A working method
AI is a technology, not a product. Most people use it like a vending machine — paste a prompt, take what falls out, blame the tool when it's wrong. The work is one level deeper: in how you use it, not which one you pick.
These are the five moves we teach in our workshops and apply in every system we build. Each one has a real demo behind it.
Ask the tool about its job before you ask it to do work.
Type the task into ChatGPT and hope. When the output's wrong, blame the tool.
Claude told us exactly what data it would need to draft those emails — five fields. Those five fields became our form. The output was right on the first call.
“If you think technology will solve your problem, you don't understand your problem and you don't understand technology.” — Roger Needham
Tell the model what NOT to do. That's the sharpest instrument you have.
Listing what you want and hoping the model figures out the rest. It defaults to whatever's most common in its training — which is exactly what you don't want.
We told Claude what NOT to do (don't pitch product, don't name-drop, don't lean on jargon) before we told it what to do. The 'don'ts' list is the headline output, not a footnote.
Headings and bullets in your prompt. The shape of the response mirrors the shape of the prompt.
Asking in one long paragraph and getting one long paragraph back. Then re-prompting to add structure. Then re-prompting again.
Named output slots. Fixed counts. Consistent labels. The model fills the slots — every time, in the same order.
Purpose. Audience. Scope. Tone. Four questions to ask before you type.
Skipping straight to 'write me an X about Y'. Getting something generic. Editing for an hour.
The four PAST questions become labelled output fields. Generic prompts become tailored briefs because the framework forces the right inputs.
PAST stands for Purpose, Audience, Scope, Tone — the free framework behind this move. Read the PAST Framework →
The data layer is a variable too. Clean inputs beat clever prompts.
Sweating the prompt while the model has no context. Spending hours tweaking words when the real lever is what you put on the input side.
Same model. Same five seconds. Add your style, your inventory, your data — the output becomes a working advisor instead of a vending machine.
We build personalised AI tools for solopreneurs and small operators who already know their work — we just turn the stuff they do well into systems that compound. Start with our /discover brief — it takes ten minutes, and you'll get back a Business DNA Brief naming your highest-leverage move.