Dreamforce 2026 marked a decisive pivot toward Salesforce’s headless and agentic strategy. That wasn’t a surprise. Salesforce had been laying the breadcrumbs since its TDX developer event back in April, and I’ve been tracking the buildout since then in Ahead of Dreamforce, Salesforce Loses It’s Head (In a Good Way). But, due to that, I also heard some grumbling on the show floor that the historic Customer 360 applications didn’t get enough airtime this year.
I don’t disagree with that read, but the more important story is that Salesforce is using agents to unlock the value of those existing applications, and to become the center of a customer’s agentic strategy. Here are the announcements that stood out.
How Headless 360 Turns Agents Into a New Class of User
Agents are becoming a first-class user profile, and every SaaS platform now has to figure out how to expose its value to them. Most vendors have shipped a handful of MCP endpoints and called it done. Salesforce went further: under the AIforce banner, it is positioning a harness that is designed to interoperate with other agent harnesses rather than lock customers into its own. A Claude user, for instance, does not have to hand-configure Salesforce’s roughly fifty MCP servers one by one. Instead, Salesforce and Anthropic’s new Claudeforce partnership ships a prebuilt plugin with 37 skills at launch that handles workflow and MCP orchestration outside the client, which according to Salesforce should make for a cleaner user experience and lower token overhead on the client side. Salesforce also has harness-specific integrations coming for Slack and OpenAI, plus a toolkit for building your own.
What Agentforce’s Named Digital Workers Change for Sales and Service
Salesforce is folding its Qualified and Fin acquisitions into Agentforce as named digital workers with actual job titles, rather than generic building blocks a customer has to assemble. Until now, Agentforce has mostly been a build-and-manage environment for agents. These new “hires”, Piper (from Qualified) on inbound pipeline generation, Fin on customer service, and Hunter on outbound sales, are designed to be tuned to a company’s own business rules out of the box. The proof point that stuck with me: Salesforce CEO Marc Benioff said onstage that Hunter generated $500 million in pipeline for Salesforce last quarter alone. That is a company-stated, self-reported number rather than an audited one, but it is still a real signal that these roles are doing more than demo work.
Where Salesforce’s Agent Pricing Experiment Stands
I have been fairly critical of the state of agent pricing this year (see Is AI Agent Pricing Getting Better? Grading My 2025 Predictions), but Salesforce deserves credit for continuing to experiment rather than standing pat. This year, Salesforce is sticking with its Flex Credit system but is also adding user-based bundles of Customer 360 apps paired with a preset amount of AI Flex Credits. That is an interesting hedge: Flex Credits look more like hyperscaler-style consumption pricing, while user-based packaging is the model Salesforce actually built its SaaS business on. It is not yet clear which pricing model wins long term, and the answer will probably vary by business case, but real choice is an advantage while enterprises adjust to agent economics.
Why Slack Is the Proving Ground for Headless Agents
Salesforce pitches its headless strategy as bringing whatever agent or coworking client a customer wants. Slack, though, keeps innovating in its own right, and it has clearly become the proving ground for a lot of the underlying headless technology. After meeting with the Slack team, what stood out most is how agents inside Slack add a collaborative angle that other coworker tools, which still operate as individual productivity tools, do not really match.
What Koa Signals About Purpose-Built Models
Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman both shared the Dreamforce stage this year, in the middle of the industry’s public debate over whether frontier labs should slow down. Against that backdrop, Salesforce and NVIDIA announced Koa, Salesforce’s first CRM reasoning model, post-trained on NVIDIA’s open-weight Nemotron 3 Super using synthetic data modeled on nearly three decades of CRM work rather than customer data, according to Salesforce. I dug into what that combination could mean for cost and accuracy in Salesforce Brings a New Angle to the Model Wars.
Will These Moves Accelerate Adoption?
The core Customer 360 applications did not get much airtime at Dreamforce this year, but nearly everything that did get a marquee announcement was aimed at driving more value out of those apps rather than replacing them. Salesforce is betting that it can grow its existing base and attract new customers with this approach. I believe that is the right instinct: the durable asset is still the backend data and workflow logic Salesforce has hardened over decades, and these announcements are a new way to unlock it.
That instinct still has to survive contact with customers. I was reminded at the event that Salesforce once had to sell two things at once: the viability of CRM as a category and SaaS as a way to buy software at all. It largely won that argument.
Salesforce’s current challenge with agents is similar. The entire Salesforce ecosystem has to convince buyers that agents deliver ROI beyond the core applications, and that Salesforce, not a hyperscaler, can be the core agentic infrastructure layer. None of that undercuts the core bet: treating agents as infrastructure rather than an add-on is the right long-term posture for an incumbent SaaS vendor facing agentic disruption. But now that the technology direction is clear, Salesforce will also need to adapt its go-to-market and partner strategies to match it, in a way that still resonates with its existing, application-centric stakeholders.
