Bookio
Bookio's team left with one prompt that replaces ten
The team already used AI, everyone differently and nobody sure what they were allowed to paste in. One day, four blocks and six hands-on exercises turned ad-hoc prompting into a method the whole team shares.
Participants work on their own tasks, on their own tools. What they build during the day is what they use the next morning.
01The starting point
Where it started.
The team was already using AI. Everyone in their own way, in a different tool, and nobody was confident about what they were allowed to paste into it.
In detail
That produced three concrete problems. The same task returned different quality depending on who wrote the prompt. Nobody could say which data may go into a public tool and which may not. And people kept rewriting the same prompt from scratch instead of building it properly once.
The goal was not to teach the team that AI exists. It was to move them from ad-hoc prompting to a repeatable method that produces the same quality regardless of who is at the keyboard.
Quote
“Exercise one is designed to go badly. The framework that fixes it comes afterwards, and that order is the difference between hearing about it and being able to use it.”
02How the day was built
What the training covered.
Four blocks, each pairing a short explanation with an exercise done on the participant's own tool and own task.
Why this format
Foundations, including the cost of a token
How a model generates rather than retrieves, what a token is and why Slovak text costs more than English, which data never leaves the company, and what each of the four common tools is actually good for.
One composition framework
Role, goal, context, instructions, format. Participants watch a single example move from a bad prompt to a professional one, then turn it into something that runs on a single word.
Security practised, not lectured
Participants try to break their own agent with prompt injection and mostly succeed on the third attempt. Then they write a real work email with sensitive data masked. A rule you have tried once holds better than a rule in a policy document.
03Six exercises
Step by step.
Write a prompt however it comes to you, then rewrite it using the composition framework, then turn it into something that runs on a single word. Three exercises that move a participant from a one-off request to a reusable tool.
How it runs
Then try to break your own agent, write a real work email with AI, and finally write it again with sensitive data masked. The order is deliberate and cannot be rearranged without losing the effect.
In practice
Block four draws the line participants most often get wrong: where a good prompt stops being enough and the task needs a workflow, retrieval over company documents, or an agent with access to real systems.
04What the team took away
What changed.
What a team has after one day, measured by what they can do the next morning without asking anyone.
- One structured prompt template that gets filled in per task and reused, instead of rewriting from scratch
- Three things that never go into a public tool, plus a masking technique for when real content is needed
- A decision rule for tool choice: which one suits long documents, sourced research, or working inside an existing toolset
- The ability to tell a task a prompt solves from one that needs a workflow or company data
- The same output quality across the team, regardless of who writes the prompt
Why it worked
Mistake before instruction. The first exercise is meant to disappoint, and only then comes the framework that fixes it. Teams that are shown the answer first tend to nod along and change nothing.
Security was practised as a habit rather than delivered as a list of prohibitions. Masking data on a real email takes two minutes and sticks, where a policy document is read once and forgotten.
Quote
“Participants worked on tasks that were waiting for them the next day, not on a training scenario. That is why the prompts survived the workshop.”
Next step
Your process could be the next case.
A 30-minute call. You tell us about one process, and we say whether it fits the platform, a product build or a consulting project.