1.You have to break work down before you can delegate it
"Hand the preparation for this event over to AI" is impossible, but "draft the event notice," "group the participants' questions by type" and "make a checklist of supplies" can be delegated. The smaller you break the work down, the more clearly you can see which pieces AI does well and which a person must do.
Ask two questions about each piece. Are language patterns the core of this piece (drafting, summarizing, organizing, converting formats)? Would a mistake cause serious harm (amounts of money, dates, commitments, judgments about people)? Start by delegating the pieces where the answer to the first question is "yes" and to the second is "no," and the risk of failure is low while the time saved is large.
| Piece | Delegate to AI? | Why |
|---|---|---|
| Draft of the notice | Yes | Writing patterns are the core, and a person can revise it |
| Organizing participants' questions by type | Yes | AI is good at classifying and summarizing (remove personal information first) |
| Finalizing the venue and date | Person | It is a decision involving commitments and costs |
| Budget calculation | Support only | Use AI for the table layout, and a spreadsheet for the calculation and checking |
| Final send-out to participants | Person | A person does the final check before sending |
2.Four ways to collaborate
There are four broad ways to work with AI. Even for the same task, switching the approach to fit the situation improves the results.
- Drafter: the AI writes the first draft and a person revises it — this saves time in front of a blank screen. But decide your main points first so the draft doesn't steer you
- Reviewer: the AI critiques what a person wrote — have it find missing points, awkward sentences and counterarguments. This raises quality while keeping your own thinking intact
- Idea partner: it lays out many possibilities — candidate titles, solutions, lists of questions. Get many, and a person chooses
- Study helper: it explains concepts you don't know and gives you problems — covered in detail in Lesson 10
- Wrong: Write a proposal. (Then submit the AI draft almost unchanged.)Right: Write the three-line core argument of the proposal yourself first, have the AI polish the structure and sentences, then have it act as a reviewer to find weaknesses, and fix them.When a person holds the center of the thinking and the AI helps with expression and checking, the result is truly yours.
3.Turn repeated tasks into templates
For repeated tasks, such as a weekly report, replies to customer inquiries that arrive every day, or organizing the minutes after every meeting, making a prompt template pays off. A prompt you have refined well once produces results of the same quality every time, and you can share it with others.
A good template separates the parts that never change (role, format, constraints) from the parts that change every time (material, dates). Mark the changing parts with square brackets. If you tweak the template a little each time you don't like the result, within a few weeks you will have a template that barely needs any adjustment.
- The parts that don't change: role (a meticulous meeting secretary), format (three sections: decisions, action items, open issues), constraints (don't add anything that isn't in the minutes).
- The parts that change: [meeting date], [attendees' roles], [meeting notes].
- Fix the columns "owner (role) and deadline" for action items so the same table comes out every time.
- On the last line, add "List anything unclear separately as a list of questions" to bring out the points that need checking.
4.A checklist for reviewing delegated results
When a piece you delegated to AI comes back, don't use it right away; check it briefly. If you set the checklist in advance, you can review quickly by the same standard every time, and when several people work together, the quality of review stays consistent.
The key to checking is telling apart "what the AI newly created" from "what came from the material I gave it." For content from your material, you only need to check that it matches the original, but if numbers, dates, names or commitments that weren't in the material have appeared, check those first.
- Facts: do the dates, amounts, places and names match the original material exactly?
- New content: have any numbers, commitments or quotations you didn't provide crept in?
- Missing content: is anything that must be included (how to apply, deadline, contact information, etc.) missing?
- Tone and audience: is the tone right for the reader, and is there any wording that could be misunderstood?
- Information protection: has any personal information or internal material slipped into text that is going outside?
- Step 1: Compare it line by line with the notes: the admission fee, date, time and number of people match the notes.
- Step 2: Find the new content: the day of the week "(Sat.)," the "first-come, first-served" method of admission and "parking available" were not in the notes.
- Step 3: Check and decide: check the day of the week on a calendar yourself, and delete first-come, first-served and parking unless they were actually decided. Whatever is in the notice becomes a promise to participants.
- Step 4: Fix the template for next time: add "Don't add anything that isn't in the notes; if some information seems necessary, list it separately as questions at the end" to the prompt.
5.Automation and where people check
Once your prompt templates are stable, you can go a step further and chain several steps together. For example, if you build a flow like "classify inquiry emails → draft replies by type → human review → send," people only need to focus on reviewing. If the work tools you use have built-in AI features, such connections are relatively easy to set up.
The key to automation is deciding in advance "where a person checks." Always put a human check right before anything that goes outside (emails to customers, public posts), anything hard to undo (payments, deletions, contracts) and judgments that affect people (evaluations, selections). Consider letting a flow run all the way through automatically without a check only when the template is very simple and the harm from a mistake is small.
📌 Key points
- Only when you break work into small pieces can you see which pieces to hand to AI and which a person should do
- Start by delegating pieces where language patterns are the core and mistakes cause little harm
- Drafter, reviewer, idea partner, study helper — choose the approach that fits the situation
- Turn repetitive tasks into prompt templates that separate the fixed parts from the changing parts
- Put a human check before anything that goes outside, anything hard to undo and judgments about people
🤖 Try asking AI like this
Copy a prompt and replace the [ ] parts with your own situation. Don't take the answer on trust — check it against this lesson.
When you want to find the pieces of your work to hand to AI
My job is [description of your work]. Break this work down into steps and organize them in a table, and for each step mark "Good to hand to AI / AI for support only / A person should do it," with a reason. Also mark the points where a person should check.
When you want to make a prompt template for a repetitive task
Every week I do [the repetitive task]. Make me a prompt template for this task. Separate the parts that stay the same every time from the parts that change, and mark the changing parts with [square brackets]. Fix the output format too.
When you want a critical review of your draft
Below is a draft [type of writing] I wrote. Don't change my main argument, but find three gaps in the logic, three sentences with weak support and three likely counterarguments. [draft]
- General principles commonly found in guides to using generative AI at work
Reached every goal above? Mark the lesson complete.
Storage is unavailable in this browser, so this lasts only for this page.🤖 AI Basics
- 1What Is AI? — Rules and Learning
- 2Machine Learning Basics — Data, Model, Training, Evaluation
- 3Neural Networks, Intuitively — Small Calculations Add Up to Judgment
- 4How Do Large Language Models Produce Text?
- 5Prompt Basics — Saying Exactly What You Want
- 6Advanced Prompting — Roles, Examples, Steps, Format
- 7Hallucination and Verification — How to Doubt What Sounds Plausible
- 8Privacy, Copyright and Ethics — Using AI Responsibly
- 9Working with AI — Delegate, Check, Connect
- 10Learning Strategies for the AI Era — The Basics Shape Your Questions