At the end of last year, we made the ‘bullish’ case for Salesforce’s flagship product, Agentforce, which required a degree of faith. Salesforce’s agentic AI strategy is ambitious, heavily marketed, and clearly central to the company’s future, but real-world adoption was considered tentative. Pricing was still confusing, reliability came into question, and for many customers, there was generally uncertainty around what the product would actually look like.
That initial skepticism was understandable. Enterprise software rarely changes overnight, and AI adds an extra layer of uncertainty, particularly in environments where trust and predictability matter. As Robert Sösemann, Senior Principal Architect (AI) at Aquiva Labs, told SF Ben in October: “No one wants to be the first to take the risk.”
Fast forward to February 2026, as the conversation has shifted slightly. In their recent Q3 earnings call, Salesforce presented numbers that suggest meaningful traction for Salesforce is well underway – not just in terms of deals closed, but in customers running agents in production and coming back for more.
Agentforce Has Moved Beyond Experimentation
For much of 2025, it was easy to dismiss Agentforce adoption as exploratory. Internal demos and carefully framed success stories did little to convince skeptics that the product had crossed the line from experimentation to genuine enterprise use. However, that argument is a little bit harder to make now.
Just over a year after launch, Salesforce says Agentforce has reached 18,500 customers, with more than 9,500 on paid plans – making it the fastest-growing organic product in the company’s history. Customer growth is reportedly running at close to 50% quarter-over-quarter, while the number of customers running Agentforce in production increased by 70% in Q3 alone.
Revenue tells a similar story, with Agentforce currently generating around $540M in annual recurring revenue (ARR), which is up 330% year-over-year (YoY). When combined with Data 360 (formerly Data Cloud), Salesforce’s broader AI and data platform is approaching $1.4B in ARR – a figure that some would say was optimistic not too long ago.
Perhaps the most telling signal, however, is where that growth is coming from – more than half of Agentforce bookings are apparently driven by existing customers purchasing additional credits, which suggests that it must be delivering enough value in live environments for customers to be expanding their usage.
While this doesn’t present concrete proof that Agentforce has reached maturity or that its long-term role in enterprise workflows is quite there yet, it does look like it’s moving beyond just being an AI experiment into a product that customers are willing to pay for, deploy, and scale – even if it still does have some limitations.
Why There’s Still Skepticism, Even as Control Improves
Despite Agentforce’s accelerating adoption, skepticism has not vanished. Some of the questions raised last year are still as important now, particularly around trust, scale, and where agents truly belong within enterprise workflows.
To Salesforce’s credit, the company has been increasingly deliberate about how Agentforce operates in production environments. Rather than positioning agents as fully autonomous decision-makers, Salesforce has placed greater emphasis on control, limiting where agents can improvise, and ensuring critical actions follow defined, predictable paths.
As Madhav Thattai, COO of Salesforce AI, told SF Ben in January: “We rely on LLMs for reasoning and intent, but one key observation from working with customers is that as processes become more complex, their ability to reason consistently can start to waver.
“That’s why our approach to process execution is the best of both worlds; using LLMs for flexible reasoning and communication, while ensuring deterministic execution where accuracy really matters.”
Even so, questions remain about how far Agentforce can really extend. While agents already prove useful as interfaces, assistant, and orchestration layers, it is less clear whether they will ever be widely trusted to perform fully automated work in regulated industries such as healthcare, financial services, or HR.
We previously discussed on SF Ben how accurate agents need to be in order to scale. If we’re to follow the Six Sigma framework, anything less than practically perfect (99.999%) accuracy will likely prevent Salesforce’s desired growth for the product – especially if the 93% agent accuracy claim made by Salesforce CEO Marc Benioff is anything to go by.
There is also the challenge of scale. Early success with Agentforce tends to come from well-scoped use cases with clear boundaries and oversight. Whether that approach can translate seamlessly across large, complex Salesforce orgs – many of which are already managing significant technical debt – is still uncertain. As agent usage grows, governance and operational discipline are just as important as the agents themselves.
Skepticism, then, should not be read as resistance or doubt in Salesforce’s direction. Rather, it reflects the realities of enterprise adoption and the care required when introducing new forms of automation into critical business processes.
Final Thoughts: A Bullish Case, With Conditions
In early 2026, it feels fair to say the bullish case for Agentforce is holding. Adoption is growing, customers are putting agents into production, and Salesforce appears to be learning from real-world use rather than simply pushing ambition. The direction looks right, and the momentum suggests Agentforce is becoming a meaningful part of Salesforce’s platform rather than a short-lived AI experiment.
The harder questions, however, sit further out. Salesforce would likely hope that, in five years’ time, agents are embedded across most customer workflows and performing reliably at scale. What remains unclear is what that scale actually looks like in practice.
As agent usage grows, risks such as agent sprawl, governance overhead, and operational complexity could emerge, particularly in large orgs already carrying significant technical debt. The long-term challenge may be less about building smarter agents and more about managing them responsibly.
There are also unresolved questions around trust. Will hallucinations ever be reduced to a level that makes agents acceptable in regulated industries, or will there always be ceilings on autonomy where human oversight is required?
Those answers will depend not just on Salesforce, but on how AI itself evolves. For now, the bullish case remains intact.
