Agentforce is officially turning one year old today.
Salesforce unveiled Agentforce on September 12, 2024, with CEO Marc Benioff hailing it as the dawn of a “new era of highly accurate, low-hallucination intelligent agents” that drive customer success.
“This is what AI is meant to be,” the Salesforce founder said at the time, as he outlined his “bold” vision to empower one billion agents with Agentforce by the end of 2025.
But while Salesforce often touts the success of its AI suite, the recently revealed Q2 ‘26 results forecast Q3 revenue below Wall Street estimates, which might be interpreted as a signal for lagging monetization for Agentforce. Salesforce stock dropped by nearly 7% after hours following the results.
Agentforce is only just a year old, and Salesforce is continuously ramping up the AI product, with ‘2.0’ and ‘3’ versions being revealed in the months following its launch.
Here, we take a look at how Agentforce has evolved, what the ecosystem thinks of it, and – as Dreamforce ‘25 approaches – what a possible ‘Agentforce 4.0’ might look like…
Dreamforce ‘24 and Agentforce 1.0
The ecosystem already knew about Agentforce before Dreamforce 2024. One of our first stories on the product, posted on August 29, 2024, revealed how the word ‘Agentforce’ was mentioned 39 times throughout the Salesforce Q2 ‘25 earnings call, signalling how important it would become.
At Dreamforce, Salesforce officially unveiled it to the world.
Looking back one year on, it seems almost funny that we were outlining things like ‘What is an Agent?’ and how these differ from chatbots, but it’s important to remember that these were still pretty fresh concepts – at least for the Salesforce ecosystem – at the time.
Now, pretty much everyone knows what Agentforce is. It’s become a central pillar of Salesforce’s marketing, and Marc Benioff regularly talks about how agentic AI is going to revolutionize the world.
Agentforce was front and center at Dreamforce last year. We at Salesforce Ben wrote at the time: “This year felt less like Dreamforce and more like Agentforce – and we wouldn’t be surprised if the name changed.”
At launch, Agentforce “1.0” came with a bunch of other announcements, including:
- The Atlas Reasoning Engine: AKA the “brain” behind Agentforce, which considers what needs to be done, generates (and refines) a plan, and takes action.
- Agent Builder: A tool that lets users create and customize agents, enabling them with automation, APIs, and code on the Salesforce platform.
- Agentforce Partner Network: Salesforce partnered with software vendors to extend agent-building capabilities, like adhering to the same ethical guardrails, data sharing, and actions.
The Einstein Copilot had been available for some time, but the launch was no mere rebranding of an already existing product.
The ability of an AI to take action was a remarkable step forward in the field, and for Salesforce veteran Peter Chittum, this next step felt almost inevitable.
“It validated what I already expected when I used the predecessor (Einstein Builder, etc.). It was clear Salesforce was moving toward an agent framework,” Peter explained.
“Given the broader AI conversation about agentic AI, Agentforce was the logical next step. I’d seen similar frameworks before – LangGraph, CrewAI (Microsoft), and others – so Salesforce going there made sense. It arrived a bit earlier than I expected, but directionally it was exactly what I thought they’d do.”
When asked about her initial thoughts on Agentforce, Salesforce Business Analyst Leanne Botwright – who shares a birthday with the AI product – told Salesforce Ben: “My first thought was, ‘Hmm, this is nice… but how does this really work for me and my users?’”
Robert Sösemann, Senior Principal Architect at Aquiva, responsible for Customer AI projects, told us: “It felt a little shiny but lacked relatability and real-world use cases that would help people instantly see the value. I wanted to see how it could solve actual challenges in my role, not just the promise of potential.”
Melissa Shepard, Salesforce CTA & MVP, said: “I was excited to see that Salesforce is trying to stay ahead and keep up with the trends. AI is a huge shift in technology, just like how the internet/web was, and I, for one, am happy that they are embracing it and trying to stay relevant.
“We know what happens when companies don’t do that: they end up like Siebel Systems! I know there’s a lot of hesitancy out there in the ecosystem, but I feel as though people should be happy that Salesforce wants to stay up with the trends and not get lost in the shuffle and become irrelevant.
“After all, that could affect our ability to continue working in the ecosystem if they suffer the same fate as Siebel. I just want to see them go about it in a way that helps and not hurts.”
Agentforce 2.0
Agentforce 2.0 was announced just two months after the initial launch and unveiled in San Francisco, with the new version boasting a library of pre-built agent skills for Slack, Tableau, CRM, and AppExchange partner-developed skills.
A number of pre-built agent skills were introduced in the newer ramped version, including:
- New CRM skills: New skills for sales teams, like Sales Development and Sales Coaching, which allow the creation of agents that nurture leads based on someone’s particular rules of engagement.
- New MuleSoft features: MuleSoft for Flow was intended to make it easier to create low-code workflows that span any system, with pre-built connectors for creating multi-system workflows at speed.
- Tableau Skills for Analytics and Insights: New Tableau Topics and Actions provided data visualizations and predictions for a better understanding of agent responses.
- Partner Skills Through AppExchange: Salesforce said that Agentforce, backed by the “first-ever enterprise ecosystem” of agent skills, lets customers extend Agentforce with custom Topics and Actions.
- Agentforce Recommends Skills: Users could now create new agents “in seconds” with natural language descriptions, and Agent Builder allowed the composition of new agents by auto-generating relevant topics and instructions while pulling from the library of skills and actions already available to you.
Agentforce-Slack compatibility was also enhanced: Slack users could now start a conversation directly from the Agentforce Hub, or @-mention Agentforce agents through DMs or in channels.
Agent Builder also brought in pre-built Slack Actions like “Create Canvas” or “Message Channel”, and Slack Enterprise Search was introduced, meaning Agentforce could draw from conversational data, enhancing the relevancy of responses and actions.
The Atlas Reasoning Engine also got a boost, thanks to new capabilities in Data Cloud, providing Agentforce with greater context.
Agentforce 2dx
The first Agentforce update of 2025 came at Salesforce’s smaller annual company event, TrailblazerDX. It was at the main keynote where the CRM giant revealed Agentforce2dx, which was described as being able to operate autonomously in the background of “any business process”.
Salesforce demonstrated how Agentforce could integrate seamlessly within your org’s data systems, business logic, and user interfaces, allowing agents to act quickly when called upon.
Several new tools and features were also introduced, including:
- Agentforce API: Integrating Agentforce into back-end processes, other systems, and directly into applications.
- Agentforce for the command line: Salesforce CLI commands to build, test, and deploy agents.
- Agentforce Invocable Actions: Embedding Agentforce within Salesforce business logic, like Flow and Apex.
- MuleSoft for Agentforce: MuleSoft Topic Center allows developers to use natural language to create topics and actions from MuleSoft APIs, while the MuleSoft API Catalog centralizes access to APIs across MuleSoft, Salesforce, and Heroku.
- Agentforce Steps in Slack Workflow Builder: Enabling developers to embed Agentforce into no-code automations in Slack.
- Agentforce Employee Template: Allows customers to create multiple employee agents that can be configured and deployed across any line of business.
- Agentforce Surfaces: Deliver rich content within Agentforce across all Digital Engagement channels, adding dynamic, interactive components and media that are channel and device-specific.
- Agentforce Cards: Lightning web components can be embedded within the response of Agentforce actions.
- Tableau Semantics: Create organized data structures – semantic models – which Agentforce can take advantage of.
There was also the introduction of new tools for both pro-code and low-code users to help admins and developers build, test, and deploy agents, as well as a free Agentforce Developer Edition for developers to experiment with agents and explore Data Cloud features.
Moreover, we were introduced to the AgentExchange – Salesforce’s agentic answer to their well-renowned AppExchange. The platform has expanded to 80 applications from 73 providers, and now boasts over 200 partners, including major companies like Google Cloud, DocuSign, and Box.
Flexi-Credit Pricing Update
When Agentforce first launched, Salesforce went ahead with a headline pricing of $2 per conversation. On paper, this made sense for customer service chats, but in reality, it quickly became one of the ecosystem’s biggest gripes.
Conversations can run long, branch into side threads, and any conversation would count – even if it wasn’t meaningful. This made costs unpredictable and hard to budget for, which ultimately was going to stall adoption if they continued with it.
This led to a rethink in May 2025, when Salesforce rolled out a new flexible pricing model.
The centerpiece was Flex Credits, which charge by action rather than conversation. Instead of paying $2 for a potentially sprawling exchange, customers now buy bundles of credits (100,000 for $500), with each action costing 20 credits – in other words, about $0.10 per action.
Salesforce also launched a Digital Wallet – a dashboard that tracks usage, sends alerts, and helps forecast spend. The old conversation model didn’t disappear, but Flex Credits became the go-to option for broader use cases.
This change landed well for many, with Rob Thomason, Senior Salesforce Solution Architect at Gruve, stating at the time: “It is encouraging to see Salesforce adjust their pricing strategy to what is happening in the marketplace. It is certainly not the norm for Salesforce to discount offerings. They just don’t do it ever. The ‘bridge too far’ just got a little shorter for some folks.”
Still, not everyone was convinced. Credits are clearer than conversations, but customers still need to guess how many actions a process really takes, and juggling multiple pricing models adds fresh complexity.
Robert Sösemann said: “The $2 per conversation made sense at first, since it was framed around customer-facing service agents replacing costly humans. But the model never fit in-org use cases or AppExchange apps, which is where my real interest lies.
“The newer pricing addressed that gap – but honestly, Agentforce should be included in the core license, like Flow or Apex. Salesforce licenses are already expensive, LLM tokens are cheap and getting cheaper, and Salesforce isn’t even running top-tier reasoning models.
“To drive broad adoption, it should come bundled with a generous token allowance, with only heavy overuse triggering extra costs.”
Melissa Shepard said: “I think the pricing model is a bit tough, and I know that they are up against competition from other platforms where you can build and integrate agents that do it a lot cheaper. There’s definitely value in staying within the Salesforce ecosystem for running agents, but I feel as though it’s going to need to be more cost-effective for companies to start to adopt it.”
The shift was welcomed by the ecosystem, but it also showed how fine the line is between flexibility and confusion with AI pricing.
Agentforce 3
By the time Agentforce 3 was introduced in June 2025, Salesforce was under some pressure to prove that Agentforce was ready for prime time.
Early adopters of the AI tool had made it clear that they needed more control and visibility, as well as confidence that they could scale agents without running into budgeting nightmares.
Agentforce 3 delivered a number of new enhancements to address these customer concerns, including:
- Command Center: Real-time visibility into what agents are doing, with tools to monitor, fine-tune, and catch inefficiencies.
- Open standards (MCP Support): making integrations and partner tools easier and less bespoke.
- Global scale improvements: Lower latency, model failover options, and expanded language/geography support.
- New per-user SKUs: Unlimited action usage for employee-facing agents, shifting the model away from unpredictable metering.
From our perspective, this really felt like Salesforce was taking the right steps to address the issues raised by the ecosystem, offering transparent solutions that would help push adoption rates up.
We spoke to Salesforce SVP Sanjna Parulekar in June, who said: “The idea is that you now have full observability and governance across all of your Agentforce agents, down to the individual interaction level.
“For example, imagine you have an agent running on your website, and you want to know if a competitor is being mentioned on your help site. How does your product compare to another? With these new tools, you’re not just able to see that someone asked that question; you can replay the full back-and-forth and inspect exactly how the agent responded. That kind of visibility is incredibly powerful because it helps you continuously refine and improve how your agents operate.”
Dreamforce ‘25… and Agentforce 4?
While speculation is usually out of our remit, all of these product updates, as well as the upcoming Dreamforce 2025 event, have got us thinking about Salesforce’s next move. Will we see an Agentforce 3.5 or 4.0?
For Robert Sösemann, another update with minor feature updates probably won’t move the needle – the next announcement needs to be a game-changer.
“The constant ‘mini-releases’ with new names feel a bit silly. Some are substantial – like the DX release with CLI and packageability – but many features don’t actually work, as we found with agent templates. If there’s an Agentforce 4, it should ship features that really work.
“My biggest pain point is the weak reasoning orchestration: it often fails at tool use and defaults to ‘prompt injection’ when corrected. Agentforce 4 should be a real AI breakthrough, not just a UI tweak – something big enough to match Dreamforce.”
Anastasiya Rastorguyeva, a Salesforce Consultant at VRP Consulting, said: “I think the real focus should go beyond speed and security – which 3 has already addressed well – and move toward trust, orchestration, and cross-ecosystem intelligence.
“It’s not enough for AI to be reactive. The next step is an agent that can anticipate needs, orchestrate workflows across teams, and even reason across systems beyond Salesforce.
“Another key area is transparency. Users need to clearly understand why an AI is recommending something, not just what to do. Building explainability and bias-awareness into Agentforce ‘4.0’ would help establish the kind of trust that drives real adoption.”
Mellisa Shepard predicts: “We may see more predictions on work that needs to be done instead of relying on complete instructions on everything that needs to be done, sort of like suggestions on ‘how to best complete this task or process in the most optimized way’, something like that.”
Peter Chittum provided a practical view of what the next step for Agentforce should look like. The real challenge isn’t about implementing shiny new features, but continuing to help their customer base deal with decades of technical debt and poor documentation.
“The problem for Salesforce isn’t features – it’s the technical debt in their customer base. Most orgs haven’t documented the semantics of what they’ve built, and without that, it’s hard to make consistent, effective agents.
“More than new functionality, Salesforce needs an effort like the Lightning adoption push in 2017: specialists, tools, best practices, and a global campaign to help customers clean up their metadata and establish a semantic source of truth.
“If Salesforce wants Agentforce to drive the next $100B in revenue, they’ll need to invest heavily in helping customers escape 20 years of accumulated technical debt so they can actually use agents in meaningful ways.”
Final Thoughts
Oh, how time flies. Agentforce has just turned one, and while Salesforce can celebrate its ambition to make this AI tool a cornerstone of the platform, it’s clear they’re not quite where they’d hoped to be.
Adoption has been slower than expected, the user base remains relatively small, and investors are pushing hard for Salesforce to start showing real results from its flagship AI bet.
As Peter hinted, Agentforce may have been introduced before Salesforce was fully ready, and much of the past year has been about keeping things on track while filling in the gaps. That makes the next phase critical.
If we do see an Agentforce 4, the ecosystem will be watching closely – and perhaps a little skeptically – to see whether it represents the real leap forward Salesforce needs to win wider adoption.
