Dreamforce 2026 arrives at an interesting point for Salesforce.

A year ago, the company was still making the case for an enterprise built around AI agents. This year, Salesforce has a much bigger body of numbers and customer deployments behind that argument. 

Agentforce has crossed $1.5 billion in annual recurring revenue, Agentforce and Data 360 together are approaching $3.9 billion in ARR. And Salesforce says 7 billion Agentic Work Units have now been delivered across Agentforce and Slack.

There is another important development just weeks before the event. 

Salesforce and Anthropic have expanded their partnership through Claudeforce, bringing Claude into Salesforce workflows and allowing Salesforce data, business logic, actions, and governance to be accessed through another AI interface.

That changes the backdrop for Dreamforce 2026. The big story is likely to be about how far the company can take it into everyday enterprise work.

Here are the areas worth watching at this year’s Dreamforce events.

Agentforce will have more numbers to defend

Salesforce is going into Dreamforce with substantially more evidence of Agentforce adoption than it had a year ago.

In its latest quarter, Agentforce ARR exceeded $1.5 billion, up more than 240% year over year. Salesforce also reported 3.2 billion Agentic Work Units in Q2 alone, taking the cumulative total to 7 billion. 

Slackbot users grew more than 150% quarter over quarter.

These figures tell you that Salesforce is beginning to measure AI usage as an operating layer of its platform.

That makes the customer stories at Dreamforce especially important. Salesforce has already pointed to deployments such as PenFed’s 76-agent rollout, which is expected to save nearly $1.6 million, and UCLA Health’s customer-facing Agentforce deployment, which took eight months to move from testing to launch. 

At this year’s Dreamforce event, it will be worth looking beyond the success metrics and into how these deployments actually work. For example: 

• How much data preparation was involved? 
• Which systems had to be connected? 
• How were agents tested? 
• Where did humans remain in control? 
• How long did adoption take? 
• What happened when the agent failed?

Those details will reveal much more about the maturity of Agentforce. 

The economics of AI agents will become harder to ignore

There is a financial layer to all this that you will need to understand.

Agentic systems consume models, data, APIs, and compute. The more work an agent performs, the more important usage measurement becomes.

Salesforce has already introduced Agentic Work Units as a way to quantify agent activity. Its Q2 results also show just how quickly that activity is growing.

That creates a new set of questions for technology and finance teams:

• What does a typical agent-driven workflow cost? 
• How does that cost change as usage scales? 
• Which tasks are economical to automate? 
• How much human work is actually removed? 
• What happens to the cost model when agents operate continuously rather than when an employee actively uses an application?

Dreamforce should provide more clues about how Salesforce expects customers to think about these. 

The same applies to model choice. 

Salesforce is working with multiple AI providers, and its expanded Anthropic partnership makes that increasingly visible. The enterprise AI stack is becoming less dependent on a single model, which puts more attention on the layer connecting models to business data, permissions, workflows, and actions.

Will Dreamforce offer more clues about how Salesforce expects customers to make those tradeoffs? We’ll find out! 

AI will move closer to the systems where work happens

Salesforce is pushing its technology beyond the traditional CRM screen. Its MCP allows Slackbot to work with Salesforce data and connected applications, including retrieving customer information, updating records, and triggering workflows within Slack, subject to existing permissions and controls.

That makes Slack more than a collaboration tool in Salesforce’s AI strategy. It can become a place where employees find information, interact with agents, and initiate business processes without switching into another application.

For example, a sales team could discuss an account in Slack and bring relevant CRM information into the conversation. A service team could surface customer context without opening another application. Actions that once required a trip into a CRM interface could increasingly begin inside a workplace conversation.

This also raises a broader question about where enterprise work should happen.

Organizations rarely operate with one application. Many already use combinations of Slack, Microsoft Teams, Zoom, contact center platforms, and specialized business systems. If CRM and AI capabilities increasingly appear inside collaboration tools, companies will have to think about how those systems coexist, and which interface becomes the employee’s primary entry point to business information and actions.

For technology leaders, the interesting part is the architecture behind that experience, which raises several questions:

• How are permissions carried across systems?
• What gets logged?
• How are sensitive actions approved?
• How does an organization maintain a reliable record when work starts in Slack but changes a system of record somewhere else?

Those are the kinds of questions that should make the technical sessions at this year’s Dreamforce conference worth attending.

Data may be the less glamorous part of the AI story

AI demonstrations tend to focus on what an agent can do. Enterprise deployments depend heavily on what the agent can reliably access.

Salesforce is putting Data 360 at the center of its AI strategy, which makes data architecture an important part of the Dreamforce conversation.

For companies considering Agentforce, that means looking beyond the model itself:

• Is customer information accurate?
• Where does the data live? 
• Who owns it? 
• Which systems can an agent access? 
• Are permissions consistent across those systems? 
• How is information retained and audited?

These questions can determine whether an AI project works outside a controlled test environment.

Security will also have a significant presence at the event. Salesforce has a dedicated Trust and Security program covering agent governance, data protection, security training and controls for the agentic environment.

Sales teams will be watching the human side of automation

Salesforce is positioning agents to handle activities such as lead qualification, follow-up and pipeline administration. That could give salespeople more time for customer conversations and higher-value work, but it also changes how sales organizations divide responsibilities between people and software.

Look for examples involving seller productivity, pipeline quality, forecasting, response times and revenue. 

Also pay attention to the work that remains with salespeople. 

That’s because automation can change a job without eliminating the job. A sales representative may spend less time entering information and more time reviewing AI-generated recommendations, handling exceptions or managing important customer relationships. 

Dreamforce should offer a clearer picture of where Salesforce sees that balance heading.

Expect greater emphasis on reusable agent skills 

Another theme worth watching is the move from individual agents toward reusable capabilities.

The Dreamforce catalog includes a session on Agentforce Skills that shows how multi-step business tasks can be packaged and deployed across Agentforce, Slack, and third-party agents.

That suggests a future in which orgs build a library of governed capabilities rather than creating every agent workflow from scratch.

For developers and enterprise architects, this could become an important design question:

• Which processes should become reusable skills? Which needs custom development? 
• How are skills tested before wider deployment? 
• Who owns them after launch? 
• How are permissions and approvals handled when the same capability is used through several interfaces?

The answers will determine whether agentic AI becomes manageable at scale or creates another layer of enterprise complexity.

Dreamforce 2026 is also a test of Salesforce’s bigger bet

Salesforce has spent the past year expanding the definition of what its platform is.

The company’s latest results show that AI and data are becoming a meaningful commercial category. Its Anthropic partnership shows that Salesforce wants its systems to be useful through AI interfaces beyond its own applications. Its Slack strategy puts CRM actions into conversations. Its data investments aim to give those agents enough context to do useful work.

Dreamforce brings those pieces together.

The event takes place September 15-17 at Moscone Convention Center in San Francisco, with the Salesforce+ program running September 15-18. The virtual program will include more than 400 sessions, hands-on training, live launches, and interviews.

For anyone evaluating Salesforce’s AI strategy, the most valuable part of Dreamforce may be seeing whether all these pieces work together outside the keynote stage.

The technology has moved considerably since last year’s agent pitch. 

Now the harder test is production: useful work, reliable data, controlled actions, manageable costs, and measurable results.

That is where Dreamforce 2026 should get interesting.