Dreamforce 2026 Updates: From AI Agents to Enterprise Architecture

Introduction

Forget what you thought you knew about enterprise software. Dreamforce 2026 just tore up the rulebook.

For three days in San Francisco, Salesforce put a very different vision of enterprise software on display. AI agents were no longer presented as tools waiting for a prompt. They were built into how work gets done, across applications, data, and everyday workflows.

The shift was hard to miss. The conversation moved from what agents can do to where they can work, what they can act on, and how quickly enterprises can deploy them.

It also put last year’s vision into perspective.

At Dreamforce 2025, Salesforce introduced the idea of the “Agentic Enterprise” alongside a whirlwind of new concepts and renamed clouds. The direction was clear, but the practical questions remained: How do we actually build this? How do we govern it? And what happens when agents move beyond the demo?

Dreamforce 2026 was about answering more of those questions. Salesforce brought more of the machinery needed to make the vision work: new infrastructure, models, partnerships, and interfaces designed to let agents operate across applications, data, and workflows.

And that points to a bigger change in enterprise software.

The application is no longer necessarily where the work begins or ends. An agent can sit across the systems and data that make up a business process, moving between them as the task demands.

Enterprise software is beginning to look less like a place people work in and more like a layer that can work alongside them.

Here’s what stood out at Dreamforce 2026, and what Salesforce’s main keynote revealed about where enterprise AI is heading next.

Top 9 Announcements from Dreamforce 2026

1. Agentforce Gets More Specific With Seven Job-Ready Agents

Salesforce expanded Agentforce with seven job-ready agents designed around specific areas of work across sales, service, commerce, employee experience, demand generation, customer experience, and the back office.

• Seven Agents, Seven Jobs: Casey handles customer service; Paige supports IT and HR; Carter assists shoppers; Hunter handles outbound sales; Marshall manages supply chain processes; Piper handles inbound lead generation; and Fin handles customer experience workflows.
• Built Around the Work: Each agent comes with prebuilt skills, actions, and data models for its role, while companies can adapt these capabilities to their own processes, business rules, and security requirements.
• Long-Horizon Work: The new Agentforce runtime lets agents pursue goals over days and weeks, enabling them to continue working after an initial interaction ends.
• Memory and Continuity: Memory preserves context and progress, durable execution keeps plans running across sessions, and dynamic steering lets agents respond to new information and user direction.
• A More Connected Agentforce: Agentforce Coworker lets employees teach agents how they perform tasks, Multi-Agent Orchestration coordinates specialized agents, and Agent Optimizer helps teams test and refine agent performance.

2. AIforce: Bringing Salesforce Into the AI Tools People Already Use

Salesforce introduced AIforce, a new interface layer that brings Salesforce data, business logic, workflows, permissions, and governance into the AI tools where people already work.

• A New Interface Layer: AIforce brings Salesforce capabilities into interfaces such as Claude, Slack, and Agentforce Coworker, extending CRM data and actions beyond the traditional Salesforce UI.
• Headless by Design: The Headless Toolkit exposes Salesforce capabilities through MCPs, APIs, plug-ins, and skills, allowing them to be used across different applications and AI experiences.
• Work Without the CRM UI: Users can query Salesforce data, update records, and trigger workflows without opening the traditional Salesforce interface.
• Governance Stays Intact: Actions still follow existing Salesforce permissions, business rules, and governance controls, with requests routed back through Salesforce.
• Three Ways to Bring Salesforce Into the Workflow:

– Claudeforce: Connects Salesforce with Claude through a prebuilt MCP server and 37 sales skills, giving Claude access to Salesforce context and actions. No new permissions model is required; Claude operates within the user’s existing Salesforce permissions and business rules.

– Slackforce: Infuses Salesforce intelligence directly into Slack conversations, utilizing Slackforce Surfaces to turn text chats into dynamic, interactive live canvases and data dashboards.

– Agentforce Coworker: Puts an AI teammate inside Lightning, giving employees another way to work with Salesforce data and processes.

– The Interface Paradigm Shift: The core Salesforce UI officially becomes just one interface among many, while the underlying data, permissions, and business logic remain the central system underneath them all.

Related Read: What Has Changed in Slack-Salesforce Integration in 2026? 

3. Koa: Salesforce Builds a Reasoning Model for CRM

Salesforce and NVIDIA introduced Koa, Salesforce’s first CRM reasoning model for Agentforce, built on NVIDIA Nemotron and trained specifically for complex enterprise workflows.

• Purpose-Built for CRM: Koa is post-trained on synthetic scenarios modeled on nearly three decades of Salesforce CRM knowledge across 14+ industries.
• Built for Multi-Step Work: It is designed to reason through workflows, select the right tools, and take actions step by step, rather than simply generate an answer.
• No Customer Data Used for Training: Salesforce says Koa’s training corpus uses synthetic scenarios, not customer data.
• Salesforce-Controlled Model: Salesforce controls the model weights and runs post-training and inference within its own trust boundary.
• Focused on CRM Actions: On Salesforce’s CRM benchmark, Koa matches or exceeds leading model performance on CRM actions while producing three times fewer errors.

4. Enterprise AI Harness: The Architecture Behind Salesforce’s Agentic Push

Salesforce introduced its Enterprise AI Harness, a composable architecture designed to give agents the context, reasoning, actions, governance, security, and models they need to operate across the enterprise.

• Six Trusted Capabilities: Brings together Trusted Context, Agency, Action, Governance, Security, and Models in a common architecture.
• AI Control Plane: Gives enterprises one place to discover, register, manage, evaluate, observe, and control agents, and AI across Salesforce and third-party systems.
• Model Choice: Lets enterprises connect different models and route work based on factors such as accuracy, performance, cost, and business requirements.
• Composable by Design: Capabilities can be used together or independently and can work with Salesforce technology, existing enterprise technology, or both.
• Built for an Open Ecosystem: Salesforce is making the architecture accessible through MCP, APIs, Skills, and Plug-ins, extending it into AI experiences such as Claude, Slack, and Microsoft Teams.

5. Headless 360: Turning Salesforce Capabilities Into Building Blocks for Agents

Salesforce is extending Headless 360 across its platform, turning capabilities traditionally accessed through Salesforce applications into reusable services that authorized AI agents can discover and use.

• Every Cloud Goes Headless: Sales, Service, Commerce, Marketing, MuleSoft, Informatica, Tableau, and other Salesforce capabilities can be exposed beyond their traditional applications.
• MCP-Native Access: The Headless 360 MCP Server lets AI agents discover and invoke Salesforce capabilities through open standards.
• Business Logic Travels with the Capability: Agents inherit the metadata, permissions, workflows, validation rules, and governance already established in Salesforce.
• 100+ Reusable Skills: Salesforce is packaging business logic and orchestration into reusable Skills and plug-ins for developers and agents.
• Built for Any Surface: These capabilities can power AI agents, web, and mobile applications, embedded experiences, and conversational interfaces.

6. AgentExchange: Building an Ecosystem Around Enterprise Agents

Salesforce is expanding AgentExchange beyond the traditional AppExchange model, bringing together apps, agent skills, plugins, MCP servers, tools, and AI-native experiences that can extend what agents can do.

• From Apps to Agent Capabilities: AgentExchange now includes skills, plugins, MCP servers, tools, and new AI-native experiences alongside traditional apps.
• Headless Toolkit Opens the Door: Partners can use Salesforce’s Headless Toolkit to bring Salesforce data, business logic, and actions into their own applications and AI experiences while respecting existing permissions and security controls.
• A Growing Partner Layer: Companies such as Lovable, Replit, and Vercel are building new interfaces on Salesforce, while Gamma, Granola, Legora, Rillet, and others are bringing Salesforce context into the tools where teams already work.
• MCP Connects Agents to More Systems: AgentExchange includes MCP servers from companies such as Qualtrics, Demandbase, Glean, Parallel, and Coveo, giving agents access to information and capabilities beyond Salesforce.
• Skills and Plugins Become Reusable Building Blocks: Salesforce now has more than 170 skills and plugins available through AgentExchange, covering areas from administration and sales to Data 360 and developer workflows. Partner-built capabilities are planned for the marketplace as well.

7. Google Cloud and AWS: Salesforce Opens the Enterprise AI Stack

Salesforce expanded its partnerships with Google Cloud and AWS around a simple enterprise reality: AI does not live in one cloud, model, or application. The new integrations connect Salesforce data, agents, models, and workloads across cloud environments without requiring customers to move their existing systems.

• Salesforce + Gemini Enterprise: Agents on Salesforce and Google Cloud can reason across shared data and act across platforms without custom integrations.
• Salesforce on Google Cloud: Salesforce workloads can run on Google Cloud infrastructure through Hyperforce.
• Salesforce in Amazon Quick: Teams can access Salesforce data, context, and skills through Amazon Quick, including pipeline, account, and service information.
• AWS Agents in Slack: AWS agents can work inside Slack, bringing capabilities such as DevOps and security assistance into existing team conversations.
• More Model Choice: Agentforce customers can access models via Amazon Bedrock, including models from Anthropic and NVIDIA, with OpenAI models planned for Agentforce.
• Zero-Copy Data Access: Salesforce Data 360 can connect with AWS data sources, including S3 and Glue-managed Iceberg tables, Aurora, RDS, and SageMaker Lakehouse, without duplicating the data.
• Voice Across Platforms: Agentforce Voice and Amazon Connect Customer are adding A2A support for real-time, bidirectional communication between AI agents.

8. Missionforce Brings Agentic AI Into Government Operations

Salesforce expanded Missionforce with purpose-built AI for government, alongside new partnerships with OpenAI and NVIDIA. The focus is controlled AI deployment across policy, operations, and fieldwork, including restricted environments.

• OpenAI Models: Frontier models will support Missionforce through Salesforce Government Cloud and Amazon Bedrock, with authorized workflows also accessible through ChatGPT.
• NVIDIA Models: Agencies can train and deploy mission-specific models on their own data and infrastructure.
• Policy to Rules: Missionforce Policy Engine converts approved policies into structured, testable rules, with human review before deployment.
• AI in Restricted Environments: Missionforce Operations supports workflows such as procurement and invoice audits in private-cloud and air-gapped environments.
• TSA Puts Agentforce to Work: The Transportation Security Administration’s Ace handles around 100,000 traveler conversations each month, resolving 96% of routine inquiries without escalation. The deployment shows how government agencies can put agentic AI to work in a high-volume public-facing service.

9. Agentforce in Action: From Engineering to Employees to Fans

Beyond the platform announcements, Dreamforce 2026 showed Agentforce moving into real-world work and customer experiences. Siemens, Adecco Group, and Live Nation offered three very different examples of what that looks like in practice.

A. Siemens: Bringing Engineering Intelligence Into Sales and Service

Salesforce and Siemens are extending Agentforce into industrial sales and service by connecting it with Siemens’ Teamcenter Service Lifecycle Management (SLM). The idea is to bring engineering-grade product knowledge into the customer workflows where sales and service decisions actually happen.

• Engineering Knowledge at the Point of Work: Teamcenter works with Agentforce to bring product and engineering context into sales and service workflows. Service technicians can identify the right spare parts for a specific machine, while sellers can quote technically valid, manufacturable upgrades.
• 100% Inbound Lead Engagement: Siemens now engages all inbound leads across 132 countries using Agentforce, with separate engagement and qualification agents handling outreach, assessment, and routing.
• Agentic Partner Onboarding: Siemens plans to use Piper for partner acquisition, Slackbot for onboarding orchestration, and Agentforce Operations powered by Regrello to complete steps in SAP.

B. Adecco: Bringing an AI Coworker Into Daily Work

Adecco is rolling out Agentforce Coworker across more than 40 countries, giving employees access to Salesforce data, systems, and organizational knowledge through a single interface.

• 40+ Countries: The rollout provides employees a common way to access business information and work across Salesforce systems.
• Recruitment and Sales Workflows: Teams can find prospects, prepare briefs, enrich data, track leads, and launch recruitment agents directly from Coworker.

C. Live Nation: Bringing Agentforce to the Fan Experience

Live Nation is using Agentforce to handle venue questions and support fan experiences across its U.S. venues.

• 24/7 Venue Support: Its Venue Agent answers questions about parking, accessibility, door times, and venue information.
• One Agent, Multiple Venues: Data 360 helps the agent identify the right venue information without requiring a separate agent for each location.
• Early Results: Live Nation’s BottleRock pilot handled 37,000+ fan interactions, while its current Venue Agent answers 95% of questions without human handoff.

Prepare for 2027: When Agents Become Infrastructure 

Taken together, the announcements at Dreamforce 2026 point to a change that is easy to miss when each launch is viewed on its own. Salesforce is building for a world in which people, applications, data, models, and agents can work across the same business process.

That raises a few practical issues for the year ahead:

• The application interface matters less. AIforce and Headless 360 allow Salesforce capabilities to surface in Claude, Slack, custom applications, and agent workflows. The CRM remains the source of business data and logic, but employees do not always need to enter Salesforce to use it.
  

• Business context carries more weight. Koa is built around CRM work. Siemens is bringing engineering knowledge into sales and service. Missionforce is turning approved policy into executable rules. In each case, the useful part of AI comes from knowing the rules, history, products, and processes behind the task.
  

• More agents create a management problem. Different agents may use different models, tools, data sources, and permissions. The Enterprise AI Harness, MCP connections, and governance controls point to the infrastructure needed to keep those moving parts under control.
  

• Enterprise architecture has to absorb constant change. Models will be replaced. New agents will appear. New interfaces will come and go. Business logic and governance cannot be rebuilt each time. That makes reusable capabilities, shared controls, and open connections notably important.

Where Does This Leave Things? Dreamforce moves at vendor pace. Adoption moves at the pace of an organization’s data, permissions, processes, and internal agreement. The work ahead is to define what data agents can trust, who can access it, which system takes precedence, and where agents must stop.

That is ultimately what Dreamforce 2026 changes. AI is moving closer to the systems where work actually happens, and the architecture around those systems now has to support it.