Short answer: start with work that involves a lot of text and comes back often, and have a person check the result. Below are five uses that the makers of the big models and the Dutch Data Protection Authority themselves give as examples.
1. Drafting replies to customers
The Dutch Data Protection Authority sees employees using AI assistants to answer customer questions. Anthropic names customer support as one of the areas where AI agents work well, because a conversation often ends in an action, such as updating a ticket.
Start like this: let AI draft a reply based on your own FAQ, and have an employee send it.
2. Summarising long documents
Case files, minutes, reports, long email threads. The Dutch Data Protection Authority, Anthropic and Google all mention summarising as a typical use.
Watch out: do not put personal data into a tool you have no agreement with. See AI and privacy.
3. Writing and translating text
A first draft of a quote, a product description or a social post, then a translation. Anthropic uses "write marketing copy, then translate it" as an example of a simple AI workflow.
Watch out: check facts, prices and names. Language models can make convincing mistakes. See How does a language model work?
4. Sorting incoming requests
AI can classify an incoming message (question, complaint, quote request) and send it to the right person or process. Anthropic calls this routing and gives customer service as an example.
5. Having small tools and scripts made
The big models can write and explain code. Google mentioned coding when it launched Gemini, and Anthropic names coding as one of the two most promising areas for agents, because you can test the result. Think of a script that converts an export from your accounting software, or a small internal form.
Watch out: always have code that touches customer data or money reviewed by someone who understands it.
Where to start
- Pick one task that comes back often and that you can check well.
- Agree on which data may and may not go in.
- Make sure your team knows how the tool works and where the pitfalls are. Under the EU AI Act, businesses that use AI have to take measures for this anyway. See The EU AI Act in brief.
Sources
- Autoriteit Persoonsgegevens: Let op, gebruik AI-chatbot kan leiden tot datalekken (6 August 2024, Dutch)
- Anthropic: Building effective agents (December 2024)
- Anthropic: Introducing Claude (14 March 2023)
- Google: Introducing Gemini (6 December 2023)
- OpenAI: Introducing ChatGPT (limitations)
- AI Act Service Desk: Article 4, AI literacy
Accessed 21 October 2026.