Estimated reading time: 7 minutes

Key points:

  • Besides answering questions, AI can now carry out tasks itself: working in your files, writing and running code, generating images and drafting a plan.
  • The difference with chatting is that you give it a goal and the AI takes the in-between steps itself, instead of you typing every step.
  • This is called an AI agent. It works within limits you set, and you stay responsible for the result.
  • You need no programming knowledge for it. You ask in plain language.
  • Start with low-stakes tasks, where a small mistake does little harm, and build trust from there.

Contents

  1. From answering questions to doing the work
  2. Running tasks on your files
  3. Writing and running code, even without programming knowledge
  4. Generating images
  5. Drafting a plan based on your project
  6. What to watch for: control and trust
  7. What you need to get started
  8. How we can help
  9. Frequently asked questions
  10. Sources

Most people know AI as a chat window: you ask a question, you get an answer, and what you do with that answer, you do yourself. That is where most people stop. The tools can do more by now. They can also act: carry out a task from start to finish on their own.

Below you will read what that means in practice, with examples, and what to watch for when you let AI not just think along but also do the work.

From answering questions to doing the work

Do you mainly know AI from asking questions and getting answers? This article goes a step further: from thinking along to doing the work.

With chatting you type an instruction and wait for an answer. With the newer approach you give a goal, and the AI takes the in-between steps itself: retrieve information, make a choice, carry out an action, check the result. That is called an AI agent.

The difference is in what the AI does without your intervention. "Summarise this report" is chatting. "Update this report with the figures from these three files and put the changes in a separate list" is a task the AI handles itself. You simply read the outcome, you don't have to steer every step.

Running tasks on your files

The clearest step beyond chatting is working in your own files. Tools like Claude Cowork run on your computer with your folders, documents and programs. ChatGPT and Copilot can handle your files too, each in their own way.

A few examples:

  • A folder with twenty invoices from which you want the amounts and dates in a single overview.

  • A report you have updated with new figures, including a list of what changed.

  • A stack of meeting notes you have summarised into one list of decisions and action points.

You describe what you want, the AI does the work, and you check the outcome. It mainly saves time on work that is repetitive and involves many separate files.

Writing and running code, even without programming knowledge

AI can write and run code, and that is no longer just for programmers. Tools like Codex and Claude Code make it possible to automate small jobs without being able to program yourself.

Think of a script that merges a number of files into one report every month, or a simple tool that cleans up a list you would otherwise update by hand. You explain in plain language what you want to achieve, the AI writes the code and runs it, and you see the result. For recurring work that is now done manually, there is often a surprising amount of time to gain here.

Wireframe illustration of an artist's easel with a blank canvas and palette, representing AI generating images

Generating images

Making images is now built into the tools themselves. You describe what you need and you get an image back: an illustration for a presentation, a visual for a social post, or a first concept to make an idea tangible.

For anyone without a designer in house, this is a fast way to get usable images. For refined work a designer's hand still makes the difference, but for a first version or an internal presentation it is often more than enough.

Drafting a plan based on your project

A lesser-known use that pays off well: let the AI help you set up an approach for a project. This is where it comes down to you. How usable the proposal becomes depends on how well you explain your goal and your context.

Give it only a few documents and the instruction to make a plan, and you get a generic story back. Explain what you want to achieve, what the constraints are, what is already in place and where you are running into trouble, and you get an approach that actually makes sense: steps, an order, a first timeline. You use that as a starting point and sharpen it yourself.

So the difference between a vague and a usable result lies in your input. The sharper you describe your goal and context, the more usable the proposal you get back.

What to watch for: control and trust

An AI that acts on its own can also make mistakes on its own. That is why a few habits come with it.

Start with low-stakes tasks. Updating an internal summary or tidying up a folder can do little harm. Sending out a quote or queuing a payment is a different story; there you want to see every step before anything goes out the door.

Keep sight of what the AI did. You want to be able to check which steps were taken and to undo what went wrong. Good tools show you that.

And remember that you stay responsible for the result, not the tool. The AI does the work, you keep control. You build trust by starting small and expanding as it proves itself.

Wireframe illustration of a four-stage pipeline, representing starting small with one AI task and expanding step by step

What you need to get started

For this kind of task you usually need a paid account, and for working in your files often a desktop version of the tool. Which tool fits best depends on your work and on the systems your organisation uses.

Beyond that, the same applies as with regular use: start with one concrete task you already do anyway, and see what comes out.

How we can help

In the workshop Claude as your daily work partner you get hands-on with this yourself. You let the AI carry out tasks on your own files and build your own reusable method that you then call up with a single command. In one afternoon you experience the difference between chatting and getting things done.

Want to build further, for example automating a recurring process in your organisation? Then we think along about what lends itself to it, how to set it up safely and how to keep control of what the AI does.

Read more in this guide: how to use ChatGPT or Claude in your daily work, ChatGPT, Claude or Copilot: which fits which work and how to give AI good context and goals.

Frequently asked questions

Do I need to be able to program to let AI carry out tasks?

No. You describe in plain language what you want to achieve. Having it write and run code also works without you programming yourself.

Is this the same as an AI agent?

Yes. An AI agent is an AI that does not just answer, but is given a goal and takes the in-between steps itself: retrieve, choose, act, check.

Can AI really do tasks on its own without me checking every step?

Within limits you set, yes. For low-stakes work you can let it run and only look at the outcome. For work with consequences you want to follow along. You stay responsible for the result.

Which tool can do this?

ChatGPT, Claude and Copilot can all three do this, each in their own way and strong on different points. Which one fits you depends on your work and the systems your organisation already uses.

Is it safe to do this with my files?

That depends on the tool and the version. If you work with sensitive or personal data, choose a version and setting in advance where your data is not used for training and, where needed, stays within Europe.

Sources