Seven Everyday Tasks You Can Hand to an AI Assistant (And Three You Should Not)

An AI assistant will draft almost anything you ask. The real test is whether you would put your name to the result. Seven everyday tasks pass that test: first drafts, summaries, minutes, rewrites, explanations, meeting questions and formulas. Three do not: confidential data in a public tool, decisions about people, and unchecked figures or legal wording.

In short

• The question is not what an AI assistant can do. It is what you are willing to sign your name to.
• Hand it the tasks where a draft saves time and a wrong answer is easy to spot: emails, summaries, minutes, rewrites, explanations, meeting questions, formulas.
• Keep three things with a person: confidential data in a public tool, decisions about people, and final figures or legal wording.
• Poor output is almost always a poor request. State the role, the task, the source material, the format and the constraints.
• It drafts. You decide. Nothing leaves under your name until you have read every line of it.

It is ten past nine on a Monday. Your inbox holds a forty-page supplier report, a request for last week's minutes, and a board paper due Thursday that nobody has started. You open an AI assistant, paste in the report and type "summarise this". Thirty seconds later you have a page of tidy bullets, and no idea whether any of them are true.

An AI assistant is a chatbot that writes text in response to a written request, and it is very good at producing something quickly. The question is not what AI can do. It is what you are willing to sign your name to.

Seven everyday tasks pass that test. Three do not. The line between them is what happens if the assistant is wrong and nobody notices.

Seven tasks where a draft is worth having

These seven share two features. The assistant has enough to make a reasonable attempt, and you can tell at a glance whether the attempt is any good.

• First drafts of emails and documents. Give it the purpose, the reader and the three points that matter. A mediocre draft you can fix in five minutes beats a blank page you stare at for twenty.
• Summarising long documents. The forty-page supplier report becomes ten bullets. Then read the two sections that actually matter before you rely on any of them.
• Turning notes into minutes. Your scrawl from the Thursday meeting becomes a clean record with decisions, actions and owners. You check that nothing was added and nothing lost. Assistants are quietly good at inventing an action nobody agreed to.
• Rewriting for a different audience. The engineers' update becomes a short note for the board, or a plain paragraph for customers. The facts stay. The register changes.
• Explaining unfamiliar terms. That clause on page nine of the contract, in plain English. A starting point, not the final word.
• Preparing questions for a meeting. Paste in the agenda and the background and ask what you should raise. It often surfaces the angle you had missed.
• Drafting a spreadsheet formula to check. Describe the calculation, take the formula, then test it on a few rows where you already know the answer.

It drafts. You decide. That division of labour is not a caveat. It is the entire reason the tool is useful.

Three tasks that stay with a person

• Anything confidential in a public tool. A public AI assistant is one anyone can sign up for, and the text you paste in may leave your organisation the moment you press enter. Client details, staff records, unpublished results, commercial terms: these go only into a tool your organisation has approved for that purpose. If you would not email it to a stranger, do not paste it into one.
• Decisions about people. Hiring, promotion, performance ratings, disciplinary matters. These affect someone's livelihood, and they need judgement, context and a name attached to the outcome. The assistant can tidy the job description or help you structure a difficult conversation. The decision is yours.
• Final figures and legal wording, unchecked. An assistant will produce a confident, fluent sentence that is simply wrong, and it will not warn you. Any number that will be reported, and any clause that will be signed, is reviewed by someone qualified before it goes anywhere.

The OECD AI Principles and the NIST AI Risk Management Framework both expect people to remain accountable for outcomes, with human oversight in proportion to the stakes. High stakes, close look. Low stakes, quick glance. The seven-and-three split is that principle applied to a Monday morning.

The fault is almost always in the request

Here is what people get wrong. When the output is poor, they blame the tool. Nearly every time, the request was "write something about the budget", and the assistant did exactly that: something, about the budget, for nobody in particular.

A good request, sometimes called a prompt, covers five things.

• Role: who it should write as. "You are an experienced project manager writing to a client."
• Task: the one thing you want done. "Draft a short update on the delay to phase two."
• Source material: the facts, pasted in or attached. Leave this out and the assistant fills the gaps with plausible guesses.
• Format: the shape of the output. "Three short paragraphs, then a bulleted list of next steps."
• Constraints: what to avoid or include. "No technical jargon, under 200 words, do not promise a new date."

Picture a finance team explaining a budget variance to department heads. With all five elements, the first draft is usable. Without them, it is a generic page that gets rewritten from scratch.

If the first attempt misses, say what is wrong and go again. "Shorter, and lead with the decision" is a perfectly good follow-up. Treat it as a conversation, not a vending machine.

What you do before you press send

Read the whole thing, not the first paragraph. Check every name, number, date and claim against your source material. If you cannot verify a claim, cut it.

Then make it sound like you. Assistant prose is smooth, polite and faintly anonymous, and readers notice when every message from you arrives in the same voice. The draft saves you the first twenty minutes. The last five are still yours.

Finally, have one conversation with your team and write down the answer. Which tool is approved. What data never goes in. Who signs off on anything that leaves the building. Most of the benefit, and almost all of the protection, comes from that conversation.

The assistant will write whatever you ask. Your name goes on it either way.

Learn this in a course

• Prompt Engineering for Generative AI and AI Agents: a practical method for writing requests that produce reliable, usable drafts across everyday tasks. Five days, classroom or live online.
• Generative AI Essentials for Business: a clear grounding in what AI assistants can and cannot do, and how to use them safely at work. Five days, classroom or live online.