Heading into its major annual conference next week, Salesforce presses forward with a bigger set of numbers than it had a year ago.
Agentforce's annual recurring revenue is past $1.5 billion, up 240% year over year. The company also says it's completed 7 billion Agentic Work Units, its own measure of AI-finished tasks, across Agentforce and Slack combined.
For anyone running automation or productivity strategy, Dreamforce 2026 represents an opportunity to pick apart which announcements will actually deliver real ROI, and which ones remain symbolic developments in the agentic workplace.
To help you get started, here are five topics and next steps you should keep in mind:
TL;DR
Do you use a Slack stack? Is your CRM hosted on Salesforec? Ahead of Dreamforce 2026, it's worth picking out the key product launches that will directly impact your business.
Claudeforce puts Anthropic's Claude behind Agentforce, Slack, and the Atlas Reasoning Engine as the default model.
Automation volume gets reported a lot at events like this. Whether that volume reflects real task completion is a separate question worth asking directly.
1. UCLA & PenFed: Use Case Studies To Your Advantage
Salesforce's Q2 FY2027 results put combined Agentforce and Data 360 ARR at nearly $3.9 billion, up more than 210% year over year. Data 360 took in 104 trillion records during the quarter, a 355% increase. Slack had its fastest quarterly growth in net new order value since the acquisition, and Slackbot users grew more than 150% in three months.
What Is an Agentic Work Unit?
Salesforce's own metric for discrete AI-completed work, like updating a record, closing a case, or triggering a workflow. It's meant to measure agent output beyond simple usage numbers.
Large growth figures like these say something about adoption, but not much about whether the work is landing well in practice.
Major conferences like Dreamforce are an opportunity to see end-users and practitioners reflect on their adoption journey.
For example, PenFed's 76-agent rollout across operations, mortgages, IT, and HR is projected to save the credit union close to $1.6 million this year. Equally, UCLA Health got a customer-facing Agentforce use case live within eight months of setting up its testing and approval process.
Both are worth asking about directly on the show floor, particularly what it actually took behind the scenes to get there. UC Today's full guide to Dreamforce 2026 goes deeper on these earnings figures and customer stories.
2. Claudeforce Swaps in a Different Brain for Automation
Salesforce and Anthropic have jointly announced Claudeforce, putting Claude's reasoning model to work inside Salesforce's automation tools. Claude now runs the Atlas Reasoning Engine, is the default model for Agentforce Vibes and Agentforce Coworker, and is available inside Agent Builder for anyone building custom agents.
The first customer-facing piece, Salesforce in Claude, comes with 37 built-in sales skills for things like meeting prep, deal health checks, and pipeline updates. It's live with a handful of pilot customers now, and Salesforce expects an open beta in September.
Why It Matters for Automation
Changing the reasoning model behind an agent changes what it can actually get done on its own, not just how polished the output sounds. The Marc Benioff and Dario Amodei keynote is likely where Salesforce lays out its clearest argument for what that shift means in practice.
If you're harnessing Salesforce in your sales team, or you're just a major fan of Anthropic's products, this is one to keep an eye on for Q4.
3. Headless 360 Is Betting on Reuse Instead of Rebuilding
Headless 360 hasn't gotten as much attention as Claudeforce, but it might matter more long-term. The idea is that Salesforce's data, workflows, permissions, and business logic become available to outside AI agents through open standards, so an organization isn't rebuilding the same integration every time a new AI tool shows up.
The rollout includes a Headless 360 MCP Server, a Data 360 MCP Server, a Slackbot MCP Client, and over 100 reusable Agent Skills and Plugins.
What Is MCP?
Model Context Protocol, an open standard that lets AI systems connect securely to outside data and tools. It means an agent can pull information or take action across different platforms without a custom integration built for each one.
The pitch is that the same underlying capabilities work across Salesforce, Claude, ChatGPT, Cursor, and other platforms, without security controls getting rebuilt from scratch each time.
