Top 7 Game-Changing Announcements From Dreamforce 2026

AIforce was arguably one of the biggest announcements of Dreamforce 2026. It introduces a live interface layer that brings Salesforce data, workflows, business logic, permissions, security, and governance into the AI interfaces where people already work.

Instead of always opening Salesforce first, employees can interact with business information through interfaces such as Claude and Slack. They can ask questions, update records, and trigger workflows while still working with Salesforce context and permissions.

• AIforce is designed around dynamic, composable interfaces.
• Existing Salesforce permissions and business rules remain part of the experience.
• AI can reason across connected business information instead of simply retrieving individual records.
• The approach also supports MCPs, APIs, plug-ins, and skills for building further experiences.

Koa is Salesforce’s first CRM reasoning model, developed with NVIDIA and built on NVIDIA Nemotron. Rather than being designed as a general-purpose model, Koa is focused on the complex, multi-step reasoning required for enterprise CRM work.

The model was developed using synthetic data based on nearly three decades of CRM knowledge. Its purpose is to help agents understand business context, select the right tools, and work through multi-step processes more effectively.

• Koa is designed specifically around CRM tasks.
• It supports longer and more complex business workflows.
• The model works as part of the wider Agentforce architecture.
• NVIDIA technology is being used to support its underlying model and computing capabilities.

Koa adds an important layer to the Dreamforce 26 conversation because the future of enterprise AI will require models that understand more than language. They will need to understand how actual business processes work.

Slackforce Surfaces brings Salesforce and other connected data into interactive experiences inside Slack. Users can describe what they need in natural language and create live dashboards, reports, and interfaces that teams can explore and act on together.

• Pulls live context from Salesforce, Slack, and other connected tools.
• Lets teams filter, explore, comment, and take action.
• Keeps information connected to the conversation where work is already happening.
• Turns Slack into a more active workspace for data and business decisions.

Slack Code Brings AI Development Into the Team Workspace

Slack Code creates dedicated code channels where teams can work with AI coding agents, review plans and code diffs, and see live previews together. It supports agents including Claude Code, Devin, GitHub Copilot, ChatGPT, and Vercel agents.

Hunter and Piper show how Agentforce is moving beyond simple prompts. Hunter can work on outbound sales goals across days and weeks, while Piper focuses on inbound leads, engaging prospects, qualifying them, and helping turn conversations into pipeline.

• Hunter works across research, outreach, account context, and sales goals.
• Piper handles inbound conversations across websites and inboxes.
• Both are designed around completing business work rather than simply answering questions.
• Long-horizon capabilities allow agents to continue working toward defined goals over time.

Agent Optimizer focuses on what happens after an agent has already been created. Instead of treating deployment as the final step, it helps organizations observe agent performance, identify problems, and improve how those agents operate.

This creates a more continuous approach to agent management, connecting building, testing, monitoring, and optimization rather than treating each stage as a completely separate activity.

• Agents can be monitored after deployment.
• Performance issues can be identified more systematically.
• Improvements can be made based on observed results.
• The approach supports the idea of continuously refining digital workers.

For businesses moving deeper into agentic workflows, this matters because having more agents also means having more things to monitor. DF26 showed that scaling AI will require management and optimization alongside creation.

Claudeforce extends the Salesforce and Anthropic relationship by bringing Salesforce intelligence directly into Claude. Salesforce in Claude includes a prebuilt MCP server and 37 prebuilt sales skills, reducing some of the setup traditionally required for connecting AI with enterprise systems.

The experience allows users to work with Salesforce information through Claude while keeping business context and governed actions connected to the underlying Salesforce environment. This makes Claude another interface through which Salesforce capabilities can be accessed.

• Salesforce context can be accessed through Claude.
• 37 prebuilt sales skills are included.
• The prebuilt MCP server reduces manual integration work.
• Sales teams can work with CRM information without constantly switching between separate environments.

Claudeforce also reinforces one of the strongest Dreamforce 2026 themes: the AI interface does not always need to belong to the system that stores the business data.

Headless 360 pushes the platform toward a model where Salesforce capabilities can be accessed without depending on the traditional Salesforce interface. APIs, MCP, and other tools allow data and business functionality to reach different applications and AI experiences.

This creates more flexibility for developers and businesses because the same underlying business logic can support different interfaces. AI agents, applications, and other experiences can work with Salesforce capabilities without requiring every user to follow the same screen-based workflow.

• APIs and MCP expand how Salesforce capabilities can be exposed.
• Business logic can operate independently from a fixed interface.
• Developers can create experiences around existing Salesforce capabilities.
• Headless architecture supports the broader move toward AI-first interactions.

Together with AIforce, this announcement makes the direction of Dreamforce 26 easier to understand. The platform is not only becoming more intelligent; it is also becoming less dependent on one specific way of accessing it.