Dreamforce 2026 Has One Job: Prove Agentforce Actually Works

Dreamforce 2026 feels like another big moment for Agentforce. Salesforce has spent close to two years putting its flagship AI product at the center of basically everything it does, and by now, most people in the ecosystem should have a pretty good idea of what it’s supposed to do.

This year has seen the conversation change from “what can Agentforce do?” to “where is the real-world value?” Salesforce can and has pointed to growing adoption numbers and billions of agentic interactions, but there still seems to be a missing use case, or perhaps a missing winning customer story, that really moves the needle for most Salesforce customers.

And while adoption is reportedly growing, there’s undoubtedly some fatigue creeping in too across the ecosystem. After two years of launches, demos, name changes, and AI-heavy keynotes, another perfectly polished Agentforce demo probably isn’t going to win over any skeptics. So with Salesforce’s Super Bowl right around the corner, it’s time to prove that Agentforce can really work at scale.

Where Are We Actually at With Agentforce?

How many of you remember how ambitious that initial Agentforce pitch was? When Salesforce unveiled the product in September 2024, company CEO and co-founder Marc Benioff talked about a new era of autonomous agents, with the rather bold target of empowering one billion agents by the end of 2025.

As regular readers might know, a lot has happened since. Agentforce has gone through numerous iterations, Salesforce has built out the tooling needed to deploy and manage agents, and the commercial numbers have become difficult to dismiss (even if it’s hard to visualize them). Earlier this year, I argued that Agentforce had moved well beyond pure experimentation, with more customers putting agents into production and existing customers buying additional capacity.

There’s no doubt that Agentforce has grown up, but has it really delivered on some of its original promises? Timo Kovala, Lead Architect for Marketing and AI at Capgemini, thinks the answer is rather nuanced and complicated. The underlying product itself, he told SF Ben, is actually pretty solid.

“If you look at the platform itself, it’s not a bad product,” Timo explained. “It’s genuinely, I would say, a decent product for what you would use it for, which is grounding autonomous AI agents in your CRM data.”

The bigger problem, however, might be that Salesforce’s promises to customers have continually evolved: “The promise itself has evolved and switched over time…. If the point of reference keeps shifting, that’s causing this bit of confusion that we’re seeing across social media, between partners, between customers.”

Agentforce may be a lot more capable than it was a couple of years ago, but Salesforce has also described it in so many different ways along the journey that working out exactly what success looks like has become very difficult.

This is where some of the Agentforce fatigue we’ve seen recently starts to make sense. Timo believes that the fatigue is very real. But importantly, it isn’t just a Salesforce-only problem.

He said: “We have fatigue across the board, and it’s not only within the Salesforce ecosystem. You can see it everywhere.”

Alongside this, there’s another outstanding frustration for many Salesforce customers. While almost everything around the platform seems to have become about AI, many of the issues businesses were dealing with before Agentforce arrived haven’t just magically disappeared. 

As Timo put it: “Their actual business challenges – things like disconnected processes, misalignment, siloed ways of working, technical debt – these sort of fundamental issues haven’t really changed all that much. Agentic AI is either not addressing these to its full potential, or it’s disregarding these very real challenges.”

So while people aren’t blatantly rejecting Agentforce, there is only so long Salesforce can keep talking about what agents could do before customers start asking what they are actually doing today.

Salesforce’s AI Strategy Is Now Much Bigger Than Just Agentforce

Adding to the ongoing confusion is the fact that Salesforce’s AI strategy now extends way beyond Agentforce itself.

Over the past year, Salesforce has grown much closer to Anthropic, bringing Claude into Agentforce while also allowing Claude to access Salesforce data and trigger actions across the platform. At TDX earlier this year, Salesforce took this further with Headless 360, effectively opening up its data, workflows, and business logic to whoever AI tools customers want to use.

In simple terms, Salesforce no longer really cares where you interact with its platform. An employee could work through Claude, ChatGPT, Slack, or another interface, while Salesforce just provides the data, permissions, and workflows underneath.

It’s an interesting strategy, but it leaves many asking where Agentforce exactly fits into that. When I asked Timo, he believed that Salesforce needs to become much clearer about the different forms of agentic AI it is now talking about.

“You never know what level of agentic AI Salesforce is referring to when they’re talking about Headless 360 and when they’re talking about agentic AI in general,” he said.

For Timo, one particularly interesting opportunity lies in what he describes as “headless agents”. Rather than users talking directly to Agentforce, an existing Flow, Apex class, or Lightning Web Component could quietly invoke an agent when reasoning is required, before continuing with the wider process. 

Timo said: “For Agentforce, I would definitely see much more potential in what I call these kinds of headless agents, where you have, let’s say, a Flow invoking an AI agent or an LWC or an Apex class that invokes an AI agent that uses its reasoning to determine next courses of action.”

This arguably gives Agentforce a much clearer purpose. Claude or another assistant can be where employees actually work, while Agentforce becomes part of the underlying machinery connecting AI reasoning with Salesforce data and processes.

But, as mentioned, Salesforce needs to explain that clearly. Timo would like to see Dreamforce distinguish between developers using AI to build, admins and architects using it to understand their orgs, autonomous customer-facing agents, and agents invoked inside existing workflows.

Otherwise, Salesforce may continue to have an increasingly capable AI platform that customers struggle to understand.

Dreamforce Needs Real Customers, Not Another Perfect Demo

So, this leaves us with the big questions – what does Salesforce need to successfully showcase at Dreamforce?

I recently prompted people on LinkedIn to share what they wanted from this year’s event, as well as what they realistically expected Salesforce to deliver. There was a pretty clear theme across all responses, and it was that people want less of the highlight reel and more of what Agentforce actually looks like in the real world. 

That means production deployments, measurable ROI, implementation costs, governance, human handoffs, and perhaps most importantly, what happens when things go wrong. 

Several people specifically wanted to hear about failed implementations and the work required to turn them around, rather than another polished customer success story.

There was also a noticeable gap between what people wanted and what they expected. While people wanted more honesty and practical details, the expectation was that the main keynote would remain heavy on vision and adoption numbers, with the genuinely useful information left for breakout sessions and conversations around Dreamforce. But this could be the year that Salesforce changes that.

Timo believed that a part of the problem with getting Agentforce into production is the expectations that agents should behave like traditional Salesforce automation. With Flow or Apex, extensive testing can give you reasonable confidence that the same input will continue producing the same outcome. But really, gen AI doesn’t work like that.

“Instead of planning or controlling risks in advance, you should move the focus to planning for contingencies,” Timo said, adding that businesses need to assume an agent “will not always behave as you would expect.”

Alongside that, it’s worth asking whether those customers will even have much of a choice. As Salesforce builds an agentic layer into its products, frameworks, and vision for the platform, opting out of AI altogether may not become an option – whether every customer is ready for it or not.