Short answer: An AI assistant drafts and rewrites text, summarises long threads and explains documents you paste in. It also turns messy notes into tables or CSV, writes spreadsheet formulas from a plain description and builds plans around your real constraints. It works fastest on one-off chores. You still supply the facts, the approval and the send button.
Thirty scattered email replies can become one clean guest list in about ninety seconds. You paste the thread, ask for a table of names and dietary notes, then count the rows. That is the whole shape of everyday AI automation. There is no code, no setup and no schedule to maintain.
For decades, automating a chore meant writing a script. The chore had to repeat often enough to repay the hour spent coding it. Most people, most of the time, therefore automated nothing at all.
An assistant changes that arithmetic completely. A one-off automation now costs a single sentence, so disposable tools finally make sense. You describe the job, use the result once, and throw the tool away.
What follows is an inventory rather than a manifesto. It names the jobs a general-purpose assistant handles well today. It also names the checks that take under a minute, and the work that should stay on your own desk.
The five-minute test for handing over a chore
Not every task deserves delegating. One question sorts them quickly. Can you describe the finished result in two sentences? If the answer is yes, an assistant can usually produce a workable first version.
The right mental model is delegation, not magic. You are the editor and the assistant is a very fast junior colleague. Its drafts arrive in seconds and are mostly right, which is a different thing from being finished.
Three signs a chore is a good fit
- The output is words or structure. Drafts, summaries, tables, lists, formulas and file formats all sit inside a model’s core skill, which is predicting text.
- You already hold the facts. When you paste the source material, the assistant reshapes what you gave it instead of inventing anything.
- A mistake would be visible. A missing row, a broken formula or a wrong tone shows up within seconds of reading.
Three signs to keep it yourself
- The task needs today’s facts. Prices, opening hours and stock levels change after a model’s training cutoff, the date on which its learning stopped.
- The stakes are legal, medical or financial. Use the assistant to explain the vocabulary, then take the decision to a qualified person.
- Sincerity is the message. Condolence, apology and praise carry weight because you wrote them yourself.
The brief matters more than the tool you pick. A one-line request returns an average answer, while a short brief naming audience, length and format returns something usable. The four parts worth including are laid out in writing prompts that get better answers.
Inboxes and documents: the largest daily win
Most people spend more time on written admin than on any other desk task. This is where an assistant repays attention first, because every step is text in and text out.
Turning a long thread into decisions and actions
Paste a forty-message thread and ask for three lists. You want decisions made, actions owed with a named owner, and questions still open. The result is a working summary rather than prose, and it reads in half a minute.
The same pattern converts a recorded meeting into minutes once the audio has been transcribed. Accuracy in the transcript sets the ceiling for everything after it. That is why our notes on meeting transcription for teams begin with the microphone rather than the model.
Ask for the summary in the shape you will reuse. A table of owner, task and due date drops straight into a project tracker. A numbered list suits an email. Naming the destination saves a whole round of editing.
Drafting a reply you would actually send
The blank page costs more time than the typing does. Ask for three versions of a difficult reply: firm, neutral and warm. Pick one, edit it, and send it in your own words.
Tone work is the most dependable request of all, because the facts are already yours. Ask for half the length, plainer vocabulary or a softer opening line. The federal plain language guidelines are a useful standard to name inside the prompt itself.
Tip: Paste two or three of your own past emails and ask the assistant to match that voice. One real sample beats a paragraph of adjectives such as professional or friendly.
Reading a document you did not write
Rental agreements, insurance terms and school letters share one problem. They are written for the writer’s protection, not the reader’s speed. Paste the text and ask for obligations, deadlines and costs as three short lists.
Then ask for each unfamiliar term to be explained in one sentence. Keep the original open beside the summary and match every bullet point back to a clause. That habit turns a plausible summary into a checked one.
Turning mess into structure: tables, CSV and formulas
Format conversion is the least glamorous category and the biggest time saver. It is pure structure, and structure is exactly what language models handle well.
From pasted notes to a spreadsheet column
Paste a block of scribbled notes with names, emails and dates buried in prose. Ask for a table with named columns, or for CSV, the comma-separated values format defined in RFC 4180. CSV pastes cleanly into Excel, Numbers and Google Sheets.
The same trick moves text between formats. Ask for Markdown for a wiki, HTML for a web page, or plain text stripped of stray formatting. Naming the target format is often the entire prompt.
Formulas described in plain English
Describe the outcome instead of the syntax. Say average column B, but only where column A reads Complete, and you get an AVERAGEIF formula with each argument explained. It works in reverse too. Paste an inherited formula and ask what it actually does.
Check anything unfamiliar against the documented spreadsheet function list. Assistants also write and explain regular expressions, the search patterns that find text by shape rather than by exact wording.
Planning against constraints you actually have
A plan is structured writing, which is why assistants are unexpectedly good at producing one. You supply the constraints and the assistant supplies the shape. Honest constraints produce usable plans.
- Meals for a week. Give the real limits: thirty minutes on weeknights, one vegetarian day, and whatever already sits in the fridge. Ask for a shopping list sorted by supermarket aisle.
- Travel days. A three-day itinerary grouped by walking distance, built around your pace and budget. Verify opening days and ticket prices yourself.
- Study and training. A syllabus or a 10k programme spread across the weeks before a fixed date, with revision passes and rest days built in.
- Events and run-sheets. Timings, a draft seating plan, and the sequence of reminder messages nobody enjoys writing.
- Household admin. The dispute letter for a wrong bill, the note to neighbours about building work, the instructions for a house-sitter.
Ask for the reasoning alongside the plan. A meal plan that explains why Tuesday is the slow-cooker night is easier to adjust than a bare list. You can then move one item without unpicking the rest.
Every plan is a bet about your future attention. The discipline of budgeting hours is far older than software, as we argue in what horology teaches about time management. An assistant does not supply that discipline. It only cuts the drafting time from an evening to a minute.
How to check the work in under a minute
Delegation without checking is not automation. It is guessing. Each category has a check that costs seconds, and matching the check to the task is the whole skill.
| Task | What you supply | What comes back | Your one-minute check |
|---|---|---|---|
| Thread summary | The full message chain | Decisions, actions, open questions | Confirm every action names an owner |
| Draft or rewrite | Audience, length, tone, source text | A first version to edit | Read it aloud once before sending |
| Notes into a table | The messy text block | A table or CSV file | Count rows in against rows out |
| Spreadsheet formula | A plain description of the result | The formula and an explanation | Test it on a column you already know |
| Plan or itinerary | Budget, dates, pace, constraints | A day-by-day schedule | Verify prices and opening hours |
| Document explained | The contract or policy text | Plain-language bullet points | Match each bullet to a clause |
Two failure modes deserve names. The first is drift, where rows are quietly dropped from a long list. Counting catches it immediately.
The second is confident invention, where a model states a plausible fact that is simply false. Where that risk concentrates, and how to spot it, is the subject of why chatbots make things up.
Build the check into the request itself. Ask the assistant to state how many rows it produced, or to flag any entry it was unsure about. A model that marks its own uncertainty gives you a shorter list to inspect.
Free tiers handle almost everything described in this article. Long documents, heavy daily use and file uploads are where paid plans begin to matter. That trade is weighed in free versus paid AI tools.
What stays on your own desk
Automate the drafting. Never automate the deciding.
Some tasks could be handed over and still should not be. Draw the line before you are in a hurry, because convenience always argues well in the moment.
- The send button. An assistant drafts and you dispatch. Nothing leaves in your name unread.
- Binding commitments. Contracts, tax figures and medical choices belong with a qualified human being.
- Words that carry a relationship. The effort is the message, and a borrowed draft quietly spends it.
- Claims you cannot verify. Unchecked facts mean the work is still a draft owed its review.
- Sensitive data. Passwords, identity numbers and other people’s details do not belong in a prompt, a boundary we set out in responsible AI use.
A week of ten-minute experiments
Reading about this teaches less than trying it. Give the habit five short sessions and judge it on the results rather than the novelty.
- Monday. Have one awkward email rewritten in three tones, then send your own edit of the best version.
- Tuesday. Turn a page of handwritten notes into a table with named columns.
- Wednesday. Describe one spreadsheet calculation in words and test the formula on a column whose answer you know.
- Thursday. Ask for a week of dinners under your real constraints, finished with an aisle-sorted list.
- Friday. Reduce your longest email thread to decisions, actions and open questions.
What a week like that teaches is calibration. You learn where drafts arrive excellent, where they arrive merely decent, and where they mislead. No article can hand you that feel.
Tip: Keep a plain text file of prompts that worked, with a one-line note on each. Reuse beats reinvention, and a saved brief is the closest thing to a script you will need.
Key takeaways
- One-off automation is now rational. If you can describe the finished result in two sentences, it is worth handing over once.
- Words are the deepest well. Drafts, rewrites, tone shifts and thread summaries deliver the most time back per minute spent.
- Structure work is the quiet superpower. Messy text into tables or CSV, and formulas written from a plain description.
- Every task has a one-minute check. Count rows, test formulas on known columns, verify real-world details yourself.
- Keep decisions on your desk. Approval, commitments, sincere messages and sensitive data stay with you.
Calibration is the durable skill here. Models will improve and features will shift, but the loop stays the same: describe, receive, check, correct. Practise the loop wherever you already work, a no-cost account at ASKAI.FREE included, and it transfers to whatever arrives next.
The reward is not doing more things. It is returning your attention to the work only you can do. Our using AI well topic page and the wider AI Assistants hub both push that idea further.
