Nobody in the AI space posts this, which is exactly why it needs writing. There are four situations where AI is genuinely the wrong tool. Slower. Worse. Sometimes actively harmful. Knowing them makes you better at using AI, not less so.

1. Anything truly heartfelt

The eulogy. The wedding speech. The card to a friend going through something hard. The note to your kid that they will keep in a drawer for thirty years.

You can ask AI to draft these. The output will be technically competent. It will also be empty. The receiver will not know why, but they will feel it. Heartfelt writing is heartfelt because of who wrote it, what they noticed, what they chose to say. Outsource the choosing and there is nothing left.

If you are stuck staring at a blank page, that is a feature of the situation, not a bug to be optimised away. Sit with it.

2. Confidential data

Client information you signed an agreement about. Patient details. Anything covered by your professional code of conduct. Anything you would not put on a postcard.

Most consumer AI tools train on user input by default, or store it long enough that you cannot tell. Even tools that promise not to train often log conversations for moderation, and those logs sit on servers you do not control. The convenience of pasting a tricky paragraph in is not worth the risk that it surfaces somewhere it should not.

If you genuinely need AI on confidential work, it has to be a setup with proper data agreements and an enterprise tier. Otherwise: write the email yourself.

3. When you're learning

If your goal is to understand something, you have to do the thinking yourself. You can use AI to explain a concept, sure. But if every essay, every problem set, every email is drafted for you, the muscle never builds.

This is the version that catches people out. AI feels like learning because the output is right there. But reading a correct answer is not the same as being able to produce one. The next time you face the same problem without the assistant, you will not know what to do — and you will not know that you don't know.

If it didn't think, you didn't learn.

Use AI after you have tried, to check your reasoning. Not before, to skip it.

4. Your actual voice

Your writing voice — the way you phrase things, the rhythm of your sentences, the specific words you reach for — is yours. It took years to develop, mostly by accident. AI does not know it. The default AI voice is a smoothed-out average that reads like every other AI output, and the longer you let it draft for you, the more your voice softens into theirs.

Use AI to outline, to check, to react. Be careful about letting it draft. Especially be careful about the personal things — your social posts, your newsletter, your About page. Those are the places people are deciding whether they like you.

So what is AI good for?

Leverage. The repetitive bits. The first draft of something nobody will read closely. The structured task with a clear answer. The bit that is technical but not personal.

Use AI for leverage, not everything. The skill is knowing the difference.

Was something on this list a surprise?

The Kent AI Meet Up is where these conversations happen first. Free, in person, and built for people who want the honest version of how AI fits into their work and life.

See upcoming meet-ups →