Businesses are gaining a new way to measure how autonomous artificial intelligence agents interact with customers across websites, messaging services and social platforms. Salesforce and technology partners introduced the Agentic Enterprise Index on August 8, 2026, as companies move beyond simple chatbots and begin allowing software agents to answer questions, recommend products, resolve service issues and complete tasks on a customer’s behalf.
A new measure for machine led commerce
The arrival of the index reflects a major change in the way consumers and companies meet online. A customer may no longer begin with a search box or visit a brand’s website directly. Instead, the first conversation may take place with a personal AI assistant that compares products, checks availability, asks follow up questions and selects an option based on the customer’s instructions.
For businesses, that shift creates a measurement problem. Traditional analytics can show how many people visited a page, opened an email or clicked an advertisement. They are less equipped to explain what happens when an autonomous agent reads product information, negotiates a purchase, requests a return or contacts customer service without the person manually completing every step.
The Agentic Enterprise Index is intended to track that activity across digital and social channels. Salesforce says its 2026 analysis uses aggregated activity from organizations that deployed Agentforce agents in production throughout the period studied. The company reported that participating organizations nearly tripled the number of activated agents, reduced average creation time by 53 percent and saw agent skills expand by as much as 350 percent for complex tasks.
[salesforce](https://www.salesforce.com/news/stories/agentic-enterprise-index-insights-2026/)
Salesforce has presented the index as a view into how companies are deploying autonomous agents and measuring business value. Readers can review the company’s wider enterprise artificial intelligence research and announcements while assessing how the new data applies to their own customer operations.
From answering questions to taking action
Earlier generations of customer service software generally waited for a person to select an option, type a question or submit a form. Autonomous agents can perform several connected steps after receiving a goal. A shopper might ask an agent to find running shoes under a certain price, compare delivery dates, check a return policy and place an order after approval.
That difference changes the meaning of a customer interaction. One conversation may involve product discovery, inventory checks, payment authorization and post purchase support. If companies count only website visits or human initiated clicks, they may miss much of the commercial activity taking place through software agents.
The index is therefore aimed at questions that business leaders are beginning to ask. How many agents are active? Which departments use them most often? How quickly can a company create a new agent? Are employees trusting the systems? Do customers still need to be transferred to a human representative?
Salesforce reported that weekly employee usage tripled while customer escalation rates remained steady even as usage expanded. It also said retailers using agents recorded four times higher online sales growth in the group studied. Those findings are company reported and relate to organizations that met the index’s inclusion requirements, so they should not be interpreted as proof that every business will receive the same result.
[salesforce](https://www.salesforce.com/news/stories/agentic-enterprise-index-insights-2026/)
Why social channels matter
Social platforms add another layer of complexity. Customers now ask questions in public comments, private messages and community groups, often expecting a response within minutes. An autonomous agent can monitor those conversations, identify recurring concerns and route sensitive cases to a human employee.
A customer who posts that a delivery has not arrived may be seeking a refund, an explanation or simply reassurance. The words alone may not reveal the desired outcome. A reliable agent must read the context, identify the relevant order, respect privacy rules and avoid exposing personal information in a public reply.
Social interactions also carry reputational risk. An inaccurate answer can spread quickly through screenshots and reposts. A system that responds too aggressively may appear dismissive, while one that gives a vague answer may frustrate the customer. Measuring agent performance across social channels will require more than counting the number of replies. Companies will need to examine accuracy, tone, resolution rates, escalation quality and whether customers felt heard.
What the index may reveal
A shared measurement system could help companies compare their progress and identify weak points. Useful indicators may include:
- The number of active agents and the tasks each agent is authorized to perform.
- The percentage of customer requests resolved without human intervention.
- The time required to create, test and approve a new agent.
- The rate of customer escalations, repeat contacts and unresolved complaints.
- The accuracy of recommendations and the frequency of incorrect actions.
- The level of employee and customer trust across different channels.
These measures should be read together. A high rate of automated resolutions may look positive until a company discovers that customers are contacting it again because their problems were not actually solved. Speed is valuable, but speed without accountability can produce a poor experience at a much larger scale.
Consumers need control and clarity
For consumers, the growth of autonomous agents may make online services faster and more convenient. A personal assistant could monitor a subscription, find a replacement item or request a refund while the customer is working, traveling or caring for a family member. Research and industry analysis increasingly describe agents as systems that can recommend products, check stock, manage subscriptions and complete purchases within limits set by the user.
[bain](https://www.bain.com/insights/agentic-ai-in-retail-how-autonomous-shopping-redefining-customer-journey/)
Convenience will depend on control. Customers should know when they are speaking with software, what information the agent can access and which actions require approval. They should be able to review a transaction before money changes hands, revoke permissions and reach a human representative when the issue involves money, safety, health or a disputed decision.
Privacy is equally important. An agent may combine purchase history, location, preferences and conversations to make a recommendation. That information can improve relevance, but it also creates a detailed profile of a person’s habits. Companies participating in the agent economy must explain how information is collected, stored and shared, and they must provide meaningful choices rather than burying those decisions in complex terms of service.
Businesses face a new discovery challenge
Companies have spent years improving search rankings, web design and advertising so that human customers can find them. AI agents may become a new gateway to the same products and services. A brand will need accurate product data, clear pricing, reliable availability information and policies that machines can interpret without confusion.
That does not mean companies should write only for software. Consumers still need understandable descriptions, accessible customer service and the ability to make independent choices. The strongest digital strategy will serve both audiences by presenting information that is structured enough for an agent to read and clear enough for a person to trust.
Smaller businesses may face a particular disadvantage if access to advanced agents becomes expensive. Shared tools, open standards and fair platform policies could help independent retailers compete without building a full artificial intelligence department. Industry groups and technology providers will need to explain not only what the index measures, but also who can use the information and under what conditions.
The next test is accountability
The Agentic Enterprise Index arrives as companies seek evidence that autonomous systems are producing practical results rather than attention grabbing demonstrations. Its data may help executives decide where agents are useful, where human judgment remains essential and which workflows are not ready for automation.
We should judge the technology by the quality of the relationship it creates with customers. An agent that saves time while preserving consent, accuracy and access to human help can be genuinely useful. An agent that hides its limits, makes unauthorized decisions or forces customers through endless automated loops will deepen distrust.
The most meaningful measure will not be how many agents a company deploys. It will be whether those agents solve real problems while respecting the people whose money, time and personal information they handle. The new index gives the industry a starting point for that conversation, but public trust will depend on what companies choose to measure, disclose and correct.
