Forward-deployed / Learning zone
Prompt engineeringa standalone module
Lesson 01

What prompt engineering actually is

TL;DR

Prompt engineering is the craft of writing input a language model will reliably act on. It's not a magic incantation and it's not "asking clearly." It sits somewhere between programming and technical writing. A language model is a probabilistic system that generates the next likely token given the ones so far. Your prompt is the input that steers the probability distribution — small changes in wording shift the output in ways a search engine never would. The mental model to hold: you are not talking to a person, and you are not calling an API with a fixed contract. You are conditioning a statistical process. That framing changes what you write, why you rewrite it, and how you know it worked.

🎯 For the AI PM (or coding-agent user)

Why it matters — Whether the model returns useful output or plausible-sounding nonsense depends on the input you gave it, more than the model you picked. Treating prompt writing as an afterthought is treating your product's quality as an afterthought.

What it changes in your decisions — You stop debating models before you have debated prompts. You treat prompt iteration as real design work with a version history, not a one-off keystroke.

Ask yourself — "If someone else wrote this prompt tomorrow, would they get the same output — and do I know why?"

Risk if ignored — You blame the model for a prompt problem, swap providers looking for relief, and rediscover the same failures with a different logo.

The mental model

The mental model

You are conditioning a statistical process, not talking to a person

Every arrow is a place you can lose quality — and a place you can steer it back.

Your intent — what you want
Prompt — what you type
Language model — probability over next token
Output — one sample from that distribution
↺ reread, rewrite, steer the distribution — the output feeds back into the next prompt

The gap between the first box and the last box is where prompt engineering lives. Your intent is clear in your head. The prompt is a lossy translation. The model is a sampler over possible continuations, not a search engine looking up the right answer. The output is one draw from that distribution. Every arrow in the diagram is a place you can lose quality, and every one is a place you can steer it back.

Three habits that separate a good prompt from a lucky one

What "engineering" actually adds

Three things distinguish engineering from typing:

The rest of this module is the specific craft. Lesson 2 breaks a prompt into its six named parts. Lessons 3 through 6 climb the ladder from everyday productivity into techniques power users depend on. Lessons 7 and 8 cover what changes when the model can act — call tools, run code, edit files. Lesson 9 is the diagnostic playbook when something goes wrong.

Failure modes

Practitioner checklist