It has been over six months since Salesforce first unveiled Agentforce, and just as we start the second quarter of 2025, there is certainly a wealth of new updates, advancements, and challenges that have cropped up that we can look back on.
However, looking back is only one part of the equation – the other is looking ahead to the future. That’s why in this post, we’ll explore expert opinions on the future of Agentforce and what could potentially hold this mammoth of a platform back.
The Current Status of Agentforce
Since its official announcement at Salesforce’s flagship conference Dreamforce last year, Agentforce has developed, and developed fast. At the end of 2024, Salesforce announced that Agentforce had a new version, cleverly named Agentforce 2.0. This new update gave the platform new reasoning, integration, and customization features to “supercharge” Salesforce’s flagship AI.
Then, early last month, Salesforce unveiled Agentforce 2dx, a new and further improved flagship offering. The latest update with this current version gives Agentforce the ability to operate autonomously in the background of “any business process,” according to the cloud giant.
As outlined in our previous article, 5 Reasons Why Agentforce Could Transform Salesforce’s Business in 2025, it’s evident from this year’s TrailblazerDX keynote that Salesforce hopes Agentforce is going to have a transformative effect on the world, providing the ‘digital labor’ necessary to solve an alleged global labor shortage.
This becomes especially important when it’s likely that there is no current labor shortage, and Salesforce is likely eager to not solve an existing problem necessarily, but counter one before it starts.
Power for Good
Although it has only been half a year since Agentforce has been in the world, it is definitely worth mentioning that the platform has been utilized successfully by a variety of different organizations with different use cases. In fact, Salesforce claims that they have closed 5,000 Agentforce deals since October, and yes, it is still early days, but the impact is already being felt.
A couple of weeks ago, I had the opportunity to attend an event held by Elements.cloud, where they Agentforce success stories were shared by the team, as well as their overall experience with the platform so far.
They spoke about their successful experience with using an HR agent for internal team holidays, guiding us through the different processes and highlighting the importance of proper process mapping.
Outside of this, finding organizations that already utilize Agentforce is getting easier by the day – RBC Wealth Management, Megnity, and Nebula Consulting are just some of the examples.
In fact, Sarah Kelleher, Nebula Consulting’s CEO, said that their very first agent was now live and already helping users find answers faster and that her team “learned a lot through the process [and are] still finding out what it can do and what is still not quite there.”
However, like with any new technology still in its relative infancy, there are always bound to be drawbacks, as I imagine Sarah’s team is finding out, and those drawbacks are just as important to consider as the benefits.
The Drawbacks of Agentforce
The most common complaints about Agentforce fall into three categories: the consistency and accuracy of its work, its pricing, and its current lack of desirable features.
Consistency and Accuracy
With any AI technology, especially in something as complex as agents, hallucinations are something to be considered and almost expected in the early stages of building and testing.
According to IBM, an AI hallucination occurs “wherein a large language model (LLM) – often a generative AI chatbot or computer vision tool – perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate.”
Hallucinations are just one of the “symptoms” of prompt-based AI work, alongside general inaccuracies or the AI technology – like an agent not understanding what you’re asking of it. This is something that Dominick Defazio, the CEO and founder of SalesCraft, has experienced with his work with agents so far.
Dominick, whose company had built a beta agent for a large debt relief company, created an agent for the company’s website to qualify new inbound leads. “The main limitation that they were concerned about was the inability for the agent to be consistent in following through on its qualification process,” he told Salesforce Ben. “[However], we got around that limitation by writing really good instructions.”
The art of prompt writing is not an easy feat, and it’s something that is vital to get right if any AI agent is going to successfully understand what is being asked and perform tasks.
“If you don’t know how to write instructions in a way to generate consistent responses, then you’re not going to have a consistent agent experience,” Dominick said. “Most people, I think, who pick up Agentforce are not using best practices for writing instructions.”
This is something that Comfort Ossai, a Salesforce Admin, also agrees with. “As a Salesforce Administrator, crafting well-structured prompts helps AI generate precise, actionable, and insightful responses,” she said. “Different prompting techniques such as Role-Based Prompting and Chain of Thought Prompting guide AI to provide more structured and step-by-step solutions.”
Pricing
Agentforce’s pricing is perhaps the most discussed challenge that the platform currently faces, primarily due to the confusion surrounding it and professionals finding it just too expensive.
Andrew Ermon, an Independent Salesforce Consultant, said that the usage-based pricing model that Salesforce has elected to roll with might seem good on the surface but is likely to cause more harm than good down the line.
“Salesforce’s ‘pay-as-you-go’ pricing sounds good, until you realize usage-based models often result in bigger bills, less transparency, and more risk for customers,” he said.
“When you can’t easily track AI token usage in real time, costs can spiral, especially if your bots run amok. This pricing shift might look flexible on paper, but it mostly benefits Salesforce’s bottom line.”
Lindsey Schweigert, a Salesforce Product Owner, said that she had been voicing her concerns on this topic, especially warning against finding use cases for Agentforce based on specific Salesforce use cases.
“I have put several posts out there with my concerns about building a business case for Salesforce when you don’t have a business case for Service Cloud because you need service licenses in order to use Agentforce,” she said. “The last company I worked at had no need for Service Cloud, and therefore, the cost for Agentforce was astronomical.”
This high cost, as many professionals have also pointed out, needs to be addressed in relation to smaller businesses and nonprofits, as they are often more likely to have smaller budgets that won’t be able to touch what Agentforce requires.
It’s something that Mohammed Jouad, a freelance Salesforce Business Analyst, agrees with wholeheartedly.
“Small businesses deserve access to Salesforce just as much as larger enterprises,” he said. “A subscription-based model, where businesses pay a clear monthly fee for specific features and services, would make [Agentforce] more accessible.”
He suggested that instead of the current usage-based model, Salesforce could offer tiers based on functionality, meaning that businesses wouldn’t need to pay for what they don’t use. Essentially, clear, predictable pricing would encourage more small businesses to adopt Agentforce.
Missing Features
Lastly, one of the last complaint focus areas for Agentforce is what features it does/doesn’t have. For example, Dominick lamented over the fact that Agentforce’s instruction UI isn’t the most functional, as you can add multiple instruction inputs to your agent in a completely wild order, and this makes things difficult to navigate down the line.
“The issue with that is that the agent doesn’t read those instructions in a consistent order,” he said.
He suggested that it should be the same as having a single Apex trigger with a handler class instead, saying: “The same kind of logic applies to the UI with agents, and I don’t think that the current [UI] is supporting that process.”
Dominick also mentioned that the customers he had recently been working for did not go ahead with an agent in the end due to Agentforce’s lack of voice capabilities. Salesforce has hinted that this is coming full-scale soon, though, so that will definitely be something to look forward to.
Final Thoughts: Changes for the Future
Agentforce is undoubtedly a force to be reckoned with, and the potential it has to create a lasting mark on the world of tech as we know it is immense. However, if the platform is to get to these heights, then Salesforce needs to be aware of what Agentforce’s users are saying and what needs to change to make using it as efficient as possible.
Salesforce may have closed 5,000 Agentforce deals since October, but over 3,000 of them were paid deals, which means we might not have the clearest picture yet of how well the product is actually selling yet.
When it comes to the other two main complaint focuses, professionals can make use of Salesforce’s free AI certifications in order to get better at Prompt Engineering and remain vocal about changes/additions to Agentforce features both online and at events.
A mixture of ecosystem involvement and Salesforce support will be needed to get Agentforce where it needs to be, but the good news is that this is certainly not an impossible task.
