The Playbook

Don't get replaced by AI. Become the one they can't lose.

By Dan Donovan, @askdandonovan

AI isn't a free replacement for people. It's useless without someone who can direct it and stand behind what it makes. Here are the five moves that make that person you, each with a copy-paste prompt or a concrete step you can run today.

5Moves
~7 minRead
$0Cost
No codeNeeded

First, the reframe. AI doesn't replace people. It replaces tasks. It can spit out a hundred options in a second, but it can't tell you which one is any good, it can't own the result, and it can't be the person your team actually trusts. That gap is where you become irreplaceable. Not by out-typing the machine, but by becoming the one who directs it and stands behind what it produces.

Do this in the next 10 minutes

Run one real task through a proper brief

Pick something you'd normally write yourself this week (a team update, a client email, a summary). Instead of a one-line ask, paste this brief and fill in the blanks. This one prompt is Move 1 in action:

Copy-paste brief 
ROLE: Act as an experienced [your job title].
TASK: Help me write [the thing, e.g. our monthly team update].
CONTEXT: Here's last time's version: [paste it]. Here's what's changed this time: [3-5 bullets].
FORMAT: Match the tone of the last one, about [200] words, and end with the top 3 priorities.

Then list 3 things you weren't sure about, so I can fill them in.
The 5 moves
1. Orchestrate, don't operate
Brief the AI properly, don't do the task
2. Build real taste
3 questions to judge any output
3. Own the output
The 5-point fact-check checklist
4. Be the first call
The 5-minute pre-meeting move
5. Put judgement where it counts
Make one call you'd normally push up
1

Orchestrate, don't operate

The lowest-value thing you can do now is the task itself. The highest is directing it. The whole skill is the brief: a good one has four parts (Role, Task, Context, Format). That's the backbone of every serious prompting framework, and it's the difference between beige filler and something you'd actually send.

❌ What most people type"Write our monthly team update."

That gets you generic mush. Here's the same request, briefed properly (this is the prompt from the box up top, so steal it):

✅ The 4-part brief 
ROLE: Act as an experienced [your job title].
TASK: Help me write [the thing].
CONTEXT: Here's last time's version: [paste it]. Here's what's changed: [3-5 bullets].
FORMAT: Match the tone of the last one, about [200] words, end with the top 3 priorities.

Then list 3 things you weren't sure about.
First move: take one task you'd normally grind through today. Don't do it. Run it through the brief above, then direct and review. You just changed jobs.
2

Build real taste

AI hands you a hundred versions in a second. It can't tell you which one is good. Taste is that call, and it's yours alone. You don't need to be born with it. You need three questions you run on every output before you use it:

  1. Would I have written this myself, or does it sound like a robot?
  2. What's the one line I'd cut if my boss was reading it over my shoulder?
  3. Where is it generic? Could this exact sentence sit in anyone's version?
First move: on your next AI output, run those three questions, then cut it down to only the parts that pass. What's left is the good half. That's taste.
3

Own the output. Trust, then verify.

AI is confident and often wrong. A 2026 Stanford index clocked hallucination rates anywhere from around 20% to over 90% depending on the model and task. You don't need to be technical to catch it. You need to know where it lies. Scan every AI draft against these five:

  • Fake sources & links. Any study, citation or URL. AI invents references that look real. Click every link; if a source or person doesn't show up anywhere else, it's made up.
  • Invented numbers. Specific stats and percentages. If it didn't come from something you gave it, treat it as a guess.
  • Wrong names, titles & dates. Stated confidently, often wrong. Double-check anything factual before it goes out.
  • Too much confidence. A flat, categorical claim on something genuinely uncertain usually means it's filling a gap with an invented answer.
  • The 3-way test. Ask the same question three different ways. If the details change between answers, it's guessing, not recalling. Go verify it.
First move: on your next AI draft, run the checklist and find the one thing it got wrong before anyone else spots it. That catch is the whole job.
4

Become the first call

You don't need a huge network. You need a small circle who trust your judgement. The fastest way to earn it: walk into the room already having done the thinking. Five minutes before your next recurring meeting, run this:

5-min pre-meeting prompt 
Here's the data/notes for my [weekly team meeting]: [paste it].
Pull out the 3 things actually worth flagging, and for each, one plain-English sentence on why it matters. No jargon.
First move: do this before your next meeting and walk in with the three things. Be the one who already looked into it.
5

Put your judgement where the stakes are real

Judgement is a muscle, and it only grows under real weight, but it doesn't have to be scary to start. You don't need to volunteer for the huge risky project. You need one small call that's normally not yours to make.

First move: this week, find one small decision you'd usually push up to your boss, and just make the call yourself. Own the outcome. That's the muscle, started safely.
The one-line version: everyone can type a prompt. Almost nobody can brief the machine properly, tell good from noise, catch what it gets wrong, and be trusted with the result. Be that person, and you don't get replaced. You get promoted.