Dreamforce is usually a performative exercise, one in which Salesforce uses the gathering to solidify messaging around products that act as tentpole explanations of its strategy, even if many of those announced products aren’t going to be GA for the better part of a year. This year was markedly different. Product announcements were less flashy, but potentially far more consequential to the company and its buyers. This year’s Dreamforce was a window into a Salesforce that is changing quickly, but is still grappling with the diffusion of AI agents into enterprise processes in ways it cannot control. For example, the 2026 ISG AI Impact survey found that 51% of respondents said that time to complete routine work has stayed the same or increased as a direct result of AI initiatives.
Salesforce is trying to move the enterprise conversation away from agents and toward consideration of what kind of business architecture can support those agents’ actions. (It’s pretty telling that the company quietly—with no apparent formal announcement—reversed its major branding exercise of 2025. Service Cloud is Service Cloud once again, not Agentforce Service. Same with Sales Cloud, etc.)
Changing the conversation involves elements of re-rebranding, but it also makes much of the apparent direction of Salesforce’s products less tangible and more confusing for buyers in the short term. I have sympathy for this approach because it is setting the stage for an environment in which new modes of work (and the contours of the supporting tech stack) have yet to take shape. Salesforce faces an existential transformation, but in order to get there it needs to position its pieces on the game board differently. This Dreamforce event was somewhere between a “throw it all at the wall and see what sticks” moment and a measured transitional approach to rethinking CRM.
What Salesforce actually accomplished here was important and will likely be remembered as critical to that transformation. It articulated a multi-layered framework for agents in the enterprise, one in which it does not pull all the levers. In this framework, the agents themselves are components that need to be supplied with context, supported by applications and workflows and controlled by an execution layer. And atop that is an interface layer that’s very different from the traditional Salesforce approach via Lightning.
Where last year the focus was on the agent-as-product, now there seems to be a belated recognition that you can’t get businesses to sign on to Salesforce’s conception of the “agentic enterprise” without a strong control structure that is already in place (now, not later), and that’s integrated with the entire developing AI infrastructure ecosystem.
Salesforce was at pains to distinguish its role as the supplier of context, not necessarily of the AI models. Though this message, too, is a little muddy: Salesforce announced a CRM-specific model called Koa built on NVIDIA tech. So I assume that Salesforce’s position is that custom or specialized models will handle domain-specific tasks where a general-purpose model doesn’t have enough context or is too large (or expensive) for an activity. This puts a premium on enterprises’ own data and allows Salesforce to occupy the role of context supplier that AI companies further upstream cannot fill.
It was an interesting sideshow to have three of the key AI industry executives (Sam Altman, Dario Amodei and Jensen Huang) all in speaking roles, all seemingly contradicting each other gently and making the point that Salesforce knows its role as a supporting player (“context”) with critical ties to everyone in the space.
And nothing illustrates that supporting role more than the very prominent declarations about how users are now able to (even encouraged to) bypass the core Salesforce interface and access Salesforce from within other applications or through platforms like Slack. Credit where credit is due, I did not envision Slack becoming as prominent an entry point for agentic operations as it has. It acts as a powerful way for users to engage with agents, give instructions and receive results, becoming, in effect, the human/agent interface environment. Agents that live within Slack are present where the users live, making them more likely to weave their way into those users’ daily work lives.
This interface shift is an interesting echo of the long conversation in the contact center industry over who owns the CSRs’ desktop: CRM vendors or CCaaS vendors. Now that Salesforce is a CRM company with a CCaaS product, its strategy there mirrors its overall AI strategy, as a version of the context story.
Salesforce argues that conventional contact centers were designed around managing interactions, whereas an “agentic” contact center should orchestrate outcomes across
The company’s recent CCaaS pitch argues that because Salesforce already contains customer identity, history, workflow and business logic, an AI agent can do more than answer a question: it can actually modify an account, process something, update an application, initiate fulfillment, make an offer or hand the case to a human with the relevant context intact. Salesforce crossed into real CCaaS this year by adding native telephony to its stack. It’s clear that Salesforce’s contact center offering is not immediately intended to compete with longstanding dedicated CCaaS products, but to be an option for enterprises trying to apply some of the CX-related context and interdepartmental workflows to their interaction-handling engines. (Salesforce was covered in two CX-related ISG Buyers Guides this year, for Contact Centers and for Customer Experience Management.)
Salesforce’s implied argument—which holds merit—is that if it controls elements of customer context, AI agents, workflows and the human desktop, then the traditional CCaaS platform risks being pushed downward into a communications-infrastructure role. Like Amazon with Connect ten years ago, Salesforce can disrupt over time, increasing its footprint in the contact center and helping pull that center closer to a business’s core customer experience and orchestration teams.
Dreamforce tells us what Salesforce thinks it has to do and say to maintain its strength and extend its position into new elements of the enterprise stack. Right now it is going to greater lengths than usual to tell the world that the stack itself is quickly changing and to reassure buyers that despite the hype over agents, the company hasn’t lost the thread. It understands what elements it controls (context) and what its value is to the enterprise. And it has a plan, it seems, to remain an essential and integrated component of an increasingly shared CX environment.
Buyers can take some assurance from these moves. Their purchasing lives are still going to be complex and unique, spanning a multitude of complementary Salesforce and partner products. But enterprises that are savvy about their own goals for agentic automation should take some comfort from the fact that Salesforce can articulate a more solid framework for deploying agentic applications that may not be perfectly clear but does provide directional understanding.
Regards,
Keith Dawson