Between chapters 3 & 4
The Workshop
Everything you've learned, one conversation away.
Stop one-shotting.
By this point you have a toolkit: building blocks, the Socratic method, context engineering, system prompts, tokens.
In practice, good results almost never come from a single prompt. The people who get the most out of AI treat it like a collaborator. You share a goal, the model takes a swing, you push back, it rewrites, you add context, it tries again. After four or five rounds, the answer starts to land.
Prompting is a conversation.
Every technique is a move.
Inside a conversation, each technique you've learned becomes a move you can make:
Open with the five building blocks: Role, Task, Format, Constraints, Examples. (Ch1)
When your goal is fuzzy, make the AI interview you before it starts building. That's the Socratic move. (Ch2)
When stakes are high, pack the context carefully, and think as hard about what to leave out as what to keep in. (Ch3)
When the problem needs reasoning, tell the model to think step by step. (more on this in a moment)
When the first draft isn't right, iterate. Nobody nails it on the first pass, including professionals.
All of these moves do the same job: they help you set up a conversation the AI can actually be useful in.
One more move
Chain of thought, briefly.
There's a famous prompting trick called chain of thought. You add the phrase "think step by step" to your prompt, and the model's accuracy on hard problems jumps noticeably. The model hasn't gotten smarter. It just has scratch paper now, and it's using it.
The trick worked so well that researchers automated it. The reasoning models you've probably heard about (Claude's Extended Thinking, OpenAI's o-series, DeepSeek R1) are basically chain of thought built straight into the model. They think step by step before they answer, without you having to ask.
Which raises a good question. If the model already reasons on its own, what does the human still bring to the table?
The conversation. Chain of thought proved that more thinking makes for better answers. The same rule holds one level up: more turns of back-and-forth make for even better ones. A model reasoning on its own is useful. A model reasoning together with you is usually where the real quality comes from.
The workshop below is a place to practice that.
Your turn
The Ultimate Prompt Builder.
Pick a recipe on the left, turn on the moves you want to use, and work with the AI turn by turn until you're happy with the result.