Salesforce says businesses are rapidly scaling AI agent deployments, but data readiness, workflow complexity and unclear success measures could determine whether Agentforce adoption translates into better customer experiences.
Enterprise use of AI agents is accelerating, according to Salesforce, with the average number of activated agents per organization nearly tripling over the past year.
The finding comes from Salesforce’s 2026 Agentic Enterprise Index, which analyzes aggregated usage data from businesses using Agentforce and other Salesforce products between February 2025 and April 2026.
The report found that organizations in its dataset increased activated agents by nearly three times, while the average time required to create an agent fell by 53 percent. Agents are also taking on more capabilities, with the average agent able to perform six distinct business actions by the end of 2025, compared with two at the beginning of the year.
The findings add to Salesforce’s wider push to position agentic AI as the next phase of enterprise automation.
“Whether you’re spinning up agents to operate at massive scale or orchestrating them through deep, multi-step pipelines, the bottom line is they’re shipping real value,” said Joe Inzerillo, Salesforce President of Enterprise AI and Technology. “That ROI isn’t just showing up on the top line in sales numbers, but execution efficiency. We are moving from passive chatbots and predictive models to execution-driven agents that actually roll up their sleeves and drive real value.”
But there is an important qualification to the figures.
The businesses included in the Agentic Enterprise Index had to have agents activated in production every month throughout the analysis period, indicating how organizations already committed to Agentforce are using the technology, rather than providing a measure of adoption across the wider Salesforce customer base.
Salesforce said the increase in deployments is being accompanied by a rise in agent versatility.
During peak shopping periods, retail agents averaged nine skills, a 350 percent increase that Salesforce said reflects their ability to handle more complex, multi-step customer needs as demand increases.
Agents are also increasingly operating across different parts of the enterprise technology stack.
A service agent may retrieve a customer record, access sales information and make a personalized recommendation, rather than simply answering a question. Salesforce measures this activity using its Agentic Work Unit (AWU), which represents a discrete task completed by an AI agent. AWU output increased at a compound monthly growth rate of 15 percent through April 2026, according to the report.
Salesforce’s Two Models of Agent Adoption
The Agentic Enterprise Index identifies two broad deployment approaches.
Consumer-facing industries tend to favor high-volume, task-specific agents that can respond quickly to immediate customer needs. More operationally complex and heavily regulated sectors tend to prioritize versatile agents capable of handling multiple steps and applying cross-functional business logic.
Retail provides an example of the first model. Salesforce says retail agents generally perform one or two simple actions during much of the year, but expand to an average of nine actions during peak shopping periods.
The company cites Pandora’s Agentforce-powered Gemma as an example. According to Salesforce, Gemma handles 60 percent of routine support requests during peak traffic while contributing to a 10 percent increase in NPS.
More complex deployments are emerging in industries such as manufacturing and financial services. Salesforce said Siemens uses a coordinated multi-agent workflow to qualify leads across seven business units, with different agents responsible for engaging prospects, collecting missing information, applying qualification rules and routing leads.
For some CX organizations, the primary value of an agent will come from handling large volumes of relatively simple requests. For others, the value may come from coordinating complex processes that previously required employees to move between multiple systems.
Both approaches create different requirements around governance, integration and measurement. The question is whether increasing agent activity also translates into better customer outcomes.
Enterprise AI Adoption Still Has a Data Problem
That question becomes more complicated when the technology is deployed across fragmented enterprise environments.
Muj Choudhury, CEO of RocketPhone.ai and a former Salesforce Director, told CX Today in a recent interview that data remains one of the biggest areas of AI readiness that enterprises underestimate.
“Companies and enterprises in particular have a pretty unique advantage in the sense that they have a lot of historic data. And they’re also capturing data every single second of the day, whether they know it or not. And preparedness for me is ‘how do you leverage what you have today?’ But more importantly, how are you going to tap into the data that’s being generated?”
That is particularly relevant to Agentforce as Salesforce expands the range of tasks agents can perform. The more systems an agent needs to access, the more important the consistency, availability and context of the underlying data become.
Salesforce’s recent State of Commerce report points to a similar challenge. Only 27 percent of commerce organizations surveyed said their customer data was fully unified across sales, service, marketing and commerce, while 46 percent of B2C organizations reported duplicate or conflicting customer data.
The lack of unified data creates a potential gap between what an agent can theoretically do and what an enterprise can safely allow it to do. An agent may be capable of updating an order, issuing a refund or changing a customer record. Whether it should do so autonomously depends on the quality of the data, the rules governing the process and the controls around the action.
AI Is Only Part of the CX Transformation
Choudhury also cautioned against treating AI deployment as a solution in itself.
“AI isn’t everything. It’s never the panacea that’s going to cure all your ills.”
Salesforce’s Index shows that businesses are creating more agents and giving them more skills. But the number of agents deployed does not necessarily indicate whether an organization has improved the underlying customer journey.
The more useful question for CX leaders is where an agent can remove a genuine bottleneck.
That might mean automating a repetitive customer service interaction. It might mean helping a sales team qualify leads. Or it could involve coordinating several systems to prevent a field engineer from arriving at a customer site without the parts required to complete a repair.
In one example from Choudhury’s experience, an AI system listens to conversations between field engineers and customers, identifies the likely problem and checks whether the engineer has the required part before the appointment.
The AI represents only part of the overall solution.
“But the AI portion of that is about 20 percent of the overall because there’s still Salesforce, there is still a workflow management system, there are still GPS and GIS systems at play, and calendaring systems and scheduling systems.”
The agent may be the visible component, but the customer experience depends on everything behind it working together.
Are Enterprises Ready for the Next Level of Agentforce?
Salesforce said agent sophistication is increasing alongside adoption. Its Sophistication Index categorizes agent activity across five levels, ranging from reading and coordinating information through to writing data, analyzing information and parsing complex inputs.
That progression also increases the consequences of getting an agent decision wrong. An agent that drafts an email can be reviewed before it is sent, whereas an agent that changes a customer record, issues a refund or makes a financial decision has a much greater operational impact. This makes data quality and governance increasingly important as organizations move from conversational AI towards execution.
It also raises questions about how enterprises define success.
Salesforce’s commerce research found that only 32 percent of commerce organizations have fully defined AI success metrics and KPIs. Meanwhile, the Agentic Enterprise Index reports increasing Agentic Work Units, agent deployments and employee usage.
Those are useful indicators of activity, but CX leaders will ultimately need to connect them to outcomes such as customer effort, first-contact resolution, satisfaction, retention, revenue and employee productivity.
The Agentforce Adoption Test
Salesforce’s latest Index provides evidence that businesses already using Agentforce are scaling their deployments and increasing the sophistication of the work agents perform.
But it does not answer the wider question of how broadly Salesforce customers are adopting Agentforce. Its methodology focuses on organizations that had agents running in production throughout the analysis period, making it a study of sustained usage among an existing Agentforce cohort.
The next test is whether that pattern can translate into broader enterprise adoption and whether enterprises can connect agent activity to measurable improvements in the customer experience.
