The best first tasks for an AI agent are not glamorous. They are the repetitive jobs that land on someone's desk every day, follow a recognisable pattern, and eat time that could go to customers.

Below are seven such tasks. For each one you will find what the agent does and, just as important, what your staff still decide. In every case the agent does the first pass and a person stays responsible for the outcome.

1. Answering customer questions from your own documents

What the agent does. When a customer question arrives by email or through your website, the agent searches your own material: delivery terms, return policy, product manuals, price lists. It drafts a reply based on what it finds and shows which document the answer came from.

What staff still decide. A person reads the draft and sends it, edits it, or writes their own. Complaints, unusual requests and anything involving money or promises stay with your team. If the answer is not in the documents, the agent says so instead of guessing.

Good fit when: the same questions come in again and again, and the answers already exist in writing.

2. Handling incoming invoices

What the agent does. Supplier invoices arrive as PDFs in all sorts of layouts. The agent reads each one and picks out the supplier, amount, VAT, due date and reference. It can check the invoice against a purchase order and enter the details into your accounting software as a draft.

What staff still decide. Approving the payment. A person reviews anything that does not match: a different amount, a missing order number, a new supplier. Your existing approval rules stay exactly as they are.

Good fit when: you receive invoices from many different suppliers and someone types the details in by hand.

3. Sorting a shared inbox

What the agent does. A shared address like info@ or orders@ collects everything: questions, orders, invoices, job applications, spam. The agent reads each email, decides what kind it is and how urgent it looks, and moves it to the right person or folder with a short summary.

What staff still decide. How each email is actually handled. The agent only sorts. Anything it cannot classify confidently goes to a "check this" pile for a person.

Good fit when: someone spends the start of every day reading and forwarding emails.

4. Internal search across your files

What the agent does. Instructions, contract templates, old quotes and process documents are often scattered across shared drives and different systems. The agent lets staff ask in their own words, such as "what did we charge for installation in the last offer to a school?", and returns an answer with links to the source documents.

What staff still decide. Whether the answer is right for the case in hand. The agent shows where it found the information, so people can check the original. It only searches the folders you choose to give it access to.

Good fit when: new employees keep asking the same experienced colleagues where things are.

5. Drafting quotes

What the agent does. When a customer sends a request for a quote, the agent reads it, picks the relevant products or services from your price list, applies your standard terms, and prepares a draft quote in your usual format. It can also list what information is missing from the request.

What staff still decide. The final price, any discounts, and whether to send the quote at all. Pricing is a business decision. The agent prepares, a person decides.

Good fit when: many quotes follow a standard structure but still take time to put together.

6. Turning meeting notes into actions

What the agent does. After a meeting, the agent takes the notes or a transcript and produces a short summary: decisions made, open questions, and a list of action items with the person responsible and the deadline mentioned. It can create these as tasks in the tool your team already uses.

What staff still decide. Whether the summary is accurate and who really owns each task. A person reviews the list before it is shared. Make sure everyone in the meeting knows notes are being processed, and agree what kinds of meetings are suitable.

Good fit when: action points from meetings regularly get lost or are written up days later.

7. Handling booking enquiries

What the agent does. For businesses that take bookings, whether appointments, service visits or meeting rooms, the agent reads an enquiry, checks availability in your calendar or booking system, and proposes suitable times in a draft reply. It can also collect the details you always need, such as address or contact person.

What staff still decide. Confirming bookings that are unusual, large, or need special arrangements. You decide which kinds of bookings, if any, the agent may confirm on its own.

Good fit when: booking requests come by email or phone and need back-and-forth to find a time.

What these tasks have in common

Look across the seven and a pattern appears:

  • The work repeats many times a week.
  • The input is messy: emails, PDFs and free text rather than neat forms.
  • The rules mostly exist already, in documents, price lists or habits that can be written down.
  • A person stays in charge of decisions that carry risk.

That last point matters most. An agent that prepares work for a person to approve is easy to trust and easy to stop. Once results are consistently good, you can decide to let some low-risk steps run on their own.

How to pick one

Do not try all seven at once. Choose the one that costs your team the most time and has the clearest rules. A worked example: say two people each spend 30 minutes a day sorting the shared inbox. That is five hours a week on one routine task, and a reasonable candidate to start with. How to choose your first AI project walks through a simple scoring method, and what an AI agent costs helps you judge whether the numbers add up.