Most advice about AI at work is written as though you are about to redesign how your company operates. You are not. You have a job, and it has a Thursday afternoon in it.
So this is the smaller version: which tasks are genuinely worth handing over, how to find yours, and what to check before you paste anything sensitive into a chat box.
Start with the boring recurring thing
The instinct is to try AI on something difficult and interesting. That is the wrong first choice — hard tasks need judgement, and judgement is exactly what you cannot delegate.
The right first choice is something boring, repetitive and low-stakes that you do every week.
The weekly update nobody reads but everyone would notice missing. Turning meeting notes into actions. Reformatting the same report. The first reply to a familiar kind of enquiry.
These win because the work is *shaping*, not deciding — you already have the material and know what good looks like — and because the value compounds. A prompt that saves twenty minutes once is a curiosity. The same prompt every week is a day a year.
Look for the task you would be mildly embarrassed to say how long it takes.
Five that reliably work
Turning notes into something presentable
Meeting notes into actions with owners. Scribbles into a project update. A call into a follow-up email. You supply the substance; it supplies the structure and the tone.
Reading long things for you
A forty-page report, a long thread, a supplier contract. Ask what it says, what changed, what needs a decision, and what to be wary of.
Check anything you would act on. But going in knowing roughly what you are looking for beats reading cold.
The first draft of anything repetitive
Job adverts, product descriptions, FAQ answers, standard replies. Not the final version — the version that gets you past the blank page and into editing, which is much faster work.
Reformatting and restructuring
Prose into a table. A table into bullets. British spelling into American. A technical explanation into something a customer can read. Tedious, mechanical, and it is genuinely good at it.
Being a second pair of eyes
"What is unclear here?" "What would a sceptical client push back on?" "What have I not addressed?"
This is underused. It will not always be right, but it reliably catches the obvious gap you stopped seeing three drafts ago.
What not to hand over
- Anything where being wrong is expensive and nobody will check. Fine for a draft, not for a filing.
- Final judgement. Hiring, firing, pricing, whether to take the deal. It can help you think; it should not decide.
- **Work that is valuable *because* you did it.** If a client is paying for your reasoning, that is the product.
- Anything you cannot check. If you would not spot a plausible-sounding error, do not use it there yet.
Before you paste anything in
This is the part people skip, and the part that causes actual problems.
Assume anything you paste leaves your control. Depending on the tool and the plan, it may be stored, reviewed by staff, or used to improve the service. Free consumer tiers are the ones to be most careful with.
Some practical rules:
- Check whether your employer has a policy. Many now do. Find out before, not after.
- Strip identifying details. Names, account numbers, addresses. The tool does not need real ones to restructure a paragraph.
- Do not paste credentials, keys or personal data about customers or colleagues. Ever.
- Treat anything confidential as off limits unless you are on a business plan whose terms say otherwise — and check that they do.
- Know what your industry requires. Legal, medical, financial and public-sector work often has rules that are stricter than your instinct.
None of this makes AI unusable at work. It means the same care you would apply before forwarding a document externally, because that is closer to what pasting is.
Finding your own tasks
For one week, note every time you think *this again*.
At the end of the week you will have a list. Most of it will be shaping, summarising or reformatting — which is exactly the work these tools are good at. Pick the one that appears most often and only that one.
Get it working properly. Save the prompt somewhere you will find it. Then move to the second.
That is a far better use of a month than trying eleven tools and adopting none.
The realistic payoff
Not transformation. A few hours a month, taken off the tasks you liked least, with the work you actually care about untouched.
That is a modest claim, and it is the one that survives contact with a real job.
If you are still working out which tasks qualify, what AI can actually do for you covers the general shape — and once you have picked one, how to write a prompt that actually works is the difference between a useful answer and a bland one.