AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people respond to "AI will change your job" in one of two ways. They panic-collect tools, or they wait for a verdict from a headline. Both feel like action and neither tells you anything about your own week of work. The more useful question is narrower and answerable: which of your tasks are getting cheaper, which are getting more valuable, and what can you show for the second group?
The rule: stop asking whether AI will replace your job, and keep a written record of which tasks you hand over, which you direct, and which you own, then build evidence on the last two.
Be careful with forecasts, including the ones below. They describe averages across large groups, not you.
None of this tells you who "wins". It does tell you where to look: at tasks, not job titles.
The playbook has four moves. The first is a table you fill in. The other three are what you do with it.
List what you actually do in a normal fortnight, 10 to 20 tasks, using your calendar and sent items rather than your job description. Sort each one into a column.
| Column | Meaning | Example (hypothetical marketing manager) | What to do |
|---|---|---|---|
| Hand over | Repeatable, low stakes, easy to check | First-draft meeting notes, reformatting reports | Delegate to AI now; keep a spot-check habit |
| Direct | AI drafts, you decide quality and direction | Campaign copy variants, research summaries | Get fast at briefing, reviewing and correcting |
| Own | Needs trust, context, accountability or a signature | Budget calls, client relationships, hiring | Protect your time here and deepen it |
Re-do the split every quarter. A task that sat in "Direct" last year may belong in "Hand over" now.
Most of the near-term gain sits here. Directing means giving clear context, defining what good looks like, and catching errors you can only spot because you know the work. If you can't tell when the output is wrong, you are not directing, you are copying. Practise on real tasks from your Direct column and note each time the model got something wrong that you caught.
Own tasks share a feature: someone has to be accountable, and someone has to trust a person. That means judgement under uncertainty, relationships, domain context that isn't written down, and decisions with consequences. Spend deliberate time here: take on the decision rather than the preparation, sit in the client conversation rather than summarising it.
Skills you can't show are skills nobody will pay for. Each month, write one line per piece of evidence:
A year of these is a portfolio, an interview answer and a promotion case. It is also the honest way to say "I use AI well" without hand-waving.
Use an AI model as a sparring partner for the Task Split, not as an oracle. Ask it to argue against your sorting and to say what it is guessing.
Role: You are an experienced hiring director in [YOUR_FIELD], assessing candidates in 2026. You are candid and you do not flatter.
Context: Here is my role and how I really spend my time over a normal fortnight: [ROLE_AND_LIST_OF_REGULAR_TASKS_WITH_APPROX_HOURS]. These are the tasks I think are repeatable: [TASKS_I_THINK_ARE_REPEATABLE]. My employer or sector: [SECTOR].
Goal: Help me sort my tasks into three groups: Hand over (AI can do it reliably with spot-checks), Direct (AI drafts, I judge), and Own (needs accountability, trust or tacit context).
Steps:
1. Place every task I listed into one of the three groups, with a one-line reason each.
2. Name any task where you disagree with my own view and explain why.
3. Say which Direct and Own tasks are likely to become more valuable as AI tools improve, and why.
4. Describe three pieces of evidence a hiring director would want to see from me within 12 months, that I probably cannot show today.
Output: a table (task, group, reason, confidence), followed by a short list for step 4.
Constraints: Mark every claim that is a guess rather than something well established. If my task list is too vague to judge, ask me up to five clarifying questions before answering. Do not invent statistics.
Before answering, check that every task I gave you appears in the table exactly once.
Fill in: your field, your real task list with rough hours, the tasks you think are repeatable, and your sector.
A second prompt turns the Proof Log into something you can use.
Role: You are a careful career coach who writes in plain British English.
Context: My Proof Log entries for the last [NUMBER] months are below: [PASTE_LOG_ENTRIES]. I am preparing for: [PERFORMANCE_REVIEW_OR_INTERVIEW_OR_PROMOTION_CASE].
Goal: Turn the log into [NUMBER] short, truthful accomplishment statements.
Steps:
1. Group entries by theme.
2. For each theme, write one statement stating the task, my role versus the AI's, and the result.
3. Flag any entry with no measurable result and suggest what I could honestly record next time.
Constraints: Use only facts in my entries. Do not add numbers, tools or outcomes I have not given you. If something is missing, ask me.
Check before answering: every number in your output must appear in my log.
Give it 30 minutes to build the Task Split once, 10 minutes a month for the Proof Log, and one real Direct task a week as practice. Don't ask your employer's permission to think about your own tasks, but do check your company's policy before pasting client or personal data into any AI tool. Where your employer restricts tools, the Task Split and Own work still apply.
The trap is optimising for visible AI activity rather than judgement. Using a tool on everything makes you look current while moving your work into the "Hand over" column, where you are easiest to replace. The second trap is believing a flattering AI assessment. Models tend to agree with the person asking, which is why the prompt above asks for disagreement and flagged guesses.