Why our AI setups work better than a prompt
Most people type a question into an AI and hope. We set the AI up first.
A prompt is a one-off. You explain your job, your tone and your rules, you get an answer, and tomorrow you explain it all again. The answer is only ever as good as what you remembered to type.
Our packs are built differently, in three layers.
A project that remembers. Your role, your customers, how you write. Set once.
Helpers that each do one job. Chasing an invoice and writing a tender are different jobs, so they get different files. A router reads what you paste and picks the right helper. You never have to name one.
Rules that hold. Every helper knows what it does, what it won't touch, and which helper takes the jobs it won't touch. No helper can overrule the router.
What we follow
The companies that make these AIs publish guidance on how to set them up properly. Hardly anyone selling AI packs follows it. We do.
- Short files beat long ones. The AI reads only what today's job needs, so nothing important gets buried.
- Tight where a mistake costs money, loose where judgement helps.
- A worked example in every helper, because an AI copies an example better than it follows a description.
- A check before the answer comes back, and a plan for when the other person pushes back, goes quiet or escalates.
- One pack, built to install in ChatGPT, Claude, Gemini, Copilot and Grok.
What we keep to ourselves
We don't publish the wording inside the files, how the router decides, or the order the checks run in. That took the longest to get right, and it's what you're paying for.
Don't take our word for it
Open a full pack, free. Paste in a real message from your week, then ask plain ChatGPT the same thing and compare the two answers. We haven't found another pack that does all of the above. If you have, send it to us.