On paper, Salesforce’s Agentforce is a successful product. It surpassed the elusive $1B ARR last month, with customers having now processed 28.6 trillion tokens and generated 3.8 billion Agentic Work Units (AWUs). At the end of last year, the SaaS leader revealed that it had closed 18,500 Agentforce deals, 9,500 of which were paid.
The product’s adoption has been a widely-debated issue, with some believing that Salesforce’s glowing numbers tell you everything you need to know, whilst others believe the official stats only tell half the story. However, we might be debating the wrong issue entirely. In fact, it is likely that Salesforce’s biggest AI risk isn’t adoption – it’s AI exhaustion.
Is Agentforce’s Adoption Anything to Shout About?
Many of us here at SF Ben have covered the ongoing conversation surrounding Agentforce’s adoption, and in March, I asked the Salesforce community what they thought about the matter. After all, Salesforce has been stressing that Agentforce remains the CRM company’s “fastest-growing organic product in Salesforce history” with a growth rate of 50% QoQ (with token usage increasing 152% QoQ). This would mean that high levels of adoption are being seen across the board. Right?
Well, the reality isn’t that simple. Community sentiment does appear to be positive, with a genuine interest in the product, but this is not necessarily reflected in the actual adoption levels of Salesforce’s wider customer base. The product does have notable enterprise customers – we were reminded of this recently at the London World Tour – but not every Agentforce customer is or will be of an enterprise level.
One proxy for Agentforce adoption is to look at active Agentforce project work. SF Ben’s recent surveys have seen respondents report 30-34% working on Agentforce projects among developers, admins, and architects, which is definitely not low. But it leaves questions as to whether it has crossed into mainstream default usage across the ecosystem yet
That being said, Agentforce’s adoption figures only represent one part of Salesforce’s ongoing battle to establish itself as an AI leader. The ratio of paid and unpaid customers has always raised questions, but it is perhaps not the biggest risk in its goal to achieve true AI domination. Instead, perhaps Salesforce should be considering whether its customers will even want to adopt this technology if they’re confused or exhausted.
The Growing AI Agent Problem
If you look at Salesforce’s AI release history, you would quickly lose track of how many agents they have announced. This is before you try to make sense of when Einstein GPT became Agentforce, Data Cloud changed names, and which new features are actually new versus being rehashed ones. Do you know the difference between the Qualified SDR Builder, Agent Builder, and Agentforce Vibes?
I regularly write about all of these products and these tweaks, and I still find myself getting confused about what has changed and why. It isn’t just me either – you don’t have to spend long on LinkedIn or Reddit to see the community confused at the evolving AI product portfolio, why every Salesforce conversation feels like an Agentforce conversation, and more.
When the Agentforce ‘hype cycle’ conversation was at its peak last year, it could have been easy to dismiss Salesforce’s customer base and community as stubborn and unwilling to adopt new technology. However, as time has gone on, it has become evident that they have simply been reacting to often whiplash-levels of change, with customers eager to learn more about the new AI technology but finding themselves being held back.
Confusing costs, updates, and the fear of having to redirect projects halfway through when Salesforce announces something new have all been reasonable obstacles. At some point, we have to stop and ask ourselves: how many agents are too many? Has Agentforce become too confusing to keep up with, let alone implement?
Final Thoughts
Even if Agentforce adoption metrics continue to climb, increased usage does not automatically equate to greater customer trust in the platform. We have already seen mediocre customer satisfaction levels with Salesforce’s own Agentforce Help agent, even as usage has grown.
However, this does not mean that Salesforce can’t overcome this issue. Going forward, it would be great to see more transparency around Agentforce’s satisfaction metrics, with information on the positives and negatives. Alongside this, there could possibly be more information on how to navigate Agentforce’s extensive and continuously growing AI portfolio, because, as I mentioned, how can you expect your customers to implement this technology if they don’t understand what it could entail?
