Salesforce’s Agentic Enterprise Index finds rapid growth in agent deployments, with new focus on measurable work

Salesforce has released its 2026 Agentic Enterprise Index, a report intended to track how businesses are adopting agentic AI through its Agentforce platform and what those deployments are producing in practice. The company’s central message is that the shift is no longer just about generating text or answering questions. Instead, more organizations are moving toward AI systems that take actions, connect to business systems, and show value through measurable “work” output.

According to the index, the number of agents activated per organization has risen nearly threefold over the past year. Once an agent is created, businesses have started deploying it to real use in an average of about two days. Salesforce also says that this cycle is shortening over time, with the time to use down 53% across the analysis period, suggesting that organizations are getting faster at operationalizing new agents rather than treating them as prototypes.

To determine whether agents are doing meaningful work, Salesforce introduces Agentic Work Unit (AWU), defined as a discrete task completed by an AI agent, the point at which “raw intelligence” is converted into real work. As of April 2026, Salesforce says Agentforce agents’ AWU output was increasing at a 15% compound monthly growth rate. The report portrays this as evidence that enterprises are not only deploying more agents, but also improving their ability to generate practical outcomes.

Salesforce frames the findings as a “tale of two rollouts,” where different industries lean toward different operational approaches based on their needs and constraints. On one side are high-volume, task-specific deployments, most common in consumer-facing sectors. On the other side are versatile, multistep deployments, which Salesforce says are more typical in operationally complex and heavily regulated areas such as manufacturing and public sector, where agents must handle a wider variety of tasks that involve cross-functional business logic.

In consumer and retail use cases, Salesforce says agents show aggressive AWU output, reflecting the scale of customer interactions. Yet the company also notes that retail agents tend to be more narrowly focused, averaging one to two actions per agent for much of the year. It adds that the capability of agents broadens during peak demand, with the average retail agent able to act on nine skills during peak shopping season, a 350% increase, implying agents are configured to handle more complex and multistep requests when pressure rises.

One example cited by Salesforce is Pandora, which it says deploys Gemma, an AI concierge powered by Agentforce. Salesforce describes Gemma as resolving routine customer inquiries such as order status, shipping tracking, and jewelry care FAQs, while also offering personalized gift recommendations. The company says Gemma handles 60% of routine support requests during peak traffic and contributes to a 10% increase in Net Promoter Score (NPS), enabling human representatives to focus on more complex, higher-touch interactions. Salesforce also says Gemma connects directly to back-end order systems and product catalogs to support faster and more personalized service at scale.

Beyond retail, Salesforce describes a different deployment posture in more complex B2B and regulated environments. It says manufacturing, financial services, and public sector (labeled as HLS in the report) are building more advanced agent networks than technology and retail, with agents deployed across a wider spectrum of tasks. Salesforce also describes “headless architecture” as a practical requirement for cross-system execution, explaining that decoupling agent logic from traditional front-end user interfaces allows agents to process tasks, execute actions, and trigger workflows wherever those workflows live.

In financial services and public sector, Salesforce points to both output growth and governance needs. It says the AWU output of public sector and HLS grew 227x and 19x, respectively. As an example, it describes PenFed, a federally chartered credit union, and its AI assistant Ace for online banking. Salesforce says Ace is secured behind online banking logins, evaluates account balances, checks loan application statuses, transfers funds, and answers using a curated knowledge base. It also describes Echo, another agent intended to extend multi-action capabilities to voice, designed to replace legacy interactive voice response (IVR) and automated teller systems, and it attributes the deployment approach to robust risk controls and cross-functional legal and compliance oversight.

Salesforce adds that agents are increasingly shifting from conversation to execution, with more action-to-output behavior as deployments mature. The company defines an action call as a step outside the chat interface that triggers a real-world digital action, while an output token is defined as a unit of generated text. Salesforce also says retailers that deployed AI agents during the holiday shopping season experienced a 4x higher sales growth rate, and that businesses using agents show stronger sales growth trends, particularly in consumer settings during peak periods.

The company’s report also includes some indicates about adoption and trust, but not all operational details are disclosed. Salesforce says the average employee engaged with an agent 300% more often per week over the analysis period, Slack agents averaged 67 sessions per week, and it characterizes Slackbot as its fastest-adopted AI agent tool in company history, with 83% of employees using it regularly. It also says customers trust agents too, citing a claim that over five quarters agents handled 170 times more customer service chats than previous years and consistently solved 7 out of 10 without human help, alongside a separate figure that 77% of shoppers who engaged with onsite branded shopper agents felt more confident in their purchase.

What to watch next is whether Salesforce’s “AWU” framing becomes a broader enterprise KPI, and whether faster time-to-use translates into more agents being connected to back-end systems with tighter escalation and governance. The index’s own premise is that the future of agentic AI depends on organizations balancing speed and scale with the ability to manage complex, multistep tasks. The next edition will likely need to show whether those measured gains persist as deployments expand beyond single workflows into deeper orchestration across more business units and regulatory boundaries.