Prompt engineering training
Prompt engineering training for teams that do not code
Our prompt engineering training teaches a request in four parts, by doing: games that score every attempt, with a real AI agent in the browser. No code, nothing to install.
- Four parts of a request
- Games scored by code
- About 45 minutes a session
The four parts of a request
A request that leaves the agent nothing to guess says four things. Every session of the program trains them.
Context
The situation, who it is for, the facts.
Task
What to do or produce.
Guidelines
How: style, layout, colours, steps.
Constraints
What to keep, and what never to do.
Prompt engineering games that score every attempt
The program Prompting with AI has four sessions. Each request is read for its four parts while the agent works.
The Four Parts
- What the learner does
- Writes the request for an event page from the events team's facts, in three requests at most.
- What it trains
- Leaving the agent nothing to guess.
- What is scored
- A complete request, four parts out of four, read from the learner's own words.
One Shot
- What the learner does
- Describes a page only they can see, in a single request.
- What it trains
- Putting everything the agent needs in the first message.
- What is scored
- A likeness score against the target, with a star at 40%, 70% and 85%.
Prompt Golf
- What the learner does
- Reproduces a page within three requests of at most 400 characters each.
- What it trains
- Saying everything that matters in as few words as possible.
- What is scored
- Likeness stars, then the attempt told against par: under par is a birdie.
Change Only That
- What the learner does
- Asks for four changes to a page that is already online, and nothing else.
- What it trains
- Constraints: what must stay exactly as it is.
- What is scored
- A likeness score on what changed and what did not.
What a vague prompt costs, measured
Prompt engineering is also a matter of cost. On the platform, a single step of the agent costs at least about 21,000 tokens, so each unnecessary request adds up. In "Ask Once, Ask Clearly", "improve it" took 113,000 to 114,000 tokens; the precise request took 49,000.
- Counted step by step. Learners see what each request cost.
- A report at the end. With the tip that saves the most, never during the exercise.
- "Can you make it look better?"113,000 tokens
- A precise request, said once49,000 tokens
Who this prompt engineering training is for
Teams that do not code
Learners describe what they want in plain words and never see code.
Every function
Sessions use made-up companies and cases; your designers can build sessions on your own files.
Companies that want proof
Milestones and scores are decided automatically, so nothing waits on a trainer.
Questions about prompt engineering training
Is this prompt engineering training for developers?
No. It is for people who do not code: learners describe what they want to a real agent in plain words and never see code.
How are the games scored?
By code, not by a person: a likeness score compares the learner's page with a target only they can see, and each request is read for its four parts, quotes checked.
How long does it take?
About 45 minutes a session; the program has four sessions, after the recommended AI Essentials.
Can we try before booking a demo?
Yes. Play a session on the platform page, and try Prompt Golf and the prompt checker on this site: simulations in your browser, no AI model called.
See prompt engineering training with your teams.
A 30-minute call: we get to know your project, then show you the platform.
Book a demo