What should I use AI for? Thirty answers from real workdays

The most common question after a course day is never how the technology works. It’s this one: what should I use it for?

That’s a fair question. You know others get something out of AI, but the examples you hear are about writing blog posts, and you don’t write blog posts. You process cases, send quotes, call customers, hold meetings, read reports you didn’t have time to read.

This is a list of things people actually do. None of them require you to know anything about AI. Every point has been tested in EuroWork before being written here, with made-up but realistic material. The answers took between five and forty-five seconds and cost between two and fourteen cents each.

The camera is the underestimated button

This is the example that usually makes people pick up their phones mid-course.

You find a gadget in a box in the storage room. You can’t search for it because you don’t know what it’s called. Take a photo, press Attach a document and ask what it is and if it can be thrown away. The same works for the spare part no one knows the number of, for a component in a switchgear, and for that cable in the server room.

Same trick, more variations:

1. The whiteboard after the meeting. Take a photo and ask for a clear project plan instead of copying fifteen arrows and a messy box.
2. An old document with fancy handwriting. A marriage certificate, a protocol from the fifties, a note in an old binder. Ask for an interpretation, and for uncertain words to be marked so you know where to double-check yourself.
3. The sign, label, or manual in a language you don’t read. Take a photo and ask what it says.
4. A report page you photographed with your phone. Ask for a summary half the length, so you know in ten seconds if it’s worth reading the whole thing.

It sounds like a toy. Then you use it at work every week.

Reading what no one has time to read

This is where the biggest untapped gain lies in most organisations. Not writing faster, but not having to read everything.

5. The annual report or interim report you were sent. Attach it and say who you are: what in this concerns me, as a principal? As a procurement officer? We tried as a principal and got the school section first, the economy as indirect impact, and nothing about the water supply network.
6. Ask it to argue against itself. First: give me the four strongest reasons to go ahead with this. Then: give me the four strongest reasons not to. You get a better picture in two minutes than you do from an hour of pondering.
7. Two versions of the same contract. Attach both and ask what differs. We tried with two drafts of a purchase agreement where a clause about staff responsibility was removed and the warranty period shortened from 24 to 12 months. Both changes came back, with what they mean in practice for the buyer. That kind of clause is exactly what the eye skips over when reading.
8. Procurement request against your own references. Attach the request and a summary of what you’ve won and delivered before, and ask straight out if this is worth spending three days on.
9. The law or regulation. Ask for an explanation in language aimed at someone who isn’t a lawyer. Feel free to ask for a version a sixteen-year-old would understand, it often turns out best for adults too.
10. Applications against the job ad. Attach the job ad as a criteria list and the applications one by one, anonymised, and ask for yes or no with reasoning. The decision is still yours, but the basis is even.

Writing, and keeping your own voice

11. Let it review instead of writing. Write the answer yourself and ask for an assessment of the tone. You’ll often hear that what you wrote is good, and one concrete thing that would make it better. When we tried, we were advised to add a sentence acknowledging the recipient’s frustration, which was right. The same answer ended with a reference to guidelines we couldn’t find. Read source references with the same scepticism as everything else.
12. Shorten to seventy percent. Not because the text is bad, but because it will be read. In our test, a text of 906 characters became 601 characters, with all four facts intact. Make it a habit for everything you send to more than five people.
13. The reply to the angry email. Paste what you received and ask for a reply that calms things down instead of fuelling them. Read through carefully before sending: when we tried, the draft promised to cancel an invoice immediately, even though we had explicitly asked it not to promise anything we didn’t know. The tone was right, the promise wasn’t yours to make.
14. Change tone completely. The same text as formal, warm, self-deprecating. Seeing your own text in three tones is the quickest way to understand which one fits.
15. The response to the debate article in the local paper. Ask for the strongest counterarguments. You don’t have to use them as is, but you’ll know where you stand. Here the warning is most important: in our test, the answer built its arguments on a report and an internal memo that don’t exist. The arguments were useful, the numbers were made up. Use the reasoning, get the numbers yourself.
16. The translation that needs to hold up. Under Documents there’s document translation for text documents and presentations. The formatting stays intact, tables and layout remain, and you get back a ready file to download. If the text will live long in print or online: still have someone with the language as their mother tongue glance through it.

One thing to remember: as soon as you publish something, it’s your words. That applies no matter who or what wrote the first draft.

Specialists: when the same question comes every day

This is the feature that changes the most for those who answer similar questions over and over. In the left column, under Specialists, you build your own expert by giving it an instruction and some sources to rely on. You choose whether it’s yours alone, shared with selected colleagues, or with the whole organisation.

17. The one who answers the same questions every day. Add the range, prices, and rules as sources, and let the specialist write the answer. We tried the principle with HR material: the specialist answered with exact amounts and times from the sources, not from memory.
18. The reviewer against your graphic profile. Upload the design manual as a source. Then attach an ad and ask for a review. We tried with an ad that broke four rules, and got all four back with the rule quoted: the forbidden red colour, the logo in the wrong corner, the missing primary colour, and the all-caps heading. The kind of thing no one bothers to look up every time.
19. The reviewer against the regulations. Same principle with your procurement rules or environmental policy as a source. It reads both the rule in the source and the colours and composition in the image you attach.
20. The onboarding helper. Point the specialist to what’s already in your HR documents, and it answers the new employee without anyone writing a single new text. In the test, the right figures came back: the wellness allowance, the flex limit, and the balance thresholds, taken from the documents.
21. The house tone. Add a couple of texts you’re happy with as a source and ask it to write in the same spirit. Now it sounds like you, not like a brochure. Our test got back the habit of acknowledging the downside first, as in the example texts.

One thing that surprised us: the sources are reviewed as soon as you upload them. Our HR material was flagged as sensitive and thus tied to European models, while the tone examples went through without comment.

Meetings that don’t eat up the day

Under Documents you upload the audio from a meeting and get back a transcript. The work continues in the background, so you can close the tab and go to the next meeting.

22. No one needs to be the secretary. The person who usually takes notes can join the conversation instead, which is usually the one who has the most to contribute. In our test, the transcript picked out the right decisions and the right person responsible from the audio, and wrote "unclear, not specified in the transcript" where details were missing instead of making things up.
23. Decisions that don’t fall through the cracks. Save the transcript as a document, so it’s searchable for everyone who needs it, instead of sitting in someone’s notebook.
24. The transcript becomes a document like any other. It gets version history, a button to approve with the words that a human has reviewed the content, and export to Word, PDF, or PowerPoint.

Tasks that do themselves

25. The morning report. Set a rhythm for the task, daily or every Friday, so it runs before you come in. The result is a ready document in the feed when you log in, and the first thing you see is the Dashboard telling you what’s waiting. Monitoring a topic, a weekly summary on Fridays, a compilation on the first of the month.
26. Tasks with step-by-step control. Under The AI colleague you describe what you want done. You get a plan, and then Approve or Skip for each step before anything happens. It can search the web, search your own material, and create a document from the result. If you’ve connected your email under Connections, it can also read your calendar and inbox, but only read: nothing is sent in your name and nothing is booked behind your back.
27. Deep research. Choose Deep research when the question is too big for an immediate answer. It keeps working while you do something else, progress updates tick in in the thread, and the report comes back with a source list. Our test gave five sources, one of them scientific. One limitation: images you attach aren’t read at this stage, save them for a regular question.

Seeing the numbers with new eyes

If you attach a spreadsheet or a csv, EuroWork gives you a summary of the whole file: each column with its value range, most common values, and outliers, plus the first rows in plain text. That means you can ask about the shape of the material directly, while exact amounts are still something you check against the source.

28. Upload the file and ask for an analysis, without guiding it. Just say what role you have, for example that your job is to market the courses or to keep heating costs down in a property portfolio. You get back a review you didn’t know you wanted, and almost always two or three questions worth digging into.
29. Ask it to challenge your gut feeling. "I think the big customers order the most. Do you see anything in this file that contradicts that?" Having your own thesis tested is worth more than a summary.
30. Ask straight out, in your own words. You don’t need to know technical terms. All the questions in this list are written in everyday language, and they all worked.

This isn’t magic, it’s craft

One thing that surprises many: those who get the most out of AI aren’t the most technical. They’re the most skilled in their profession. The lawyer asks the best contract questions, the economist spots immediately if a number is unreasonable, the case officer recognises an answer that sounds right but is wrong. Your professional knowledge becomes more valuable, not less, because it’s what makes you know what to ask for and when the answer isn’t good enough.

Two habits make a bigger difference than all the tricks combined:

Don’t wait for the answer. Give the task, go do something else, come back. Deep research and meeting transcription are built for this, and both continue even if you close the tab. Tasks, on the other hand, wait for you: they stop at each step until you approve.

Read the numbers with scepticism. Reasoning and structure are the strong side. Exact amounts, years, and source references are the weak side, especially when the AI hasn’t been given any material to rely on. Give it a document and the numbers will be right, we saw that time and again. Ask it for numbers out of thin air and it can make them up with full confidence in its voice.

And what makes you dare

All of the above would be uncomfortable if you didn’t know where the data ends up. That’s why it says under every answer in EuroWork which model answered, in which country it was processed, and what the answer cost. The default is European models. The content is never used to train any model. If you choose Sensitive information, everything stays in the EU no matter what you asked for, and content that mustn’t be processed is stopped before it’s sent.

That means you can use the list above on real cases, real contracts, and real customers. Not on made-up examples.

Start with three

Pick three things from here and do them this week. For example: take a photo of something you’re wondering about, attach the longest document you’ve received this month and ask what concerns you, and give The AI colleague a task with the rhythm "every week".

In two weeks, you won’t remember how you did things before.