Based on real working patterns

The Top 5%
AI Playbook

10 habits that separate people who get real work done with AI from everyone else. No jargon. No fluff. Just the things almost nobody does.

Scroll
01

Tell AI what is broken before you start

Most people let AI waste 30 minutes chasing a problem that doesnt exist because they forgot to mention something is already broken. Tell it upfront what isnt working, what is missing, and what to avoid touching.

Try this: "Before we start, the file upload feature isnt set up yet. Dont try to test it. Work around it for now."
02

Upload old versions and ask what changed

Everyone uploads one file. Almost nobody uploads three versions and says "something broke between version 1 and version 3, find what changed." This turns AI into a detective that spots exactly what went wrong.

Try this: "Here are three versions of this document. Compare only section 4 across all three. Tell me what was added, removed, or reworded."
03

Ask AI to try to break its own work

After AI builds something, most people say "looks good" and move on. The top 5 percent say "now tell me every way this could fail." Its a free quality check that catches problems before they cause real damage.

Try this: "Before I use this, try to break it. What happens if the data is empty? What if someone runs this twice by accident?"
04

Keep a cheat sheet of decisions and paste it in every time

Most people re-explain their whole project every session. Keep a short document with your key decisions, tools, and constraints. Paste it at the start of every new conversation. It turns a cold start into a warm handoff.

Try this: "Here is our current setup: we use Mailchimp for emails, Stripe for payments, our brand colours are navy and coral, and we never discount below 20%."
05

Define exactly what "done" looks like before you ask for anything

"Write me an email" is a wish. "Write a 150-word email to our supplier, apologetic tone, referencing order 4821, asking for a replacement by Friday" is a brief. The second one gets a usable result first time.

Try this: Before asking AI to build anything, write one sentence that describes exactly what the finished thing looks like and how you will use it.
06

Ask for the same thing rewritten for three different people

Most people ask AI for one output. The top 5 percent ask for the same information rewritten for the technical team, then the boss, then the client. The AI already knows the content. Repackaging it costs almost nothing.

Try this: "Now rewrite that as: (a) a quick message to the team, (b) a 3-sentence email to the director, and (c) a one-liner for the Monday meeting."
07

Check if the explanation actually makes sense before you send it

When AI gives you a nice-sounding analogy or explanation, most people just use it. The top 5 percent ask "is this actually accurate? Where does it fall apart?" A confident explanation that is wrong will hurt your credibility.

Try this: "That explanation sounds good but if someone challenges it, where does it break down? Fix it or give me a better one."
08

Tell AI what you dont want, not just what you do

Everyone describes what they want. Almost nobody says what to avoid. "Dont mention pricing." "No corporate buzzwords." "Never use the word synergy." These rules stack up and the output gets noticeably tighter with every message.

Try this: Add a "do not" list to every brief. Phrases to avoid, topics that are off-limits, formatting that is banned.
09

Ask "what are we missing" before saying "go build it"

After AI proposes a plan, most people say "great, do it." The top 5 percent say "what have we not thought about? What breaks if this gets bigger? What would a critic point out?" This catches blind spots before they become expensive.

Try this: "Before we move forward, what are the three biggest risks in this plan and what would we need to change if demand doubles?"
10

Stay in the same conversation and keep building

The most wasted opportunity in AI is starting fresh every time. The 15th message in a focused session is dramatically better than the 1st message in a new one because AI has learned your style, your rules, and your standards along the way.

Try this: Never start a new conversation when you could continue the current one. Chain related tasks together like a working session, not a help desk.
Proceed with caution

Top 1% Tips That Give You
an Unfair Advantage

i

Ask AI to review you

Say "based on this session, what should I have told you earlier?" and watch it show you your own blind spots.

ii

Fact-check someone elses reply

Paste a colleagues response and ask "is this actually correct and where are they wrong?"

iii

Use AI to build other AI

Get it to write the instructions, personality, and rules for a separate AI agent you are deploying.

iv

Reverse-engineer things you didnt build

Upload someone elses work and say "trace through this, tell me what it does, and find the flaw."

v

Make AI write proof its own work works

Ask it to write the exact test steps a stranger would follow to verify everything it just built.

vi

Design your own failure tests

Ask AI how to deliberately break something so you can watch whether the safety net catches it.

vii

Rehearse conversations that havent happened yet

Use AI to pre-build the arguments, analogies, and defence you will need before the meeting even starts.

The AI isnt the bottleneck.
The brief is.

The people who get the best results from AI arent prompting wizards. Theyre people who know what they want and can explain it clearly. Thats it.

Get clear. Get specific. Get building.