Agentforce is seemingly Salesforce’s number one priority right now. It’s the forefront of every Salesforce event, product launch, announcement, and conversation – so much so that CEO Marc Benioff recently joked that he might even rename the company ‘Agentforce’.
But while adoption figures are rising, seemingly painting a rosy picture for Salesforce’s big AI suite, it’s still worth asking: are Agentforce customers really using it? I’m sure many of us have shared the experience of signing up to be a paying customer to some service or another, forgetting about it, and then only remembering we’re customers months down the line. So, is this what’s happened to Agentforce customers? Or are they squeezing out maximum ROI from it?
Agentforce Usage Figures Imply Concentration
Salesforce’s Q3 financial results, posted in December, revealed that Salesforce had closed 18,500 Agentforce deals, 9,500 of which were paid – with 3.2 trillion tokens processed. Just one month later, on January 12, the company’s CEO, Marc Benioff, posted on X, revealing that Agentforce had hit 11.14 trillion tokens served.
If we take Benioff’s 11.14 trillion tokens figure, and divide it by the latest number of Agentforce deals (18,500), we get a little over 600 million. Salesforce is understood to have over 150,000 customers.
An average rate of 600 million tokens is quite significant, seeming to imply that, yes, Agentforce customers really have embedded it deeply within their workflows. Still, even taking these figures at face value, most Salesforce customers are not using Agentforce. This paints a narrative that those who are using it are really finding value in it, but most customers are still not bothering with it. Let’s take a deeper look at what this might mean.
The 600 million tokens served per customer figure is a very rough guideline – and the reality is likely to be much more highly skewed. A relatively small number of larger enterprises could well be driving a disproportionate share of those 11.14 trillion tokens.
Even if we took the 600 million figure average at face value, that still paints a picture of heavily skewed usage overall. If 18,500 customers out of 150,000 have Agentforce deals, that equates to a figure of around 12%.
But, taking into account that the reality is far more likely to be quite unevenly distributed, even among that 12%, the story of “deep embedding” among a small number of customers becomes even more pronounced.
In most enterprise SaaS models, particularly the world of AI, it’s not uncommon to see a minority of customers (even within that existing minority of 12% of total customers) accounting for the bulk of tokens served.
But it’s still worth bearing in mind the pattern that enterprise AI adoption can sometimes follow:
- An initial stage of experimental pilots, sometimes amid fears of AI products being ‘overhyped’ or an ‘AI bubble’. We see small deal counts, with low usage, in this early phase.
- Early adopters embed the product within their workflows, either reaping the rewards (if the product is good) or suffering the consequences (if the product is bad). Some customers will see very high usage, while others remain skeptical.
- A penetration inflection point, which sees broad operational rollout as the product proves itself to provide return on investment (ROI).
Marc Benioff has said before that Agentforce is Salesforce’s “fastest-growing product ever”, naming several large enterprise customers like Dell, FedEx, Pandora, PepsiCo, and Williams-Sonoma.
The current figures suggest that Salesforce is somewhere between stages two and three of the above pattern, with heavy usage among early adopters, suggesting product-market fit in certain segments.
But it’s still obvious that there’s low overall penetration, meaning it has not yet crossed into mainstream default usage across the Salesforce ecosystem. SF Ben’s recent surveys have seen respondents report Agentforce adoption of between 30-34% among developers, admins, and architects.
Even if most Salesforce customers are not yet on Agentforce deals, the cloud giant can use the 11.14 trillion figure as proof of scalability and AI credibility. With AI rivals Microsoft and ServiceNow battling Salesforce for market leadership in the AI race, demonstrating the ability to properly scale matters a lot.
The figures do not necessarily paint a weakness for Salesforce, even bearing in mind the relatively low percentage of Salesforce customers currently paying for Agentforce. It’s not unusual for emerging enterprise AI tools to see relatively low penetration early on, amid a cautious environment of skepticism.
Tokens are a handy metric to keep track of, but the real story of Agentforce will be told in Salesforce’s coming quarterly results, on February 25, 2026. It’s not simply about generating revenue – though obviously any business needs to do that – Salesforce now needs to keep showing demonstrable Agentforce growth, breaking into double figures for percentage of customers on paid Agentforce deals.
‘Implementation Skills Just Haven’t Filtered Down’
Senior Industry Analyst and Founder of Keenan Vision, Vernon Keenan, told SF Ben:“It’s just speculation on my part, but based on my conversations with Salesforce executives, I’m guessing they have brought on several large customers and activated some major projects, and that would explain the growth.
“Here’s something to consider. Let’s not forget how huge Salesforce has become, with a global footprint including deep penetration in most English-speaking countries plus Japan and others. Once they get a product that people like, it can take off with startup-like acceleration.
“I think we’re seeing that in the numbers they’ve announced for Data 360, and that’s largely driven by these very large customers coming online. I expect this to continue through 2026.”
Vernon said that he agrees with the general observation that Agentforce hasn’t really penetrated most Salesforce customers, and “the few thousand who have adopted it represent a fraction of the 150,000-customer base”.
If Agentforce usage were more widespread, we would see secondary market effects in areas like the observability market with more third-party apps and products, like Wayfound.ai, to support the emergence of the new agentic software development lifecycle, he added.
“This reflects a general problem in the industry,” Vernon said. “The system integrators and consultants generally do not know how to perform the data engineering and setup necessary to support agentic deployments that have enough company and customer context.
“If you have a crack team of experts in multiple cloud architectures and other data lake technologies, plus they have two or three projects under their belt, then you can get agents to work, and you’ll be in that elite group.
“To sum it up, AI agent orchestration technology isn’t the problem; it’s that the implementation skills just haven’t filtered down to the mid-tier and smaller organizations that really need them.”
Salesforce Architect and Strategist at Capgemini, Timo Kovala, said that he has spoken to a few of those “rare power users”. Timo told SF Ben: “They’ve confirmed my hypothesis that the power of Agentforce lies in non-conversational (headless) agents.
“One customer had ten agents in prod. None of them interacted directly with user prompts – screen flows, record-triggered or schedule-triggered flows. They invoked their agents and had the agent do complex reasoning and RAG to act on changing inputs, so I’m thinking the whole product has been positioned and sold wrong.”
Timo, who is the author of A Complete Guide to Agentforce, added: “Go-to-market strategy has focused on flashy demos instead of real customer pain points. Pain points like rigid, rule-based processes that are full of duct tape and break whenever inputs change.”
Agentforce Customers Are Growing Nearly 50% QoQ
SF Ben asked Salesforce about the figures, particularly, that about how Salesforce’s Q3 financial results, posted in December, revealed that Salesforce had closed 18,500 Agentforce deals, 9,500 of which were paid – with 3.2 trillion tokens processed. Then, one month later, Marc Benioff said that Agentforce had hit 11.14 trillion tokens served.
Salesforce stressed to us the following Agentforce adoption figures, as of the end of our Q3 FY26 (ending October 31):
- Agentforce has 18,500 – 9,500 paid – customers after just one year, making it the fastest- growing organic product in Salesforce history.
- The number of Agentforce customers is growing almost 50% QoQ.
- Salesforce’s top 50 customers generated more than 200 million LLM calls – interactions when agents invoke a model to understand context and decide the next action to take – in Q3 alone, and are on track for 2 billion over the next year.
- More than 50% of Agentforce ACV came from existing customers “refilling the tank” (purchasing more credits).
- The number of Agentforce customers in production increased 70% in Q3.
- Agentforce ARR is now $540M (up 330% YoY). When combined with Data 360, ARR is nearly $1.4B.
A Salesforce spokesperson told us: “Agentforce is the fastest-growing organic product in Salesforce history. In just one year, we’ve closed 18,500 deals, with paid customers growing 50% quarter-over-quarter. The token usage isn’t just significant – it’s accelerating. In October alone, usage was up 25% month-over-month.
“In addition to net-new logos, the consumption flywheel is spinning faster than anticipated. More than 50% of Agentforce’s new bookings came from existing customers refilling the tank in Q3. This consumption-driven growth has pushed Agentforce ARR to $540M in a year.
“We are now deploying agents across thousands of businesses of every size, and across dozens of industries – including highly regulated ones, like public sector, financial services, and life sciences. These aren’t vanity metrics – we’re seeing record growth, thousands of new paid global customers, and a surge in production deployments.
“AI success is a direct result of data readiness, which is an industry-wide issue. Many customers are leveraging the Data 360 to prepare their foundations for Agentforce – a necessary step that ensures when they do flip the switch, they see the high-utility token consumption your math highlighted.
“We feel very confident about the rate of Agentforce adoption across our customer base. The high level of customer success demonstrates the huge opportunity we have in front of us.”
The cloud company also stressed how recruiters at one of their customers, Adecco, receive 300 million applications every year, and Agentforce will help them engage with 100% of applicants, which they say was “never possible before”.
Williams-Sonoma’s deployment of Agentforce allows it to offer ongoing touchpoints and lifestyle services, allowing the brand to stay engaged with customers during the planning and creating phases, not just at the point of sale, Salesforce said.
They added that PenFed recently deployed Agentforce for IT Service, and is projecting a 30% reduction in operational expenses and $2M in annual savings.
Also, Engine deploys Agentforce to scale personalized travel. Customer service agents autonomously resolve 30% of service cases and cut service costs by 10%, employee agents in Slack provide instant call support, and Sales and SDR agents automate prospecting and lead qualifying, Salesforce told SF Ben.
Final Thoughts
Tokens served only really gives us a snapshot of the whole picture, but Salesforce is keen to stress how Agentforce is its fastest-growing product – and the number of customers is growing by nearly 50% QoQ, all while the “consumption flywheel is spinning faster than anticipated”.
This week is Salesforce’s Q4 and full fiscal year earnings. We will see just how far Agentforce is spreading among Salesforce customers.
