Salesforce’s Dreamforce event was an event of epic proportions. Pre-pandemic attendance peaked at 170,000 about 10 years ago. This year’s event garnered nearly as much attention with 50,000 attendees and 122,000 online viewers. The event took over the entire Moscone Convention Center, including the North, South and West pavilions, as well as surrounding venues. If the agentic future that Salesforce envisions becomes a reality, perhaps we will send our agents to attend in the future.
The event extended and refined the agentic platform story it began at its developer conference, TrailblazerDX (TDX), earlier this year. As I wrote in my coverage of that event, Salesforce is aiming to provide a comprehensive foundation for agentic AI across the enterprise. The centerpiece of Dreamforce was its agentic enterprise architecture, Headless 360, consisting of four layers:
AIforce extends Salesforce capabilities to any work surface, including Claude, Slack, Microsoft Teams (soon), Tableau and Salesforce itself. Agentforce provides configurable and prebuilt agents. Customer 360 supplies applications, workflows and semantic definitions. Data 360 serves as the data management foundation, combining Salesforce data with other data along with the data engineering capabilities acquired through Informatica to collect, combine and govern the data.
The shift from Headless 360 at TDX to AIforce at Dreamforce is a refinement in its positioning. The Headless toolkit remains part of the architecture, providing APIs, MCPs, plug-ins, skills and developer tools. However, the message is now easier to understand: Salesforce wants customers and partners to build and use experiences on top of its platform without requiring every user to enter a Salesforce application. There may also have been concerns that “Headless agents” was a bit of a paradox since brains (and therefore heads) are an important part of intelligent execution.
Claudeforce brings Salesforce into Claude and Cowork through a prebuilt connection and is initially focused on sales skills such as pipeline review, forecasting, account planning and deal strategy. Slackforce extends Slackbot and Slack CRM so employees can ask questions, update records, create follow-up tasks and build interactive experiences without ever leaving Slack. Salesforce also discussed integrations with Microsoft Teams and other conversational surfaces. The vision is one in which users can ask for an outcome rather than navigate screens and records, while Salesforce continues to enforce the underlying rules.
With these changes, Salesforce is attempting to thwart the notion of “SaaS-pocalypse,” suggesting that the real value of the applications lies in the workflows, governance and
Beyond its “any surface” vision, Salesforce, in conjunction with NVIDIA, introduced Koa, a CRM reasoning model, built on NVIDIA Nemotron and trained with the knowledge gathered over nearly three decades of CRM experience. Koa is designed for multi-step tasks such as updating opportunities, routing cases and scheduling follow-ups. Salesforce says it controls the model weights and runs the model within its own trust boundary, giving customers a specialized option for Agentforce workloads.
Koa does not replace Salesforce’s broader model ecosystem. Instead, it gives Salesforce a way to encode CRM-specific reasoning and to control more of the inference path. That is important for customers that care about predictable actions, data residency and operational control, although the model remains in pilot with general availability expected later in 2026.
It’s worth noting that Jensen Huang, CEO of NVIDIA, Dario Amodei, CEO of Anthropic, and Sam Altman, CEO of OpenAI, all attended and participated in the event. Their participation highlights the unique position that Salesforce occupies in the AI market. It is one of the most significant players in the AI market that does not compete with the core business of any of these other key players. In fact, Salesforce acts as a delivery channel for the value that NVIDIA, OpenAI and Anthropic seek to provide to their enterprise customers.
Salesforce is eating its own dog food too. They are expanding Agentforce with purpose-built agents for sales, service, commerce, employee experience and back-office work. The company highlighted agents such as Hunter for outbound prospecting, Piper for inbound lead qualification and Casey for customer service and escalation. Salesforce’s own examples included large-scale service conversations, real-time lead qualification and employee productivity, while customer examples included Pirelli and Heathrow. The agentic focus is also becoming more ambitious, with agents that pursue goals over days or weeks, learn skills, work with other agents and improve over time. That is a far cry from the single-task assistant model and a turn toward a growing digital workforce. Our research suggests that this digital workforce will rarely be turned loose to operate on its own, with 87% of AI-enabled work involving humans to some extent through the end of 2027.
Dreamforce 2026 was a refinement of Salesforce’s agentic strategy, but it was also a significant change in how the company describes its platform. TDX established the need to manage the full agent development lifecycle. Dreamforce added the interface layer and made the case that Salesforce can provide the trusted context and business execution behind agents wherever employees work. I recommend that as enterprises develop and expand their agentic AI investments, they consider the capabilities that Salesforce provides.
Regards,
David Menninger