AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people sort themselves on AI by mood rather than by task: one week nothing goes near a chatbot, the next every email and half-formed thought does. Both feel fine because one sounds principled and the other productive. Neither gives you a deliberate line between the work you hand over and the work that stays yours.
The rule: decide task by task what AI does and what you keep, write it down, and change one thing at a time.
In late summer 2026 the writer Dan Koe published a long X article titled "We are in the middle of the digital renaissance, please take advantage of it". In it he describes three groups forming around AI:
| His label | What it looks like in practice | The cost |
|---|---|---|
| Purists | Reject AI outright and retreat to simpler, screen-light living | Opt out of real leverage, and tend to treat the whole thing as a threat rather than a tool |
| Automatons | Run every decision and thought through a model; optimise everything; avoid friction | Lose track of what they actually think, and their output starts to sound like everyone else's |
| The third group (he frames them as polymaths of a "digital renaissance") | Use AI heavily for leverage while keeping their own taste, judgement and inner life in charge | Takes deliberate effort and the discipline to say no to easy automation |
His argument is that the third group pulls ahead and the first two are the defaults people slide into. You do not have to buy the whole essay to find the split useful. What follows is our own read, turned into something you can run this week.
Skip the personality quiz. Answer these four questions about your last five working days:
| Question | Leans Purist | Leans Automaton |
|---|---|---|
| When you last got stuck on a decision, what did you do first? | Avoided AI on principle, even where it would have helped you see the options | Asked a model before forming your own view |
| Could you explain, without notes, the reasoning behind your last AI-assisted piece of work? | — | No, or only by rereading the chat |
| Is there a dull, repeatable task you still do entirely by hand? | Yes, several | — |
| Do you edit AI drafts until they sound like you, or ship them roughly as they came out? | — | Ship them roughly as they came out |
Most people are a mix: Purist in one area, Automaton in another. The ledger below works at the level of tasks, not identity, so the mix stops mattering.
If you are going to read the original, read it as an argument, not a manifesto.
The mechanism is two lists and one change.
List A: Hand off. Tasks where a model's output is easy for you to check and a mistake is cheap: summaries of documents you will still skim, reformatting, tidying meeting notes, generating options you will choose between, spotting gaps in a checklist.
List B: Keep. Tasks where the value is your judgement or taste, or where you could not easily tell if the model got it wrong: deciding what to work on, final calls involving money, people or health, the opinion at the heart of a piece of writing.
The one change. Move exactly one task this week: a dull task from "done by hand" into List A, or a lazily outsourced task back into List B. One, not five, or you cannot see the effect.
Run it like this:
This prompt does the first sort for you. Paste it into whichever assistant you use and fill in the brackets:
You are helping me decide which of my work tasks to hand to an AI assistant and which to keep doing myself. The goal is to use AI heavily without outsourcing my judgement or taste.
My role: [YOUR_ROLE]
Tasks from my last working week, one per line: [LIST_OF_TASKS]
Areas where mistakes would be costly for me (money, clients, health, legal, reputation): [HIGH_STAKES_AREAS]
Steps:
1. For each task, judge (a) how quickly I could check an AI's output for errors and (b) how much harm a mistake would cause, using only what I have told you.
2. Put each task in one of three groups: "Hand off", "Hand off with a review step", or "Keep".
3. For every "Hand off with a review step" task, name the specific check I should do.
4. Suggest one single change I could make this week, and say why that one.
Output: a table with columns Task | Group | Reason (one line) | Review step if any, then the one suggested change in two sentences.
Rules: if a task is too vague to judge, ask me about it instead of guessing. Do not move anything involving final decisions about money, people, health or legal matters into "Hand off". Before answering, check that every task I listed appears exactly once in your table.
Fill in your role, your real task list and the areas where a mistake would hurt.
Imagine Priya, a freelance bookkeeper: a Purist at work, an Automaton at home, where a chatbot plans her meals and holidays unchecked. Her tasks sort like this:
Her one change: invoice chasers move to List A. On Friday it has saved time and needed only light edits, so it stays. Next week's change goes the other way: she shortlists holidays herself and uses AI only to compare her two favourites.
| Snag | What is going on | Fix |
|---|---|---|
| Everything ends up in List B | Purist instinct wearing a "quality" costume | Force at least three entries into List A, even small ones |
| Everything ends up in List A | Speed feels like progress | Ask "could I defend this decision without the chat open?" for each item |
| Your writing starts to sound generic | AI drafted the opinion, not just the wording | Write the core point yourself first, in a sentence, before asking for help with the rest |