Why Agentforce Adoption Is Slower Than Expected – And What Salesforce Needs to Do

It’s no secret that Salesforce have gone to extreme lengths to market and sell Agentforce since debuting it at last year’s Dreamforce. A lot of our content here at Salesforce Ben has covered it’s every corner – from how to get started with it, what people think about the hype, as well as all the new upgrades, such as Agentforce 2dx, that Salesforce has announced. 

It’s been a pretty exciting time to be in the ecosystem, with it feeling like we may be on the brink of a whole new way of working with Salesforce, thanks to agentic AI. No one is more excited about Agentforce than Salesforce CEO Marc Benioff, who has claimed 2025 is the “absolute year of Agentforce” and that we’ve never seen a product grow at the speed that Agentforce has.

But there have been signs – alongside ongoing discourse within the ecosystem – that Agentforce still has a long way to go in convincing the masses of its true ability and ROI.

In February, Salesforce announced that it had closed around 5,000 Agentforce deals – with 3,000 of those being paid customers and 2,000 still testing the technology, per The Information. Many of these sales are also potentially part of wider packages, meaning customers are paying for bundles with Agentforce included but are not necessarily using the tool at all.

Despite Benioff’s optimism, Salesforce’s Chief Financial Officer, Amy Weaver, expects “modest” sales of Agentforce over the next year and that the company’s overall sales would rise 7% to 8% – its slowest ever growth rate.

Agentforce is proving to be a hard sell, despite its strong marketing approach. So, I wanted to investigate – gaining insight from industry experts – what people think of Agentforce so far, what is really holding people back from investing, and what the real reason is behind Agentforce’s slow growth rate.

Product Readiness vs. Marketing Hype

If you’ve been following Salesforce for the last year, you’ll know that it feels like Agentforce is everywhere at the moment. A lot of time and resources have gone into convincing the ecosystem that our future relies on this agent reality, and if you don’t follow suit, then you’re falling behind.

But the reality – when directly compared to the marketing – is nuanced for many who have used it.

Keir Bowden, a Salesforce technical expert with over 20 years of experience in the ecosystem, feels as though Agentforce itself is very much in its infancy.

“I’d say my experience of that version was that it felt a bit early – immature, even. There aren’t many standard functions available yet. So for any project, you really have to build a lot of foundational automation to allow the agent to take actions or do anything meaningful with your data.”

For the many who are still trialling Agentforce, the gap between promise and performance has proved frustrating. 

There have been a number of Salesforce demos that sell the idea that you can build an agent in under 30 minutes. However, the timeframe to get that agent ready for release is a different story.

Ben McCarthy, Founder of Salesforce Ben, said: “I had a demo the other day, and it bugged out halfway through. That kind of thing doesn’t help build confidence. Just because you can spin up an agent in 30 minutes doesn’t mean it’ll actually work properly in that time. Setting up, training, testing takes weeks, if not months.”

Many also feel that many of the demos Salesforce has shown are simple and consumer-friendly, not reflecting how agents can be used in complex, enterprise situations.

“Senior management are not confident that agents can perform reliably. Which is why lots of agents are being built, but only a few are signed off and in production” claims Ian Gotts, CEO & Founder of Elements.cloud. “Our experience is you can get complex agents working effectively in days.”

Keir expands on this theory and believes we need to start seeing more applicable scenarios for Agentforce.

“Everyone’s showing the same kinds of demos: book a table, return a dress,” Keir explained. “What we need are real B2B scenarios – an agent guiding someone through a complex sales or support workflow.”

This combination of what is expected and what is possible is having a significant impact on current adoption and potentially preventing people in the ecosystem from using the tool until it shows a higher level of performance. 

Rob Thomason, Senior Salesforce Architect at Gruve, believes that Agentforce still has a number of strong positives despite its early issues. However, he states that Salesforce needs to do more to encourage adoption – especially with the ecosystem feeling let down by previous products.

“Marketing-wise, there’s also a trust issue. People assume it won’t live up to the hype because they’ve been burned before. But honestly, this one is actually user-friendly. You can spin up prompts quickly. The challenge is, people don’t believe it.”

Until that confidence gap is closed, Agentforce risks being seen as a compelling idea but not yet a dependable tool in the short term. In order to help Agentforce grow, it might be time to deliver and showcase solutions that are going to work in more complex situations.

Data Debt and Implementation Challenges

The sentiment, in general, is that Salesforce should be looking to do more in order to push Agentforce forward. But what could users be doing to get themselves “Agentforce ready”?

As you may already know, messy orgs with technical debt and legacy issues are fairly common in the Salesforce ecosystem. This comes down to a number of different reasons – a long lifecycle of some orgs, the rapid evolution of the platform, and an overload of customization, among other things, has left many orgs with data that needs thorough cleaning.

In turn, this has a drastic impact on Agentforce implementation. As Jessica Oliver, Managing Director at Cosimo Consulting, puts it: “Many organizations have data and technical debt challenges. However, they have managed for years through workarounds or human intervention.”

“Technical debt and bad data don’t work with AI. It’s a really good opportunity for companies to use the benefits of AI as a reason to prioritize data and technical debt projects.”

In essence, a big roadblock to Agentforce’s growth comes down to org readiness. If these orgs aren’t cleaned up, then agents will have no meaningful data to work on.

“Loads of Salesforce orgs were built the ‘quick and cheap’ way, which isn’t scalable. Before a company can take on AI, they often need to clean that up. There are also data issues, broken processes, and understaffed teams – all of which make it hard to adopt something like Agentforce.” Ben McCarthy, Founder of Salesforce Ben

The promise of agentic AI is only as strong as the data and systems that feed it. Without clean inputs, the outputs will always fall short.

“Bad data in, bad results out,” said Rob. “You can have a $10K AI model, but if your data model is a mess with duplicate records and outdated picklists, the output will suck. It’s like putting a $1 burger on a fancy grill.”

The challenge is even more pronounced for teams operating across different markets, geographies, or business units.

“AI needs structured processes,” said Daniel Rudman, VP and Practice Lead at Gruve. “If your org is full of complexity and inconsistencies, it won’t work well. We’re trying to use AI for generating statements of work internally, but with all our different regions and use cases, it’s difficult unless you can clearly define the inputs.”

So for many companies, the blocker isn’t the agents themselves but the work that now must be done to get their orgs in check and ready for implementation. Until that is addressed, even the most complex agent will struggle to deliver.

Pricing Confusion and Risk Aversion

If there’s one thing that sparks discussion on Agentforce, it’s the current pricing model.

Salesforce’s decision to put Agentforce on a $2-per-conversation pricing model has generated some mixed reactions, with many confused and concerned, and others outright frustrated by how expensive it comes to.

In addition, the $75-per-user/month pricing for employee-facing agents has added to the unease, sparking broader debate around the platform’s affordability and accessibility.

The larger issue many have, however, is a lack of transparency and difficulties understanding what a conversation really means.

“It’s a tricky one,” said Keir. “The underlying AI costs are now very low, and it’s cheap to make requests. But with Salesforce, it feels like a premium product with a premium price tag.”

“There are so many factors that go into it. How many requests count as one? What if the user only asks one question – is that still a full-priced conversation? After 24 hours, does it reset? Is it tied to a specific customer?”

The result, he says, is a “black box” pricing model, where you’re being charged for something without fully understanding how that price is calculated and, ultimately, what you’re paying for. This, of course, is going to make executives very apprehensive to greenlight unless they have a concrete way to track or forecast spending. 

Rob echoed these sentiments, flagging issues surrounding product line items, tiers, and actual cost structure.

“Even I find the buying model confusing – so I can’t imagine what it’s like for newer admins. You go to buy Agentforce, and you’re told you need certain SKUs. Then, you’re told you actually need four more, and then it turns out you didn’t need some of those. It’s messy.”

The complexity of the pricing model and what you’re actually paying for is one thing, but another large concern is the ROI on what you end up paying for.

A significant concern for Keir is that, even if you are going ahead with purchasing Agentforce, the product is still stuck in that grey area of ability – which we touched on earlier.

“Let’s say you route everything through Agentforce and get amazing results, with every query resolved and no escalations. Great. But if every query [does] get escalated, now you’re paying twice: once for Agentforce and again for a human. In that case, you’ve paid $2 just to be told someone else needs to help.”

Still, not everyone sees the pricing model as out of line or an unfair reflection of the product. As Jess explains: “Consumption pricing is interesting. Pay for what you use! $2 a conversation doesn’t seem unreasonable when you compare to the cost of a human agent for the same task or the cost, for example, of training temporary staff during peak periods.”

“Also if AI can answer a customer query quicker than a human agent working through a backlog, then you have a better customer experience. There is more to consider than just price. It’s a cost versus benefit conversation.”

The topic of Salesforce’s discounting culture came up again in a discussion with Ben, who set a future prediction for the pricing model.

“I think they will bring the price down. There was also a spike in their stock when DeepSeek came out because it could help lower their AI infrastructure costs.”

“If Salesforce can run things cheaper, they can pass the savings on to customers.”

Salesforce Ben expert author Peter Chittum agrees with this line of reasoning, saying:

“Salesforce tends to be pretty risk-averse in their initial pricing of paid features, charging more, then adjusting down once they understand how usage and their costs will line up.”

This may be a necessary step in the future, as potential buyers look at comparisons between other agents on the market. Google’s AI agent, Vertex, is priced at $12 per 1000 queries at the moment, which is a significantly lower per-usage basis than Agentforce.

Either way, the issue stands with uncertainty and a current unwillingness to buy into an expensive model when the guarantees aren’t quite there.

As Ian puts it: “They don’t know how to price them yet, and right now, nobody wants to hand out a blank cheque – they want caps and predictable ROI. So the $2 price? It’s what you pay to learn of a narrowly scoped agent.”

The Missing Pieces: Enablement, Trust, and Ecosystem Readiness

While product maturity and pricing dominate the Agentforce conversation, there’s another significant factor to take into consideration.

Even if the tech gets better over time and the pricing situation improves, adoption may struggle to grow if Salesforce fails to invest more in enablement, partner support, and overall guidance.

Salesforce has made bold promises around low-code and fast setup, but the lived experience tells a story of fragmented documentation, inconsistent tooling, and a lack of enablement for both customers and partners.

Peter said: “I think probably the act of creating an agent and some actions are not hard and well documented. But all the blocking and tackling of getting an org ready for Agentforce, just to have it ready to create an agent, I have found fiddly and not well thought through in the setup experience.”

“The documentation is written as if it is many teams each owning a piece of the story. I’ve not found that there’s an easy single step-by-step document that takes me from a non-Agentforce enabled org to ready to build my agent.”

In essence, this sort of “patchwork” approach puts the burden on individual practitioners to work it out themselves. They will often have to do this by reverse engineering what actually works and piecing together information from different sources.

So while product immaturity is one thing, a lack of guidance on how to actually use Agentforce is another growing problem. Many have pointed to the recent disbanding of the Salesforce Well-Architected Program – a framework for best practices in Salesforce.

“Salesforce have removed the very team best placed to drive this work – the Well Architected team. In fact, this very team had made a very strong start to this in their keynote at Dreamforce 2024 showing real and concrete architectures for how to approach Agentforce projects.”

“I have to wonder if Salesforce have shot themselves in the foot in removing the very team that might have been the biggest accelerator of Agentforce adoption and project success among Salesforce practitioners.” Peter Chittum, Freelance Software Developer

And while Salesforce leadership continues to push the “this is the year of Agentforce” narrative, the on-the-ground reality is that many partners are still unsure of how to sell or deliver it confidently.

“Agentforce should be positioned as just another tool in the implementation partner’s toolbox,” Jess explained. “It’s not separate from everything else. It should be embedded in how we meet customer needs. If a company’s objective is to grow sales and we’re only looking at traditional Sales Cloud functionality, we’re missing a trick. We’re also not helping Salesforce grow AI adoption or helping customers modernize.”

The resounding message here is that Agentforce can’t scale without ecosystem clarity. Partners need repeatable playbooks. Customers need confidence. And everyone needs Salesforce to lead with transparency, not just marketing.

Reasons for Optimism: Why Agentforce Still Has Huge Potential

Despite the growing pains and skepticism around Agentforce, there’s no denying that it holds huge potential in the future.

For all the limitations it may have at the moment, there are certainly some building blocks in place that give Agentforce a good foundation to build on.

“Agentforce – and AI in general – is one of the most exciting developments I’ve seen in a while,” said Jess, reflecting the sentiment felt across much of the ecosystem. “Salesforce has done an incredible job building a strong core platform across all of their offerings, and the fact that AI can extend across those is really exciting.”

And while adoption may be slow so far, it doesn’t mean certain companies aren’t seeing the benefits already. At Elements.cloud, Ian detailed how they’ve made some impressive steps with the tool already.

“We built eight agents, and they’re all deployed in production. We’ve got another eight or so currently going through the process… Not just FAQs – we’re talking 26 steps, 5-6 actions.”

Ian’s experience shows that when Agentforce is approached methodically with a structured process, clear inputs, and real alignment between business goals and technology, it can work – and fast.

“We’ve now got experience building agents and can get from idea to deployed in days because we have a solid process, and a well documented org. We’ve built the agent-building muscles.”

There’s also growing optimism around how different companies are starting to experiment. Instead of going all-in, many are finding value through smaller, low-risk wins, which may be a good strategic move to make at the moment.

Jess outlined her first-hand experience of this, saying: “There are plenty of quick wins out there, with use cases that don’t require a massive transformation. I think we’ll see companies taking smaller steps rather than going all-in at once.”

“In the industries I work in which aren’t heavily regulated like banking, for example, there’s excitement – not fear. It’s just a question of how and when to adopt it.”

Looking ahead, while 2024 may not be the “year of Agentforce” as first suggested by Benioff, it may well be a year of breakthrough and calibration as users and companies learn to get comfortable with Agentforce. Like any sort of product, it may just need an adaptation phase on a wider scale before everyone really jumps in on it.

“Trends shift so quickly,” Keir explained. “First it was all about copilots, now it’s agentic AI. So, I suspect some companies are holding back to see if this trend sticks before investing further.”

“To me, this year feels like a year of stabilization. proving the tech works and delivers value. Then maybe sales will pick up more in 2026.”

And even with competitors circling and customers hesitant, Salesforce still has something that’s hard to replicate – trust, existing data, and deep integration with go-to-market teams.

“Agents and copilots are everywhere now, but I think Salesforce still has an edge – especially with sales, marketing, and service teams. Even if Agentforce isn’t the best technical solution, Salesforce has the users, the data, and the infrastructure already in place.”

“Yes, other agents from OpenAI or CrewAI might technically compete, but embedding them raises security and cost concerns. There’s a moat Salesforce still owns, which is trust and data control.” Daniel Rudman, VP and Product Lead, Gruve

Agentforce may not be ready for mass adoption today, but it’s evolving. Those paying attention now – testing, learning, and building the foundations – will be the ones best positioned when it is.

Final Thoughts

Agentforce is poised to have a transformative effect on the ecosystem, but it may be the case that it’s not going to hit the ground running and “take over” this year as Salesforce have suggested.

Over time, it’s likely that we will see a more mature product – one that’s affordable or offers clear value for enterprises, with better usage guidance and cleaner orgs to support it. But expecting that all to come together all at once is what’s preventing the growth and image that Salesforce has been promoting.

A lot of work in different areas still needs to be done before we see agentic AI at scale, and some things need to meet in the middle before we see its real potential.