tasks.
Rather than using a general-purpose frontier model (GPT-6, Claude, Gemini) for all Salesforce operations, Koa is optimized for: – CRM data understanding and relationship mapping – Sales motion reasoning (deal progression, opportunity scoring) – Salesforce-specific workflow logic – Quote-to-cash and service case resolution
Think of Koa as what happens when a vertical SaaS vendor trains its own domain-specific model — the same strategy Harvey AI uses for legal, Waystar for healthcare billing. Salesforce is betting that domain-specialized reasoning outperforms general models for CRM-specific tasks.
Cost advantage: Koa is expected to be significantly cheaper than frontier models for in-Salesforce tasks, since it requires less context to understand CRM-specific instructions.
AgentExchange: The App Store for Salesforce Agents
AgentExchange extends the existing Salesforce AppExchange model to the agentic era. Rather than listing SaaS integrations, it surfaces pre-built agents and agent components that snap into the AIforce/Headless 360 stack.
Use cases available at launch: – Pre-built agents for common sales motions (pipeline review, forecast generation) – Service Cloud agents for tier-1 case resolution – Industry-specific agent packages (financial services, healthcare, manufacturing)
The significance: organizations do not have to build agents from scratch. They can deploy pre-certified agents from the AppExchange in the same way they previously deployed Salesforce apps.
What the Licensing Model Shift Looks Like
Salesforce has not eliminated per-seat pricing — but it has created a parallel pricing model that acknowledges agents as users:
| Scenario | Old model | New model (2026) |
|---|---|---|
| 10 humans use Salesforce daily | 10 seats × $150 = $1,500/mo | 10 seats × $150 = $1,500/mo |
| 1 agent handles work of 5 humans | 5 seats × $150 = $750/mo | 1 Agentforce capacity license + consumption |
| Full agent-first deployment | Not possible | Headless 360 MCP access + outcome-based |
Organizations that shift to agent-first deployment will see lower per-unit costs but need to monitor consumption carefully — Agentforce usage is metered, and multi-agent orchestration at scale can generate significant consumption.
3 Immediate Actions for Salesforce Customers
1. Evaluate your Headless 360 eligibility Headless 360 access depends on your current Salesforce edition and contract. Contact your Salesforce account executive to understand what is included in your existing contract vs. what requires add-on licensing.
2. Audit your current integration architecture If you have custom API integrations connecting agents or automation tools to Salesforce, compare the maintenance overhead against switching to the Headless 360 MCP server. For many teams, the MCP approach will be simpler and more maintainable.
3. Pilot one AgentExchange agent before building custom Before investing engineering time in a custom agent, check AgentExchange for pre-built options covering your use case. Pre-certified agents reduce time-to-value and compliance risk compared to custom builds.
The Broader Signal
Salesforce’s Dreamforce 2026 announcements confirm what the market has been pricing since February: the human as the primary interface to enterprise software is being replaced by the agent as the primary interface.
This is not a distant prediction — it is the product roadmap of the world’s largest CRM vendor, announced at their biggest event of the year, with specific products and pricing available today.
For automation teams, the window to build expertise before this becomes table stakes is measured in months, not years.
For the full context on how agentic software is reshaping the SaaS model, read our AI Agents vs SaaS 2026 guide. For the model powering many of these workflows, see our GPT-6 Astra enterprise guide. For the security implications, read our AI Agent Security Crisis 2026 analysis.
Published September 26, 2026. Sources: Salesforce Dreamforce 2026 announcements, Ranosys Dreamforce recap, CX Foundation, Salesforce blog, Cloudsheer Dreamforce 2026 analysis.