For the last couple of years, the Salesforce AI conversation has largely been centered around adoption. Are people actually using AI? Are companies experimenting with Agentforce? Are these tools actually making anyone more productive? Over time, the answer to all three is arguably yes – AI has now almost certainly gone beyond a side experiment for a small group of early adopters.
But that also means some of the questions around it are getting harder. As Senior Salesforce Advisor Timo Kovala told SF Ben, “Pilots and POCs are done and dusted.” Companies are now either pushing their AI projects further or quietly putting them on hold. And once the excitement of a successful demo wears off, familiar problems start coming back into view – problems like messy data, unclear ownership, cost, trust, and the simple question of whether any of this is actually worth it.
So applying the Gartner Hype Cycle, Salesforce AI may therefore be entering its reality check era. Not because the technology has failed, but because simply proving that people are using it is no longer enough.
The Experimentation Phase Is Ending
In short, the “reality check era” means AI has to graduate from proving it can do something to proving it is actually worth doing.
That means the conversation around Agentforce moves from basic adoption, demos, pilots, and polished use cases towards real business value, cost, reliability, data quality, and whether projects can successfully scale.
Now, this doesn’t mean any of the hype has disappeared. Gartner’s 2026 Hype Cycle for Agentic AI still places the technology at the “Peak of Inflated Expectations” – the point where excitement and ambition are running ahead of proven results. Just 17% of organizations surveyed by Gartner had deployed AI agents, while more than 60% expected to do so within the next two years.
So it would be premature to say Salesforce AI has already entered Gartner’s infamous “Trough of Disillusionment”, where enthusiasm starts wearing off and problems such as poor performance and weak ROI become harder to ignore. But some of the questions associated with that stage are already clearly surfacing.
Salesforce professionals themselves are certainly no longer waiting around with AI – our latest Admin Survey found that 43.6% now use AI daily or regularly, almost double the previous year, while 71% say it has had a real impact on their productivity. The problem is that personal AI use and a successful company-wide AI deployment are very different things.
That is apparently visible on the ground too. Salesforce Admin Vicki Moritz-Henry told SF Ben that admins specifically are “definitely using AI more, though not always enthusiastically,” but that Agentforce adoption remains much more limited in the organizations she works with.
One reason is cost predictability. “It’s hard to forecast credit consumption, and that makes it a tough sell internally,” she said. Instead, admins are often turning to more targeted AI tools for things like documentation, troubleshooting, drafting, and meeting notes, where “the spend is clear up front.”
That is where Timo believes the market has changed. “Pilots and POCs are done and dusted,” he told SF Ben. “AI initiatives are either ramping up or postponed indefinitely.”
In other words, simply showing that an AI project can work is becoming less impressive. Businesses now have to decide whether it’s useful enough, reliable enough, and valuable enough to keep investing in.
That may be the first real sign of Salesforce AI entering its reality check era. Experimentation isn’t disappearing, but the bar for what counts as success is getting much higher.
AI Is Forcing Salesforce Customers to Confront Old Problems
Once a project does move beyond the pilot stage, the problems tend to become much less exciting than AI itself. For Timo, one of the biggest is data. “Data and the lack of ownership thereof remain a challenge,” he said. “Part of the problem is that data is viewed separately from its business context and origins.”
As many of you know, this is hardly a new problem. Poor data quality, unclear ownership, and disconnected systems have been around long before Agentforce. The difference is that AI gives businesses another reason they can no longer afford to ignore them.
An AI tool can look impressive in a controlled demo, which we saw a lot of at this year’s Dreamforce. But putting it into a real business process means trusting the info it has access to. If that information is outdated, incomplete, or nobody quite knows who is responsible for fixing it, the limitations become much harder to cover up.
Timo pointed to another problem that might be familiar to many right now – urgency. “When bad data doesn’t disrupt core business functions, it’s always going to be priority 2 at most,” he explained.
That is one of the less glamorous consequences of Salesforce’s AI push. Companies can buy all the new AI, but none of that removes the need to get the basics right first.
And data quality isn’t the only old concern becoming more important. Vicki sees a similar problem around security and trust, particularly when admins start giving external AI tools access to Salesforce.
“Full buy-in on agentic AI just isn’t widespread among admins yet,” she said. In her experience, the biggest barriers are comfort with setting up the integration, confidence around the security implications of giving AI access to an org, and trust that the AI is actually doing what it is supposed to do.
Agentforce Now Has to Explain What It Actually Is
Following this year’s Dreamforce, Salesforce is actively broadening what its AI strategy looks like. The company’s yearly flagship event introduced AIforce, designed to bring Salesforce data, workflows, permissions, and business logic into whichever AI interface someone happens to be working. That includes Agentforce Coworker inside Salesforce, Slackforce inside Slack, and Claudeforce, which brings Salesforce directly into Claude.
This, on the surface, makes Salesforce look more flexible, with customers no longer having to leave the AI tool they already like and use in order to get access to Salesforce data and actions.
But it doesn’t come without some potential problems, which Timo pointed out: “Agentforce is facing a market positioning problem. What’s unclear for many is why you should go with Agentforce instead of Claude, GPT, Gemini, etc. + MCP + Salesforce.”
That isn’t necessarily a hypothetical comparison either – Vicki said she is already seeing a small group of more advanced admins connect tools such as Claude directly to Salesforce through MCP and use them to perform configuration work in the org.
“That’s a real shift, but far fewer admins are operating at that level,” she said.
Claudeforce makes this particularly interesting, with Salesforce actively working with Anthropic to let people use Claude’s reasoning while Salesforce provides the data, business rules, and governance underneath it. So the question then becomes: where does that leave Agentforce?
Timo describes it as something of an “identity crisis”: “Is it a platform? A capability layer? A tool? An architecture pattern? All of the above, or something else?”
His preferred answer is much simpler – he would like to see Agentforce positioned as “a core capability like Apex or Flow” that becomes embedded throughout Salesforce, rather than something customers necessarily think of as a separate AI destination.
And AIforce arguably points in that direction. Salesforce may not need to convince every user that Agentforce should be the AI interface they spend their day in. Its bigger advantage could be making Salesforce the trusted layer underneath whichever AI they choose to use.
If that is where things are heading, the next stage of the Agentforce story may be less about building the most autonomous agent possible, and more about quietly becoming part of how work gets done across the platform.
Final Thoughts
The Salesforce AI story has gone beyond the simple “hype versus failure” narrative. There is clearly real usage, real revenue, and evidence that AI is helping plenty of Salesforce professionals work more efficiently. The question now is whether that early momentum can turn into something more durable.
That means the standard of proof is much higher than it was when Agentforce first launched. In the reality check era, AI stops getting credit for just existing. Customers need to show that projects are moving beyond experimentation, that successful deployments are being expanded, that costs and risks are under control, and that the outcomes are strong enough to justify continued investment.
For Salesforce, that may be the next big test. 2024 and 2025 were largely about proving it could build convincing AI products. 2026 has been about proving that customers will actually use them. What comes next is arguably harder, which is proving that those customers still see enough value to keep using, expanding, and paying for them once the novelty has worn off.
