The five announcements from Dreamforce 2026 worth your attention, translated from keynote language into what actually changes for the people using Salesforce every day.
Years of work go into a CRM. The data model, the processes, the business logic that finally matches how the company actually operates. And then a large part of the organization never opens it. Sales reps live in their inbox, service teams in their queue, leadership in slide decks. The value sits in Salesforce, while the people sit somewhere else.
Dreamforce 2026, held September 15 to 17 in San Francisco, took that gap head on. Across more than a dozen announcements, one idea kept coming back: people no longer need to go to Salesforce to get work done. Salesforce now goes to them.
Below are the five announcements that matter most for organizations in Belgium and the Netherlands, and what each one means in practice.
The headline of the week was AIforce, a new interface layer that sits above Data 360, Customer 360, and Agentforce. It brings Salesforce data, workflows, and permissions into the tools people already use. AIforce launches with three entry points:
Every request runs on existing permissions, so an assistant only sees what the person asking can see. Business data is not retained by the model provider.
Adoption no longer depends on people opening an app. The colleague who never touched a dashboard can now ask a question in Slack and get an answer grounded in real CRM data. That is a real opportunity for organizations that have struggled to get full value from their Salesforce investment.
It also raises the stakes on the basics. Your permission model becomes the security boundary for every new AI surface. A useful question to ask today: if an AI assistant could see exactly what each employee can see right now, would you be comfortable with that?
Large language models are smart, but they know nothing about your customers, your products, or how your business runs. Salesforce’s answer is the Trusted Enterprise AI Harness, an architecture that brings six capabilities together into one AI Control Plane:
As models become interchangeable, the real differentiator is your own data, definitions, and governance. Two companies using the same AI model will get very different results depending on how trustworthy and well-structured their data is.
For most organizations, this makes data foundation work the smartest investment of the coming year. Agreeing on shared business definitions, connecting siloed systems, and cleaning up customer data may not sound as exciting as a new agent, but it is exactly what makes every agent that follows more reliable.
Salesforce introduced a portfolio of seven prebuilt agents for the most common use cases, designed so business users can configure them without long IT projects:
This is an honest shift in approach. Building agents from scratch turned out to be hard for many teams, and Salesforce acknowledged as much. Starting from a proven agent lowers the barrier considerably, and shortens the road from idea to first results.
Siemens offers a good illustration. With 2,800 unqualified inbound leads arriving every week across seven business units, sellers struggled to know where to focus. A coordinated set of agents now engages every lead, gathers missing information, and routes it with full context. The agent did the heavy lifting, but the results came from clear qualification rules and a well-designed handoff to people. Job-ready agents still need a well-designed job.
Together with NVIDIA, Salesforce launched Koa, its first reasoning model built specifically for CRM work. Salesforce reports that Koa matches or exceeds leading models on CRM tasks with three times fewer errors. Koa is currently in pilot with organizations such as Formula 1 and UChicago Medicine.
Alongside Koa, the model menu keeps growing: Google Gemini now powers parts of the Agentforce reasoning engine, and Agentforce customers can use the full range of frontier models on Amazon Bedrock, including Anthropic and NVIDIA models.
Choosing an AI model is becoming a configuration decision, not a strategic lock-in. Organizations no longer need to bet their roadmap on a single AI vendor. The better question is which model fits which task, at what cost, and under which governance. For European organizations, it is worth keeping an eye on regional availability as Koa moves beyond its pilot phase.
Salesforce deepened its partnerships with both major cloud providers:
Agents are starting to work together across platforms, much like colleagues across departments. Your Salesforce environment is no longer an island; it is a hub in a wider network of agents and systems. That makes integration architecture and clear governance across platforms more important than ever, especially in organizations running a mix of Microsoft, Google, and AWS technology.
Readers who follow Salesforce closely will notice that the familiar names are back. On Salesforce’s own website, Sales Cloud, Service Cloud, Marketing Cloud, and Commerce Cloud once again appear under their original names, while Agentforce refers to the agent platform itself.
Look past the product names and the pattern is hard to miss. For two decades, the hard part of any CRM program was getting people into the system: the training, the adoption dashboards, the gentle reminders to log the call. This year Salesforce stopped fighting that battle and started bringing the system to the people instead.
That changes what is scarce. When anyone can ask a question in plain language and get an answer drawn from live CRM data, the bottleneck is no longer access or training. It becomes whether the answer is right. Whether the record behind it is current. Whether the permissions around it hold. Whether everyone in the building means the same thing by pipeline. The quality of the foundation stops being an internal concern and becomes something customers feel directly.
The pace is not theoretical. According to the Salesforce Agentic Enterprise Index, the average number of agents activated per organization grew nearly 3x in a single year, and service agents now resolve 7 out of 10 customer conversations without human help. Whatever pace an organization sets for itself, its customers are already forming expectations somewhere else.
Which is why the most valuable work of the coming year will probably look unglamorous next to a keynote. Agreeing on what the numbers mean. Cleaning up records nobody has owned in years. Revisiting a permission model written for a time when only a few dozen people ever logged in. None of that makes a good demo. All of it decides whether the demo holds up once it meets a real business.
These are the questions specialists across the Spire ecosystem work through every day, from Life Sciences to Manufacturing and Financial Services, and the cases show what that looks like in practice.
Get in touch with the Spire team for a conversation about where the biggest opportunities are for your organization.
Agentforce is the digital workforce: the agents themselves, ready to deploy or build. AIforce is the layer that carries Salesforce data, workflows, business logic and permissions out to wherever people already work, whether that is Claude, Slack or the Lightning interface. Put simply, Agentforce is who does the work, and AIforce is where the work can now happen.
Yes, in beta for all customers. It launches with Salesforce in Claude, a prebuilt connection that includes 37 sales skills covering prospecting through pipeline hygiene, and it was piloted by companies including Deloitte, GitLab and Legora. Each person signs in with their own credentials and sees only what their Salesforce permissions allow, with no data retained by the model provider.
Koa is Salesforce’s first reasoning model built specifically for CRM work, developed with NVIDIA on the Nemotron family and trained without customer data. Salesforce reports that it matches or exceeds leading models on CRM actions with three times fewer errors. It is in pilot now, with general availability expected in winter 2026 starting in US regions, so European organizations should plan around a later date.
In practice, yes. Sales Cloud, Service Cloud, Marketing Cloud and Commerce Cloud appear under their original names again on Salesforce’s own website, and Agentforce now refers to the agent platform rather than the applications. There has been no formal announcement, so treat this as current practice rather than a published decision.
Salesforce describes AIforce as working with what you already have: no migration, no new permissions model and no custom integration work, with admins connecting once so teams get access on day one. Agentforce Coworker in particular can be activated from inside Salesforce with a single button. Pricing and packaging were not detailed at launch, so that part is worth confirming with your account team.
Dreamforce 2026 preview: the agentic enterprise, Data 360, governance and adoption. What Spire is watching, and why it matters.
An AI agent pilot rarely stalls on technology. Discover five signs that yours has stalled, and what each one tells you to fix first
AI agent pilots are easy to start. Getting an AI agent into production is a different kind of work, and it’s where most initiatives slow down.