Salesforce used the opening of Dreamforce 2026 to make a commercial case for AI agents that can carry out multi-step business work, as public-market attention shifted from the chipmakers supplying AI infrastructure toward software companies expected to turn that computing capacity into recurring revenue.
The company said Agentforce annual recurring revenue had surpassed $1.5 billion, up 240% from a year earlier. Combined annual recurring revenue from Agentforce and Data 360, its data platform, approached $3.9 billion, a gain of more than 210% year over year, according to Salesforce’s latest quarterly results. Those figures place the company’s agent strategy at the center of its effort to show that AI spending can generate software revenue rather than remain concentrated in data centers and semiconductor supply chains.
The message arrived after a Sept. 14 U.S. trading session in which chip stocks broadly weakened while cybersecurity companies CrowdStrike and Palo Alto Networks each gained more than 13%. Salesforce and ServiceNow also attracted renewed buying interest. The session reflected a growing market debate over when large AI infrastructure budgets will produce revenue from enterprise applications.
Salesforce’s answer is to charge for completed actions. Its latest Agentforce products are designed to retrieve information, update records, seek approvals, follow up with customers and employees, and continue workflows across multiple business systems. That model moves beyond the one-question, one-response format associated with early workplace chatbots.
Salesforce puts agent actions at the center of pricing
Salesforce has expanded Agentforce with purpose-built agents for customer service, IT, HR, commerce, sales and supply-chain back-office tasks. The sales agent, called Hunter, remains in pilot and is scheduled for general availability in November.
The product lineup reflects an effort to make AI agents part of operational software rather than a standalone assistant sitting beside existing systems. Salesforce describes the agents as capable of working over extended periods, including following up across weeks and routing decisions for approval when human oversight is required.
That design changes how an enterprise software provider can measure the performance of AI. A conventional chatbot may be judged through user satisfaction ratings or response times. An agent assigned to process a return, qualify a sales lead or resolve an employee request can instead be assessed by whether the task was completed accurately and whether it reduced manual work.
Salesforce said Agentforce and Slack have completed 7 billion “Agentic Work Units,” its measure of agent activity. The company reported 3.2 billion work units in the second quarter alone, up 97% from the prior quarter. The metric gives a sense of product use, though it is not equivalent to revenue, paid accounts, or a standard industry measure of autonomous work completed.
The company is pairing those usage figures with a more explicit consumption-pricing model. Its Agentforce Flex Credits package lists 100,000 credits for $500, while a standard “Agent Action” uses 20 credits. At those published rates, a standard action would cost about $0.10.
This approach places Salesforce closer to cloud-computing economics, where customers pay partly for the resources or activity they consume. Each agent action can require AI inference, data retrieval and computing capacity, creating costs that differ from the fixed seat licenses that have underpinned much of enterprise software for decades.
Customer examples show uneven but measurable deployment
Salesforce cited several customer deployments to demonstrate how agents are being used in specific workflows. The figures come from separate organizations and do not represent a single standardized benchmark.
Engine, a travel and booking platform, resolves about 50% of chat inquiries through its agent, Salesforce said. Perk, another travel company, generates about 60% of its sales pipeline through a sales agent. Autism Queensland handles roughly 70% of administrative requests through an agent, according to Salesforce.
Retailer Hibbett uses AI in about 90% of its core shopping workflows, Salesforce said, while Anthropic’s customer-service tool Fin automatically resolves about 79% of customer-service conversations it handles. Each example points to a different task, customer base and measurement method, but together they illustrate the type of work Salesforce is targeting: repetitive, high-volume business processes with data already stored in enterprise systems.
The commercial question is whether this activity can support durable growth after accounting for the cost of operating the agents. A traditional software contract can produce additional revenue when more employees receive licenses. Agent deployment may have the opposite effect in some departments if companies need fewer human seats to process the same volume of tasks.
Salesforce’s credit system offers a way to capture revenue from that automation. If an agent executes more workflows, runs more processes or updates more records, usage could increase even as the number of employees using software directly declines. The model also shifts some risk to Salesforce, which must ensure that the price charged per action exceeds the infrastructure costs required to perform it.
Governance and cybersecurity move closer to the AI workflow
The spread of autonomous systems also makes identity management, permissions and runtime security more immediate concerns. An agent that can read internal files, access APIs, modify customer records or initiate a workflow needs tightly defined authority. A mistaken response from a chatbot can be inconvenient; a poorly governed agent with system access can create operational and security problems.
Salesforce introduced AIforce as a way to expose Salesforce data, workflows, business logic and governance controls across interfaces including Anthropic’s Claude and Salesforce-owned Slack. The initial Claudeforce release contains 37 prebuilt sales skills, Salesforce said.
The company also reported that Agentforce Coworker, a separate distribution channel, reached 100,000 user activations within 35 days of launch. Activations indicate initial access or setup, rather than confirming paid subscriptions or sustained use, but the figure suggests Salesforce is seeking adoption through multiple entry points instead of relying solely on traditional CRM deployments.
The stronger performance of cybersecurity stocks during the Sept. 14 session fits the practical demands of this software cycle. Companies deploying agents at scale will need to control what each automated system can access, how it authenticates and whether its actions comply with internal policies. That places security tools alongside workflow software in the spending conversation created by enterprise AI.
Salesforce’s Dreamforce announcements therefore focused less on another general-purpose AI assistant and more on turning automated work into a billable software unit. Its ability to maintain Agentforce growth will depend on whether customers continue to find enough value in those actions to absorb both the credit charges and the governance work required to deploy agents across critical business systems.
To see how AI reshapes trading like Salesforce reshapes workflows, explore Toobit’s AI copy trading guide now.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
