How to use AI without losing your edge

We hear more and more these days that relying too much on LLMs (Gemini, ChatGPT, Claude) actually hurts us – following the “use it or lose it” principle. It is a bit like using GPS, which makes us worse at finding our way around¹.

• 83% of AI users could not remember a part of an essay they had just written.
• Overall creativity is dropping, and thinking is becoming more standardized.
• Critical thinking is getting worse.
• Risk of cognitive decline².

At my office, e.g. the way people communicate in business emails has changed drastically. Many people have finally started using en dashes and em dashes that they never used before ;) Jira tasks are also better described (though unfortunately, they often include arguments based on made-up data).

On one hand, this is a benefit because emails are now easier to understand. They also tend to be less aggressive and less toxic.

The problem is that when we outsource our communication to LLMs, the human voice vanishes. To be honest, it just doesn’t feel right – I get the impression that a machine wrote to me, not a person. Jared Spool described it perfectly:

People don’t like being on the receiving end of AI usage.

Job applicants hate it when they learn that a company they’ve applied to has used AI to screen their application, especially if it then rejects them. At the same time, hiring managers don’t like it when they discover a job candidate used AI to embellish their résumé and other application materials.

College professors dislike it when their students use AI to do their assignments, while college students complain when they learn their professors are using AI to create their lectures.

So, what can we do?

Let's be honest: whoever hasn't used an LLM to polish a work email should cast the first stone. I do it too. But I believe there is a better way to do it.

It is worth writing the message yourself first, then giving it to an LLM with a request to fix only grammar and style errors.

This way, our brain stays active (after all, we have to put thoughts into words, build an argument, etc.), we sound like ourselves, and we learn to write better along the way!

And those who first use the brain, so, brain, brain, brain, then you use LLM, actually showed high neural connectivity.³.

For example: when I write a work email in English, I get extra practice. In my prompt for the LLM, I ask it to list and explain every correction. This way, I know exactly what mistakes I made. The model becomes my editor and English teacher.

Of course, this only works if you actually read the edits instead of just copying the result :)

But sometimes it’s okay…

Of course, there are tedious tasks where you just have to mindlessly “re-type” something, and an LLM can do it much faster. Personally, I see nothing wrong with using it then, and using the time you saved for more ambitious work.

I also often use it simply as a search engine. However, if I need reliable data, I also use a standard search engine to check the sources.

And going back to Jira tasks… well, I guess it’s probably better when non-technical people write them with the help of an LLM :D

As for other tasks

I have only mentioned minor, everyday topics so far, but this also applies to more complex work – especially tasks that require critical thinking. In my case, this might be data analysis or design. If I delegate these tasks to an LLM, I would eventually lose my edge (and I don’t want that).

AI works best as leverage. Use your own brain first – only then use the model to improve or speed up your work.