How to failwith AI
Nobody remembers the twelve-step best-practice guide. Everybody remembers the mistake that cost them a client. So this site only teaches mistakes. 90 of them, written as instructions, each with the reason it works and the damage it does.
Every rule below is a real pattern, collected from teams, classrooms and one very long year of watching smart people burn credits. Follow them faithfully and your AI results will stay at zero. Recognise yourself in three of them and you already know what to change on Monday.
Six chapters, one appendix. No sign-up. No newsletter.
Chapter 00. Foundations
Applies to every tool you will ever openStart with the tool, never with the problem.
A problem tells you what good looks like. A tool only tells you what is possible.
Motion without progress. Busy quarters, flat results.
Expect the model to read your mind.
You hold the context: the client, the constraint, the history, the reason. None of it is in the prompt.
You blame the model for not knowing what only you know.
Never verify anything.
Verification is the only step that converts plausible output into usable output.
One public error, and every correct thing you published gets doubted too.
Keep no examples of good work.
Examples beat instructions. Three good samples teach more than three pages of description.
Every project restarts from the generic default.
Judge the whole field by one bad session.
One weak result usually means a weak brief or the wrong model, not a ceiling on what is possible.
Two years of standing still while your competitor iterates.
Hide the process from your team.
Shared prompts, shared failures and shared fixes are how an organisation actually gets faster.
Six people solving the same problem privately, badly.
The chapters
Text and chat
The written word, wasted. 19 documented ways to waste the effort.
→02Image generation
Pictures nobody remembers. 14 documented ways to waste the effort.
→03Video generation
Eight seconds of nothing. 11 documented ways to waste the effort.
→04Websites and web apps
Shipping the demo. 17 documented ways to waste the effort.
→05AI avatars and voice
A face that erodes trust. 10 documented ways to waste the effort.
→06Agents, automation and teams
Scaling the mistake. 13 documented ways to waste the effort.
→The prompt cemetery
Twelve real prompts that returned nothing worth keeping, each with a cause of death, what it left behind, and the version that would have lived. None of them died of an impossible request. All of them died of a vague one.
Visit the gravesThe fastest way to get good at this is to stop doing six specific things.
Pick one chapter. Read it with your last project open next to it. Count the rules you followed by accident.
Then invert the three that stung the most. That is the whole method, and it works faster than any course.
Open chapter 01