AI agents for enterprise work are AI software sold to companies as a worker rather than a tool. The company gives the agent access to its systems and a goal, and the agent does the desk work of a role, such as checking supplier documents or answering suppliers' questions. Such products are also called AI employees or digital workers. The company pays per user, or per task the agent completes.
An AI agent is built on a large language model (LLM), the kind of AI model that reads and writes text. On its own, a language model only writes answers. An agent is also given tools: logins to the ERP, the email system and the document store. Given a goal, it decides which tool to use, uses it, reads the result and decides the next step. It repeats this until the task is done, or until it needs to ask a person.
Here is a typical case. A hotel group in Goa with six hotels is growing. Its accounts team needs two more people to check supplier invoices and answer suppliers' questions about payments. Instead of hiring them, it sets up an AI agent. The team describes the job in a conversation: check each invoice against the purchase order and the goods received note, which is the record that the goods arrived; post the invoices that match in the ERP; and answer suppliers who ask when they will be paid. In its first month, the agent checks 2,000 invoices. It posts 1,780 of them, and sends 220 to a person, each with the reason, such as a missing goods received note. It also answers 300 supplier emails. The vendor charges ₹15 for each invoice checked, so the month costs ₹30,000.
The agent needs its own login, with only the permissions its job requires, and a record of every action. Products for this are covered under AI agent identity and security. Companies also keep a person to approve anything risky, such as a payment. This is the idea of human in the loop.
Paying per task instead of per user is a new way to price software. The page on SaaS pricing models compares the two. The agent reaches the company's systems through an integration platform, or, for old screens with no other way in, through an RPA robot.
Most companies in this category are very young and very small. At the same time, the established automation and ERP companies are adding agents to their own products. It is not yet clear whether the general-purpose AI employee will remain a product of its own, or become a feature of the software that companies already run.