What It Is Good At, and Bad At

Most disappointment with AI comes from pointing it at the wrong job. It is genuinely excellent at some things and genuinely unreliable at others, and the line between them is more predictable than people expect.

Where it is strong

Working with material you supply. Summarising a long report, pulling actions out of a transcript, reformatting notes, translating. You provide the ground truth, so there is little room to invent.

First drafts. Getting from a blank page to something you can react to. Editing is far easier than starting, and this collapses the hardest part of writing.

Rewriting. Making something shorter, plainer, more formal, suitable for a different reader. It is very good at this and it is undersold.

Generating options. Twenty subject lines, ten angles, five objections you had not considered. You are the filter; it is the supply.

Explaining. Taking something dense and making it comprehensible at whatever level you ask for.

Structured, repetitive work. Categorising, tagging, extracting fields, converting formats. Unglamorous and reliable.

Where it is weak

Facts it was not given. Anything specific, recent, niche or local.

Arithmetic and precision. It is improving, and it is still not a calculator.

Knowing what it does not know. It will rarely volunteer uncertainty unless asked.

Your context. It does not know your clients, your constraints, your history, or what you can stand doing on a wet Tuesday.

Judgement with consequences. It can lay out a decision. It cannot carry it.

The pattern underneath

Notice what separates the two lists.

It is strong when it transforms something you gave it. It is weak when it must supply something you did not.

That single distinction will steer you correctly more often than any list of use cases. When a task feels risky, ask: am I asking it to reshape what I have, or to produce what I lack?

Adjust your expectations, not your tools

People switch tools hoping for a different answer. Usually the second tool has the same strengths and the same weaknesses, because they are properties of the approach rather than the product.

Better to move the task than the tool: give it more of the ground truth, and take the judgement back.