Without question, Dreamforce 2025 marked a pivotal moment for Salesforce. The company brought together 50,000 people to learn about the why and the how of creating an agentic enterprise. Yet, just as there are barriers to becoming an agentic enterprise, there were barriers to learning about it, cognitive and physical. This year’s Dreamforce featured airport-level security at every turn — my bag had to be X-rayed and checked five separate times just to reach my destination.
That said, the issue is not whether agentic AI is the right direction forward for Salesforce and other enterprise software companies; it is whether it will deliver revenue fast enough to satisfy Salesforce and its investors. Unlike sales automation, which required only a web browser to subscribe and get started, agentic AI demands something that has eluded most enterprises to date: a modicum of process and data maturity.
In a conversation last summer with Salesforce Tableau’s AI leader, I learned that while the promise of agentic AI is profound, most Salesforce customers lack the foundational elements needed to transform, to wit:
-
Defined processes and scripts for automation.
Research reported in August by Dresner Advisory Services found that 6.5% of organizations are running agentic AI in production, and these early adopters share a consistent profile. They report 100% past success in business intelligence. Seventy-five percent of respondents were early adopters of machine learning, and 62.5% have a designated data leader. Only 32% of organizations reported being completely successful with their BI implementations. Salesforce leaders did not flinch when I suggested that the aggregate number of organizations that are agentic AI-ready was 20%, and this is why only 10.5% in our research said that they are proponents and actively adopting.
Against this background, my goal for Dreamforce was to understand not only the promise of agentic AI but also what Salesforce is doing to help less advanced customers advance. The short answer? Not much.
Agentic Business Impact: ‘Extra Arms and Legs’
Parker Harris, CTO at Salesforce, declared in a pre-briefing that all companies will become agentic enterprises. He went further, forecasting that by 2029, AI will manage 40% of the work at Fortune 100 companies.
Salesforce CEO Marc Benioff, in his keynote this year, steered away from labor arbitrage, highlighting instead the potential of agentic AI to bring humans and agents together to unlock new opportunities for companies. He described Agentforce 360 as providing the “extra arms and legs” that will deliver the ultimate customer experience.
How and when Dreamforce attendees will realize this ultimate transformation is the big unknown.
Apparently, a few weeks before the event, Benioff completely reframed this year’s Dreamforce around Agentforce 360 customer success stories. Interestingly, the customers who presented were all relatively data mature and had implemented Salesforce Data Cloud.
Their stories were impressive. Athina Kanioura, chief strategy and transformation officer at PepsiCo, for example, touted the 23% reduction in labor costs the company reaped by applying agentic AI, duly noting that the greater value will come from business transformation. Laura Alber, CEO of Williams-Sonoma, introduced attendees to Olive, the company’s virtual sous-chef, underscoring that the culinary companion — designed to help customers with cooking questions — went from concept to completion in just 30 days.
Agentic Divide: Crossing the Data Chasm
But what about for the vast majority of Salesforce customers that don’t have their data and cloud ducks in order? What can they do in 30 days with agentic AI? Not much, apparently.
Benioff shared the now-notorious MIT statistic from “The GenAI Divide: State of AI Business in 2025,” which says that 95% of AI projects do not get turned into products. He said companies need many things to succeed: They need governance and data trust. They need quality data. If you don’t have your data right, you are not getting your AI right, he said.
And it’s for this reason that the rate of tech innovation exceeds adoption, Benioff said. He acknowledged that his company needs to bring companies along — and to that end, he is getting personal coaching from high-tech adoption guru Geoffrey A. Moore, author of Crossing the Chasm.
In the meantime, based on Moore’s ideas on bridging the gap between early adopters and the mainstream market, Benioff is pursuing a product market fit strategy by assembling the 23% of customers that have data and CX maturity. He is specifically looking for firms that have productized their data and are looking to create agentic enterprises — firms that want to be changed by Salesforce products.
And for those who are not there yet? Brent Hayward, head of competitive intelligence at Salesforce, was candid that there are critical issues that need to be solved — first and foremost, the mess of data that is the reality for many companies. “Throwing bad data at AI problems doesn’t work,” said Hayward, formerly CEO at MuleSoft. Customers may very well be excited about agentic AI, but none of them will get value from it with bad, untrustworthy, ungoverned data.
There are advancements that will make the journey to data maturity faster, such as zero-copy data and low-code/no-code business processes. But building a platform with integration, automation, clean data, and governance is hard enough without then having to orchestrate and manage agents.
When I asked about how companies can close the agentic AI gap other than by launching a massive data project, I got several boilerplate answers from Salesforce leaders, including limiting the scope of domains. A more candid answer came from Derek Santana, PwC’s U.S. Salesforce Alliance leader at PwC.
Acknowledging the readiness gap on implementing agentic AI, Santana explained that PwC has two operating models for clients — use case-specific or end-to-end processes. The latter is about transformation and high-value processes such as end-to-end customer experience, and this is where the value lies.
For CIOs with low data maturity, virtualization and being part of an open ecosystem can help. Hayward said Salesforce MuleSoft is building an agentic fabric and implied this will eventually include some form of agent orchestration and management function.
Parting Thoughts
Dreamforce 2025 showcased what’s possible when data, process, and purpose align. But for most enterprises, this remains aspirational.
Salesforce may have built a bridge to the agentic future — but many customers are still miles from the on-ramp. What stops these organizations is organizational immaturity. This will inevitably limit the size of the market adopter market and how fast Salesforce will move through the technology adoption lifecycle. While I am a big fan of Geoffrey Moore, nothing is going to fix the data and architecture immediately for organizations that failed to invest.
