Salesforce Dreamforce 2026: The CRM Becomes Something AI Can Operate Through

TLDR

• HubSpot UNBOUND and Salesforce Dreamforce ran in the same week, and together they show one shift. HubSpot made the CRM better at understanding the business. Salesforce is making it something AI can act through.
• AIforce is the headline. Salesforce describes it as a live interface layer that brings its data, workflows and permissions to wherever people and agents already work, launching with Claudeforce, Slackforce and Agentforce Coworker.
• Claudeforce puts Salesforce inside Claude as a plugin with 37 prebuilt sales skills, so a seller can review pipeline and update records without opening Salesforce.
• The numbers behind the story matter more than the demos. Salesforce forecasts that 20% of holiday ecommerce traffic will come from AI chat agents, and told analysts it expects over $63 billion in revenue by fiscal 2030.
• For marketing and growth teams, the takeaway is that AI is now shaping both ends of the funnel: how buyers discover us, and how our own teams act on what we know about them.

In my previous article on the HubSpot UNBOUND 2026 announcements and market direction, I argued that the real story wasn't the AI features. It was context. The CRM was becoming a system that understands how the business works, rather than a system we update.

Salesforce ran Dreamforce in the same week, and the picture got a lot clearer.

Because Salesforce isn't mainly talking about AI helping people work inside the CRM. It's building toward AI operating across the CRM, the workflows, the business logic and the permissions that sit around all of it.

HubSpot showed AI moving closer to the buyer and the marketer. Salesforce is showing AI moving deeper into the company itself. Put those together and the direction of travel for marketing and growth teams gets much easier to see.

I've been doing this for close to two decades, and I've sat through plenty of conference keynotes that promised a new era and delivered a new dashboard. Having both of these land in the same week made the pattern hard to miss, and Dreamforce is the reason I'm now fairly convinced.

What actually got announced

Salesforce shipped a lot across Dreamforce week. The individual features aren't the interesting part, same as with UNBOUND. What matters is the pattern underneath them.

AIforce was the centerpiece on keynote day. Salesforce describes it as a live interface layer that brings the full power of the platform to wherever people and agents work, launching with Claudeforce, Slackforce and Agentforce Coworker (Syncon AI roundup). The framing in the announcement is that people don't have to come to Salesforce anymore, because Salesforce comes to them in Claude, Slack, Lightning or wherever they prefer to work (Salesforce Ben).

Benioff put it on stage as a generational shift in interfaces, moving from command line to graphical to web and mobile, and now into what he called the AI interface era (ITPro live blog). Salesforce also went further and said the AI is going to replace the UI, which is a strong claim from a company whose interface a generation of ops people have built careers around.

Claudeforce is the Salesforce and Anthropic partnership that AIforce launched on. It runs in both directions. Claude becomes the reasoning model across Agentforce and Slack, and Salesforce becomes a plugin inside Claude with 37 prebuilt sales skills (TNW), covering things like deal health checks, meeting prep and pipeline updates without switching to the Salesforce app (Reworked).

Koa is Salesforce's first CRM reasoning model, post-trained on NVIDIA's Nemotron 3 Super, and notably, no customer data was used to train it (Syncon AI).

There was also an expanded Agentforce agent portfolio, plus partnerships with both Google Cloud and AWS announced on the same day, with Agentforce and Gemini Enterprise connected over MCP.

That last detail is easy to skip past. Two hyperscaler partnerships on one day, both built on MCP, tells you the plumbing standard has stopped being a developer curiosity.

The CRM becomes a place AI works through, not a place we work in

This is the part I keep thinking about.

In that earlier piece on UNBOUND, I wrote that the CRM may be becoming less of a system marketers update and more of a system that understands how the business works. Salesforce is taking that further, to a system that AI operates through.

Salesforce frames AIforce as connecting probabilistic AI models with deterministic enterprise data, workflows, permissions and business rules (Daily Political). Strip the vendor language away and the idea is straightforward. The model does the reasoning. The CRM supplies the facts, the rules and the boundaries on what can actually be done.

That last part matters more than it sounds for anyone who has watched an AI tool confidently do something wrong. Your permission model effectively becomes the security boundary for every new AI surface you open up, which is a very different conversation than picking a content tool.

Marketing has spent years asking for the CRM to be the single source of truth. This is what it looks like when that request is taken seriously and then handed to something that can act on it.

Claudeforce moves the work out of the CRM entirely

Here's the part I'd pay closest attention to.

With Salesforce in Claude, a seller can reason over live revenue context, automate pipeline updates and take governed action from inside Claude (Salesforce press release).

Think about what that does to the interface question. For 20 years, the fight in martech was about which screen people live in. CRM vendors built more surface area so you'd stay inside their product. Salesforce just said the opposite. Come to us or don't, as long as the context and the actions route through us.

The scale of the commitment behind it is worth knowing too. Salesforce plans to invest roughly $300 million in Anthropic tokens over 2026, on top of an existing equity stake valued at about $5 billion (Crypto Briefing). Companies don't put numbers like that behind a side experiment.

For marketing teams, this is where I'd resist the urge to file it under "sales stuff."

If sellers start working through an AI layer that reads live CRM context, everything we hand to sales gets re-read by a machine. Lead scores, campaign attribution, intent signals, lifecycle stages, the notes we write on a nurture track. A rep might have skimmed those. An AI reading the full record will not skip them, and it will happily act on whatever it finds.

Which means the quality of the marketing data we push into the CRM stops being a reporting problem and becomes a pipeline problem.

From assistant to agent to something that runs for days

When I wrote about UNBOUND, I described the shift from AI that helps you find information inside the CRM to AI that takes a business objective and turns it into work. Salesforce is showing the enterprise-scale version of that same move, and it has usage data to back it.

Its Agentic Enterprise Index, based on Agentforce usage from February 2025 to April 2026, reports that organizations increased activated agents by nearly 3x over the fiscal year while cutting average agent creation time by 53%.

The number I find more telling is about capability. The average number of unique skills each agent could act on rose from about two at the start of 2025 to six by the end of the year, and during peak shopping season the average retail agent expanded its skill set by 350%.

Salesforce also introduced a metric it calls the Agentic Work Unit, meaning one discrete task an agent completes. By April 2026, Agentforce agents had produced 734 million of them, with output growing at a 15% compound monthly rate (Advanz101 summary).

Look at the most common agent actions in that index too: querying records, drafting and scheduling emails, summarizing records, and generating briefs and creating campaigns (Salesforce).

Campaign briefs are already on the list. That's not a roadmap item.

The forecasts and research worth taking back to your team

This is the part I'd actually pull into a planning deck. A few of these were published in the weeks around Dreamforce rather than announced on stage, but they're the data underneath everything Salesforce said.

The company forecast. At its analyst session during Dreamforce, Salesforce issued a fiscal 2030 revenue target of over $63 billion, beating analyst estimates (CNBC). Read that as the company pricing in that the agentic bet works.

The forecast marketers should care about more. Salesforce's 2026 Holiday Predictions puts numbers on the discovery shift I wrote about after UNBOUND:

• Consumer reliance on AI assistants as the first stop in the shopping journey grew 200% in a single year, from May 2025 to May 2026.
• Half of shoppers now report using an AI assistant at some point in their buying journey, a 67% year-over-year increase, and 74% say they trust the product recommendations they get from AI chat.
• Salesforce predicts 20% of all 2026 holiday ecommerce traffic will originate from AI chat agents, a mix of consumer-facing bots, autonomous agents doing backend work, and competitor scrapers feeding price matching.
• Year-over-year engagement on brand-owned properties dropped 7%, while traditional search engines and marketplaces each fell 15% as shopping aids and newer channels like AI assistants and social surged 38%.
• One in three ecommerce sites is predicted to have a site-specific shopper agent live by Cyber Week 2026.

Underneath that, the Shopping Index shows global digital traffic up 18% year on year in Q2 2026 while order volume grew just 1%, with cart abandonment at 82% (Communicate).

That traffic-up, orders-flat gap is the whole story in one line. More visits, less intent per visit, because the evaluation is happening somewhere else before people arrive.

I know most of my readers are B2B, not retail. Use retail as the leading indicator it usually is. The behavior shows up in consumer buying first and lands in B2B buying committees a year or two later.

What the research says about our own adoption. From Salesforce's State of Sales 2026, based on a survey of more than 4,000 sales professionals: 87% of sales organizations already use some form of AI, 54% of sellers have used agents with nearly 9 in 10 planning to by 2027, top performers are 1.7x more likely to use agents than struggling teams, and sellers expect agents to cut prospect research time by 34% and email drafting by 36%.

The State of Marketing 2026 numbers are the ones I'd put in front of a CMO: 78% of marketers say they need more personalized content than they can produce and 75% are turning to AI to close that gap, but 98% hit barriers to personalization, with data issues the most common culprit. There's also a detail I find genuinely interesting: 75% of marketers using AI are satisfied with their ability to connect touchpoints, against 60% of those without it, though it isn't clear whether agents improved connectivity or deploying them simply forced teams to unify data first.

Honestly, I suspect it's the second one. Which is an argument for doing the data work either way.

And the counterweight. Before this reads like a straight adoption curve, IBM's State of Salesforce 2025-26 report found that only 33% of AI initiatives are currently meeting expected ROI, 62% of organizations are concerned about unpredictable AI-related costs, and just 21% strongly agree they have the governance structures to manage agentic AI responsibly (2-Data analysis).

Also worth saying plainly: most of the adoption numbers above are Salesforce surveying its own market. Coverage going into Dreamforce noted that public ROI figures and cost-to-run numbers for Agentforce are still largely missing (MarketScale). Take the direction seriously. Build your own baseline before you take the magnitude seriously.

The main stage told a second story worth hearing

Benioff brought Dario Amodei, Jensen Huang and Sam Altman to the stage in front of roughly 12,000 people at Moscone, and they didn't agree with each other. For once, that was more interesting than the product demos.

Amodei argued for pacing the frontier. Days earlier he had published an essay urging the industry to slow the pace of model development, and he repeated the case on stage, framing it as a way to lead by example (CNBC via Jingletree). He pointed to the automotive industry as a model, where a safety incident at one company prompts responsible competitors to improve and sometimes to organize international standards (InformationWeek).

Huang pushed back hard. He said market forces are already sufficient and no new laws or regulations are needed, calling the speed versus safety tradeoff a false choice, and framing safety as an engineering challenge rather than a legislative one (Quartz).

Altman sat somewhere in between. In his afternoon conversation with Benioff, he said the public is right to fear that a few AI companies could gain too much power, and that no company should make its safety depend on what rivals do (TNW).

Benioff's own framing was blunt. He said the industry cannot let AI become social media 2.0, and argued that companies and executives have to take absolute responsibility for what they build (Yahoo Finance).

Why does any of this matter to a growth team?

Because the pace of this debate sets the pace of your roadmap. If the model companies converge on slower releases and tighter standards, agentic features arrive more gradually and with more governance attached. If Huang's view wins out, capability keeps shipping fast and the responsibility for controlling it lands on your team.

Altman made one point I'd write on a whiteboard for any lean marketing team. He said there are probably more advantages to being a smaller company than ever, because of how fast you can move and adopt new technology (transcript). That's a useful counterweight to the assumption that enterprise AI is a game only enterprises get to play.

He also warned that companies need to defend against a coming wave of AI-enabled cyberattacks, regardless of whose models they use. Marketing owns a surprising amount of surface area there, from web properties to email platforms to the data we sync between them.

One honest gap for marketers

I'll say the quiet part here, because my audience is marketing and growth teams rather than sales ops.

Most of what Salesforce showcased leans toward sales and service. The 37 prebuilt skills in Claudeforce are sales skills. The flagship customer stories are service ones, like Live Nation's venue agent answering fan questions across 120 venue websites with the potential to automate over 300,000 inquiries a year (Salesforce newsroom).

Marketing sits downstream of this release rather than at the center of it. If you're running demand gen on Salesforce, your practical near-term wins are more likely to come from what happens to your leads once they're handed over, not from new campaign tooling.

I don't think that's a permanent state. But it's worth setting expectations internally before someone promises the CMO an agentic marketing team by Q1.

Marketing is being squeezed from both ends

This is the thing I'd want a growth team to take away from the week.

At the top of the funnel, AI is becoming the interface through which buyers discover and evaluate companies. That was the ChatGPT Ads and AEO story out of HubSpot, and it's what the 20% agentic traffic forecast is really describing.

Inside the funnel, AI is becoming the interface through which our own teams understand and act on customer context. That's the Dreamforce story.

The funnel most of us learned looks like search, click, website, form, lead, sales, customer. Every step was something we could see, instrument and optimize.

What's replacing it looks more like AI discovery, AI evaluation, human or AI interaction, AI-assisted decision, agentic execution, then a human relationship somewhere in the middle of all of it.

Both ends of that chain are now mediated by systems we don't own. The difference is that we can influence one end with evidence and positioning, and we control the other end with data quality and governance. Most teams I talk to are underinvested in both.

How I'd act on this

Some of this carries over from what I wrote after UNBOUND, but Dreamforce adds a few of its own.

Assume anything in your CRM will be read and acted on, not just reported.
Lead source, lifecycle stages, campaign fields, notes. A rep might have ignored a stale field. An agent won't.

Get a baseline before the agents arrive.
Given that only a third of AI initiatives are hitting expected ROI, write down your current cost per opportunity, response times and content output now. Otherwise you'll have no way to prove what changed.

Decide what agents may do before you decide what they can do.
Permissions are becoming the real control layer. Sort out who approves what, which actions need a human sign-off and what happens when an agent gets it wrong.

Start with narrow, repeatable work.
Campaign briefs, list cleanup, follow-up routing, meeting prep. The index data suggests agents mature by picking up more skills over time, so give yours a small, well-defined job first.

Treat non-human traffic as a real segment.
If a fifth of holiday ecommerce traffic is projected to be agents and crawlers, your analytics, your bot filtering and your conversion benchmarks all need revisiting. Falling engagement on your own properties may not mean falling interest.

Ask for cost per outcome, not per seat.
Usage-based agent work means someone needs to own the unit economics. With most teams worried about unpredictable AI costs, find out now what happens to the bill if agent activity triples.

Keep strategy human, and make it clearer than it is today.
Agents executing against a fuzzy ICP or inconsistent positioning will scale the confusion. AI execution raises the cost of vague strategy.

What I took away

In my previous article on UNBOUND, I ended with the idea that the advantage may not go to whoever has the fanciest AI, but to whoever gives their AI the best context.

Dreamforce made me want to add a second half to that.

HubSpot's announcements made me think about AI that increasingly understands the business. Salesforce's announcements make me think about AI that increasingly acts across it. Put together, they point at something bigger than another generation of marketing automation.

The questions worth asking in your next planning meeting aren't about how much content AI can produce, or even how many tasks it can automate. They're about what your AI actually knows about your business and your customers, which decisions it's allowed to make, which actions it's allowed to take, and where a human should stay firmly in control.

Context is the foundation. Action is where the advantage starts to compound.

Related questions

What is AIforce?
Salesforce describes AIforce as a live interface layer that brings its data, workflows, permissions and business logic to wherever people and agents work. It launched at Dreamforce 2026 with Claudeforce, Slackforce and Agentforce Coworker.

What is Claudeforce?
It's the expanded Salesforce and Anthropic partnership. Claude becomes the reasoning model across Agentforce and Slack, and Salesforce becomes a plugin inside Claude with 37 prebuilt sales skills for pipeline updates, meeting prep and deal reviews.

What is Koa?
Koa is Salesforce's first CRM reasoning model, post-trained on NVIDIA's Nemotron 3 Super. Salesforce says no customer data was used to train it.

What did Salesforce forecast at Dreamforce 2026?
At its analyst session, Salesforce issued a fiscal 2030 revenue target of over $63 billion. Separately, its 2026 Holiday Predictions forecast that 20% of holiday ecommerce traffic will originate from AI chat agents and that one in three ecommerce sites will run a site-specific shopper agent by Cyber Week.

Who spoke at Dreamforce 2026?
Marc Benioff hosted the main stage with Anthropic's Dario Amodei, NVIDIA's Jensen Huang and Siemens CEO Roland Busch, plus a separate conversation with OpenAI's Sam Altman the same day. Travis Kalanick and Bill Nye also appeared during the week.

When were Dreamforce and HubSpot UNBOUND 2026 held?
Dreamforce ran September 15 to 17 in San Francisco and HubSpot UNBOUND ran September 16 to 18 in Boston, so the two events overlapped almost entirely.

How reliable are the Agentforce adoption numbers?
They come from Salesforce's own Agentic Enterprise Index and its State of Sales and Marketing surveys. Useful as trend signals, but not independent research. IBM's State of Salesforce report found only 33% of AI initiatives are meeting expected ROI, which is a helpful reality check.

Does any of this matter if my team runs on HubSpot rather than Salesforce?
Yes, because the direction is the same at both vendors. AI is moving from answering questions about your CRM to acting on it, and the data quality and governance work that enables it applies either way.

Where should a marketing team start?
With CRM data quality and a clear view of which decisions you're willing to delegate. Then pick one narrow, repeatable workflow, baseline it, and measure what the agent actually changes before expanding.