How to know if AI is the right tool for the job
A short test to run before you open a chat window at all.
A common assumption: if you're not using AI for everything, you're behind. A common consequence: people use AI for tasks where it adds nothing, then conclude AI doesn't work.
Both mistakes come from the same source. There's no shared framework for when AI is the right tool and when it isn't. So here's one.
Can it be written down, and would you know?
Two questions gate everything else. The first is whether the work can be expressed as text at all. These tools take text in and give text out, and some handle images. That rules out more than people expect. "Decide what to do." "Judge whether this client is a good fit." "Have a difficult conversation with a colleague." AI can draft the email that comes after any of those. It cannot do the thinking.
The second is the one people skip: do you know what good looks like? Drafting an email in a voice you understand, summarising a document you've read, generating ideas in your own field: you'll spot a bad answer immediately. Writing on a technical subject outside your knowledge, producing legal language, translating into a language you don't speak: you won't. In the second case AI hasn't sped you up. It has given you confidence in something you have no way to check.
Clearly the right tool
You're staring at a blank page and need draft zero
Clear input, clear output: notes to summary, transcript to actions
The same kind of task across many inputs
Brainstorming where you need volume to get past the obvious
Translating, reformatting or shifting the tone of what you already have
Clearly the wrong tool
The task needs judgement you don't have
The information is sensitive and you're on a free tier
It has to be exactly right in a domain you can't verify
The work depends on your voice
The work is about relationship, not output
AI is brilliant when you can judge the output, and dangerous when you can't.
Where it actually pays
Past those two, the benefit is real in three situations and modest everywhere else. When the task repeats: ten product descriptions, twenty cold-email variants, fifteen transcripts to summarise. When the task is simply large: a fifty-page document to compress, an inbox to triage. And when you have never done this kind of thing before and need something to react to, because reacting to a passable draft is faster than starting from nothing.
If your task is none of the three, AI may still help, but the margin is thin.
Run the test before you open the tool
Every AI task carries overhead. Opening the tool, writing the prompt, reading the output, deciding whether it's good enough, refining it, pasting it back out. Small, but real. The rough rule: if the task would take you under three minutes by hand, a prompt is unlikely to beat it.
Three things to do this week
01
Check the work can actually be written down
If the task is really a decision, a judgement call or a conversation, AI can only draft what comes after it.
02
Ask whether you'd recognise a good answer
If you couldn't spot the mistake, you're not reviewing the output. You're trusting it.
03
Weigh the set-up against just doing it
A short, familiar, one-off task is usually faster by hand than by prompt.
Conclusion
01
AI fits work you can put into words and judge when you see it.
02
If you can't tell a good answer from a bad one, AI is a risk, not a shortcut.
03
The teams getting the most out of AI use it for fewer things, not more.
Go to next