Prompt Engineering for Real Work

The book

Most people who use AI tools write their prompts the same way they would write a search query, then wonder why the output needs three rounds of fixing. Prompt Engineering for Real Work is built around a single premise: the gap between a mediocre result and a useful one is almost always in how the request was written, not in the model. Marcus Hale works through sixteen areas — from writing a role-and-task prompt for the first time to managing a 100,000-token document — with real prompts and real outputs on every page, including the ones that failed. He is honest about what models still cannot do reliably, which makes the practical advice easier to trust.

What you'll learn

• How the five-part framework — role, task, context, format, and constraints — removes the most common prompt failures before they happen
• Why concrete, specific language produces better output than abstract instructions, and how to test the difference yourself
• When to show the model worked examples and when a clear description is enough
• How to feed a long document into a model without losing coherence or burying the question
• How to read a bad output and decide whether to refine the prompt or start over from a different angle
• How to build a personal prompt library that grows more useful the longer you work with it

Who it's for

Any knowledge worker who has opened a chat interface, typed a reasonable question, and felt quietly let down by what came back — whether that person works in marketing, law, finance, operations, or research. This is the book to reach for on the day you decide the tool is worth learning properly.