A narrated, hands-on workshop · ~70 min · for anyone who talks to AI daily

Stop guessing. Start briefing.

Press play and I will sit next to you while we take prompts apart — the five levers that actually move an answer, the folklore that does not, and a bench for testing your own prompts instead of trusting your gut.

13 chapters · player controls live at the bottom

The Workbench · the hero demo

Pick a task. Both sides get the same job. Edit either prompt, run them together, and watch which levers actually moved the answer.

A · the prompt most people type

Press Run both.
Simulated output — composed in your browser by a rule engine reading your prompt. No model, no network.

B · the same job, briefed properly

Press Run both.
Simulated output — composed in your browser by a rule engine reading your prompt. No model, no network.

The rule engine looks for six things in your text: a role, a named reader, an output format, worked examples, hard constraints, and pasted context. Strip one out of B and run it again.

Read this once. Nothing on this page calls a language model. Every “output” you see is written by a small rule engine that inspects your prompt and assembles a stand-in answer, so the page works offline and forever. The shape of what changes is real and worth learning. The exact words a real model would say are not being predicted here, and I will not pretend otherwise.

Ch. 1 / The mental model ▶ narrated

You are not casting a spell. You are writing a brief.

Everything you type becomes one thing: the situation the model has to continue. Not a command it obeys — a context it completes. Which is why leaving something out is never neutral. It is an instruction to guess.

🧠

It is a continuation engine

It reads what is in front of it and produces what plausibly comes next. That is not a limitation to fight. It is the lever you pull.

so: change what comes before
🎣

Silence is a choice

Say nothing about length, reader or shape and you have not left it open. You have handed those three decisions to the average of everything ever written.

so: decide them yourself
🔄

Every answer is evidence

A bad output is not a verdict on the model. It is a readout of what your prompt failed to pin down. Read it like a diagnosis, not a disappointment.

so: repair, do not re-roll
The first-morning test. Read your prompt out loud as if you were saying it to a brilliant new hire on day one. If they would have to ask you a follow-up question before starting, that question is the thing missing from your prompt. Answer it and run again.

Ch. 2 / The craft, named ▶ narrated

Role, constraints, examples, format, iteration.

Five levers. That is the entire toolkit. Not five tricks — five kinds of information that were missing from your prompt, and each one closes off a different way for the answer to wander.

🎭

1 · Role

Not flattery. A role is a vocabulary and a priority. “A support lead triaging inbox mail” tells it what matters and what to skip. “A world-class genius” tells it nothing.

test: does it change the words?
🚫

2 · Constraints

The fence. What must not appear, what must be quoted exactly, what to leave out. Constraints are the only lever that removes things, which is why they fix waffle.

test: can you check it was obeyed?
📎

3 · Examples

One finished answer beats three paragraphs describing a finished answer. Chapter four is entirely this, because it is the highest-leverage thing in the whole workshop.

test: is it a real output, not a description?
📑

4 · Format

Table, JSON, five bullets, sixty words. The shape decides whether the answer is usable or whether you spend ten minutes reformatting it by hand.

test: could a script validate it?
🔁

5 · Iteration

The lever everyone forgets, because it is not a sentence you type — it is the habit of reading the bad answer as a bug report and repairing one thing.

test: did you change one thing, or all of them?
📜

+ Context

The sixth thing, sitting underneath all five: the actual material. Paste the email. Paste the receipt. Describing your document is not the same as giving it.

test: is the real text in there?

Demo · the lever inspector

Paste any prompt you have actually used. This reads it and tells you which levers are present and which questions you left the model to answer for you. No simulated output here at all — it is just reading your text.

You will not pull all five every time. A throwaway question needs none of them. The rule is simple: the more the answer matters, and the more times you will run it, the more levers earn their keep.

Ch. 3 / Your first real skill ▶ narrated

Take a weak prompt apart in ninety seconds.

This is the move you will use every day. Six questions, asked in order, each one adding a single line. You do not need to be clever. You need to be specific, in a fixed order, until the guessing stops.

01

Who reads it?

Name a human being with a job and a reason to care. Not “a general audience”.

02

What shape?

Table, bullets, JSON, one paragraph. Pick one, name the fields.

03

How long?

A number. Sixty words. Five rows. Three bullets. Numbers are checkable; “concise” is not.

04

What must not happen?

The one failure you keep getting. Ban it by name.

05

Show one right

Paste a finished example, even a rough one you typed yourself.

06

Where is the material?

Paste the actual text. Then say: use only what is below.

Demo · the rewrite bench

Pick a weak prompt anyone might have typed. Tick the levers one at a time and watch the rebuilt version assemble itself. Untick one and watch what you just gave back.

Before

After · 0 levers

Ninety seconds, not ninety minutes. You are not writing a document. Six short lines, most of them under ten words. The cost is trivial and the difference is the gap you saw at the top of this page.

Ch. 4–12 / Locked

That's the free preview.

You've heard the setup. The remaining 9 chapters are where The Prompt Workbench actually gets built — the guardrails, the hands-on builds, the full narrated walkthrough and the end-of-workshop check.

Get the full workshop → See the catalog

One file. Yours forever. Works offline, in any browser, on any device.

Transcript

Now playing
Ready — press play to begin
0.95×
🎙
Free preview