Be Careful What You Wish For
a wish granted exactly, literally, as worded — and never as generously as intended — is the oldest cautionary tale about instructions, and it's exactly the problem prompt engineering solves.
Part 1
Part 2
What the Genie Actually Grants
the real difference between a prompt that merely gets a response and one that gets the response you actually needed.
Part 3
Before There Was a Genie to Ask
how instructing a computer worked before natural language prompting existed, and why that history explains what makes prompting so genuinely different.
Part 4
The Wish Granted Too Literally
how a single ambiguous word or unstated assumption in a prompt can send a model confidently down the wrong path entirely.
Part 5
Three Wishes, Clearly Numbered
how giving a prompt real structure — separated instructions, context, and format — reduces the ambiguity that causes so many wishes to go wrong.
Part 6
Showing the Genie an Example Wish
why demonstrating exactly what you want, rather than just describing it, is often the single most reliable way to get a precise result.
Part 7
Asking the Genie to Think Out Loud
how asking a model to reason through a problem step by step, out loud, often produces a meaningfully more accurate final answer.
Part 8
The Genie's Job Description
how a system prompt sets a model's persistent role, tone, and constraints before any specific user request even arrives.
Part 9
One Wish at a Time
why breaking a complex request into a sequence of smaller, focused prompts often produces a more reliable result than one giant, sprawling wish.
Part 10
A Wish With Room to Improvise
how tightly or loosely you constrain a prompt's output changes what kind of response you actually get, and why that's a genuine, deliberate design choice.
Part 11
The Genie's Rulebook
why explicitly stating what a model should not do is often just as important as stating what it should — and easy to forget entirely.
Part 12
When the Genie Misunderstands the Wish
a systematic way to diagnose exactly why a prompt failed, rather than randomly rewording it and hoping for the best.
Part 13
Testing the Wish Before It's Final
why a prompt that worked once on one example is a genuinely different, much weaker claim than a prompt that's been systematically tested across many.
Part 14
A Wish That Works for Every Genie
why a prompt carefully tuned for one specific model can behave surprisingly differently on another, and what that means for writing portable prompts.
Part 15
The Fine Print in the Wish
how specifying an exact output format — like structured JSON — turns a model's response from free-flowing prose into something a program can actually use.
Part 16
Wishing for the Wrong Thing on Purpose
how the same techniques that make a prompt more effective can be deliberately misused to manipulate a model into unintended behavior.
Part 17
A Library of Proven Wishes
why building a reusable collection of well-tested prompt templates saves real effort and avoids repeating the same mistakes over and over.
Part 18
Grading the Genie's Performance
how to move beyond a gut feeling that a prompt 'seems good' toward a genuine, measurable evaluation of how well it actually performs.
Part 19
The Wish-Granting Committee
why prompt engineering at real organizational scale needs the same collaborative discipline as any other shared, evolving codebase.
Part 20
The Last Wish
reassembling the whole discipline, from a single carelessly worded wish to a carefully engineered, tested, and maintained request that reliably gets exactly what's needed.