The third training got rewritten this week

The old version of the third foundations training explained how to check an answer. The new one makes you check one — a real paragraph from your own work, against a source you can actually open. Explaining a habit and installing one turn out to be different jobs, and the submissions were telling me that clearly enough. If you finished the course before this week, the training is worth ten minutes again.

The context block that fixes most generic answers

Role: You are a [role] writing for [audience].
Context:
- What this is for: [where it will be read, and by whom]
- What the reader already knows: [so it stops explaining basics]
- Constraints: [what you can't say, budget, deadline, house style]
Task: [the one thing you want back].
Format: [length, structure, tone]

Worth ten minutes: opening a second assistant

Most people pick one assistant in the first week and never open another, which quietly caps what they think is possible. This is a plain beginner's tour of Claude, and the reason to watch it is not that one tool wins — it is that long documents and long context feel different to work with, and you should know that before your next contract review. Try the same task in both once. You will develop an instinct for which to open, and that instinct is the whole benefit.

Can I paste a client contract into a chat assistant?

This one arrives in the inbox constantly, and the honest answer is: not until someone at your organisation has said which tool is approved for that class of document, in writing. It is not a question about the model's abilities — it is a question about where the text ends up and who can retrieve it later. Until you have that answer, redact the names, figures and dates and work on the structure; you will get almost all of the value and none of the exposure. If nobody at your company can tell you what is approved, that gap is the actual finding, and it is worth raising.

What a win worth posting actually looks like

A win is not "AI wrote my report". A win is the Thursday afternoon task that used to eat three hours and now takes forty minutes, and you can say exactly which part the machine did. Post that shape here — the specific ones help everyone else far more than the impressive ones, because they can be copied. If the honest version is "I tried it and it was worse than doing it myself", post that too; those are the most useful replies in this feed.

Role, context, task, format — in that order

This is the clearest ten-minute explanation of the prompt structure we teach in course two, and the examples are ordinary office work rather than a clever demo. Watch it if your answers have been technically correct and completely unusable. The order matters more than people expect: format last, because deciding the shape before you have decided the substance is how you end up with a beautifully formatted wrong answer.

The tool question people ask first, and the one they should ask

Someone reads every exercise you submit

Most courses skip this part because it does not scale, and skipping it is why most courses do not stick. Every written reflection and uploaded piece of work in this academy is read by a person, and you get a reply — sometimes a short one, sometimes a rewrite of your prompt with the reason it was failing. That is the slowest thing here on purpose. If you have submitted something and it has been more than a couple of days, message me rather than assuming it went into a void.