AI agent examples and use cases for companies
Updated:
An AI agent in a company takes a request, works out which steps it needs, looks things up in your systems, takes the action it is allowed to take and hands the rest to a person. Typical AI agent examples are a customer service agent that tracks an order and starts the return itself, a sales agent that researches a new enquiry and prepares the quote, a finance agent that works out why an invoice doesn’t match the purchase order, and an internal helpdesk agent that answers staff questions and files the request. What makes each one an agent is that it chooses its next step instead of following a fixed path. Below are examples by function, with what each agent checks, what it may do alone and what stays with a person.
What makes something an AI agent?
Anthropic, which makes the Claude models, draws the line clearly: workflows are systems where AI and tools follow “predefined code paths”, while agents “dynamically direct their own processes and tool usage”
Building effective agents. In practice, three kinds of system get called AI, and they do different jobs.
| Chatbot | AI workflow automation | AI agent | |
|---|---|---|---|
| What it does | Answers questions | Runs the same steps every time, with AI doing one of them | Decides which steps a case needs and takes them |
| Access to your systems | Usually none, or read-only | Fixed connections set up in advance | Tools it can choose to use: look up an order, update a record, send an email |
| Example | Explains your return policy | Reads every incoming invoice and enters it into accounting | Checks the order, the policy and the parcel, then starts the return or hands it over |
| Good for | Questions with written answers | Work that always follows the same path | Work where every case needs a slightly different route |
If the work always follows the same path, you probably want workflow automation, not an agent. Our AI automation examples by department cover that side.
AI agent examples in customer service
- Order status and delivery changes. The agent reads the message, finds the order, checks the carrier’s tracking and answers. If the customer wants a different address and the parcel hasn’t left, it changes it. If it has left, it says what can still be done and passes the case on.
- Returns. It checks the purchase date against your return policy and the item’s condition rules, creates the return and sends the label. Refunds above an amount you set wait for a person.
- Appointment changes. It finds the booking, checks the calendar for free slots, offers two or three and moves the appointment when the customer picks one.
This is the most proven use. LeoAI, the customer service agent we built for webshops, has handled 8,400+ conversations, 94% of them with no person needed (source:
leoai.hu). More on what it takes is on the AI customer service page.
AI agent examples in sales
- Researching a new enquiry. The agent reads the enquiry, checks whether the company is already in the CRM, reads its website and decides whether it fits the criteria you set. It creates or updates the record, drafts a reply and assigns it to the right salesperson.
- Preparing a quote. It works out what the customer asked for, checks the price list, stock and any agreed discount for that customer. If something is missing, such as the quantity, it asks the customer. The finished quote waits for a salesperson to approve.
- Keeping the CRM honest. Once a week it looks for deals with no recent activity, reads the last emails and proposes an update or a next step for each. The salesperson accepts or ignores each proposal.
AI agent examples in finance
- Invoices that don’t match. When an invoice differs from the purchase order, the agent checks the goods receipt and the supplier’s emails to find out why: a partial delivery, a changed price, a duplicate. It writes up the reason and drafts the email to the supplier or the correction. A person decides.
- Payment reminders. It checks open invoices and each customer’s payment history, sends the reminder that fits, and records a promised payment date when a customer replies with one. Disputes and anything about a contract go to a person.
- Expense claims against policy. It reads the receipt, checks it against the expense rules and the approver’s limits, and sends clean claims for approval and unclear ones back with a question.
Payments stay with people. The agent prepares and explains; it doesn’t move money.
AI agent examples in operations and procurement
- Orders that need clarifying. The agent reads an order sent by email, matches the products to your codes and checks stock. When a code is unknown or the quantity looks wrong, it asks the customer before entering the order.
- Supplier delays. When a supplier announces a delay, the agent finds the customer orders it affects, checks whether another supplier or warehouse has stock, and drafts the notices to customers. Someone approves what goes out.
- Reorder proposals. It compares stock levels with open orders and lead times and proposes purchase orders, with the reasoning for each. A buyer approves them.
Internal AI agents: IT, HR and company knowledge
- IT helpdesk. The agent answers how-to questions from your own documentation, checks the status of known problems, and files a ticket with the details already filled in when it can’t solve the issue. Access requests go to an approver.
- HR questions and leave. It answers questions about leave, expenses and benefits from your handbook, quoting the paragraph it used, and records a leave request in the HR system when the rules are clear.
- Finding company knowledge. It searches contracts, project notes and past emails to answer questions such as “what did we agree with this supplier on delivery terms?”, and links the documents it used, so the answer can be checked.
Keep agents away from decisions about people. The EU
AI Act lists AI used to filter job applications or evaluate staff as high-risk (Annex III), and GDPR restricts decisions with significant effects on people that are made by automated means alone (Article 22; see GDPR and AI). This is general information, not legal advice.
When is an AI agent the wrong choice?
More often than the marketing suggests. Anthropic’s advice to people building agents is to find “the simplest solution possible”, which “might mean not building agentic systems at all”, because agents trade speed and cost for flexibility and their mistakes can compound
Building effective agents. You probably don’t need an agent when:
- every case follows the same steps, so a rule or a plain workflow does it cheaper and more reliably;
- the job is answering questions from documents, with no action to take, which a simpler assistant handles;
- the cases are rare, so the time saved won’t pay for building and maintaining it;
- the systems involved offer no way in, so the agent would have nothing to work with.
How do you keep an AI agent under control?
An agent can act, so the limits matter more than with a chatbot. OWASP, the open security project, lists “excessive agency” among the top risks of AI applications: giving the system more access or freedom than its job needs
LLM06:2025. Another is prompt injection, where text the agent reads, such as an email from a stranger, contains instructions that try to make it do something else
LLM01:2025. The defences are mostly plain engineering:
- Only the tools the job needs. A returns agent can look up orders and create returns. It can’t delete customers or issue refunds of any size.
- Approval for anything costly. Money, contracts and messages that commit the company wait for a person.
- Limits outside the AI. Amount caps and permissions are enforced by the system it connects to, not by instructions the AI could be talked out of.
- A log of every step. What it read, what it decided and what it did, so you can check any case afterwards.
- A way out. When it isn’t sure, it hands the case to a person with the history attached.
What does an AI agent cost?
There are two parts. Building it depends mostly on how many systems it connects to, which actions it may take alone and where a person approves. Running it means AI usage fees, which for an agent are higher per case than for a single AI step, because it calls the model several times while it decides, looks things up and checks its work. Add hosting and someone to adjust it as your processes change. Compare that with the hours the routine cases take today, which is why the first step in how to implement AI in a business is measuring the work before automating it. In a small business those hours are often too few for an agent, and AI for small business covers the cheaper options.
What vSites does
We build AI agents like the examples above, connected to the systems you already use, with the limits and approvals agreed before we start. The price is fixed in writing up front, and the developer you talk to is the one who builds it. If a plain workflow or no automation would do the job better, we say so, and if you aren’t sure where to start, AI consulting answers that first.