Why this track exists
Your builders have a 10-session deep track on this site that goes all the way to eval pipelines and prompt security. This track is the other half of the deal: six short sessions for the people who sponsor AI, approve its budgets, and use it between meetings - executives, boards, managers. No code, no API keys. Just the thinking moves that separate people who get real work out of AI from people who got a funny paragraph once and gave up.
Meet your brilliant, literal new hire 8 min live
Anthropic's own documentation tells prompt writers to treat the model like a brilliant new employee - one with amnesia. Sharp, fast, well-read, and knowing absolutely nothing about your company, your project, or what you said in the hallway. Everything it needs must be in the request.
LiveMeet Meridian - a Monday you will recognise3 min▶
Meridian Retail is a 2,000-person retailer: 80 stores, an online arm, a data team of twelve. The exec team just got company AI accounts. Within a week, three things happened:
- The COO, Dana, asked for "a summary of our customer complaints" and got a beautifully written essay about retail complaints in general - none of it about Meridian.
- The CMO pasted a campaign brief and got back exactly the same bland copy three times, concluded "the AI is mediocre", and went back to the agency.
- The CFO asked for market sizing, got confident numbers with no sources, and nearly put them in a board deck.
Nobody did anything wrong by their old habits. They delegated to the AI the way you cannot delegate to anyone: no context, no success criteria, no examples. This track is the story of that exec team learning to ask properly - and what changed.
The pattern in every company: the person who says "AI is useless for real work" and the person getting hours back every week are usually using the same tool. The difference is rarely the model. It is the brief.
LiveThe golden rule - one test before you hit send3 min▶
Anthropic's documentation gives one rule above all others, and it needs no technology to apply:
Notice what this implies: prompting well is not a technical skill. It is the same skill as writing a good work request for a person - which is why the best prompt writers in most companies turn out to be the best delegators, not the best engineers.
| Delegating to a person | Delegating to a model |
|---|---|
| "You know the client, so tone it right" | Say who the client is and what tone you want |
| "Like the one you did last quarter" | Paste the one from last quarter as the example |
| "Use your judgment on length" | Say the length. It will happily write 3 pages or 3 lines |
| They ask questions when unsure | It guesses when unsure - unless you tell it to ask |
Self-studyWhy "magic words" thinking took hold - and why it faded2 min read▶
In 2023 the internet filled with "50 secret prompts" listicles, and companies hired "prompt engineers" at eye-watering salaries. Two things have happened since:
- Models improved. Anthropic's own guidance now notes that some classic tricks (special formatting tags, elaborate role-play) are less necessary with modern models. The models meet you more than halfway.
- The craft matured. What survived is not vocabulary but structure: clear goal, right context, defined output, an example when it matters. The industry now calls the wider discipline context engineering - deciding what information the AI sees - and clear asking is the layer of it every leader personally controls.
So the fashionable phrases were never the point. The thinking was. Good news: thinking is transferable, and you already do it every time you brief a person well.
The four skills of working with AI 7 min live
Anthropic teaches organisations a framework called AI Fluency with four skills, all beginning with D. Prompting - the subject of this course - is the second one. Knowing all four tells you where prompting fits, and why it alone is not enough.
LiveThe loop, not the lottery4 min▶
The single biggest behavioural difference between frustrated and fluent AI users: fluent users iterate on purpose. They treat the first answer as a first draft from the new hire - useful information about what the brief missed - and refine.
- First pass: decent brief, see what comes back.
- Read the miss: wrong tone? Missing context. Wrong content? Vague task. Too generic? No example.
- Fix the brief, not the mood: add the one thing the output shows was missing, and go again.
Two or three rounds like this routinely beat an hour of trying to compose the perfect first prompt. Andrew Ng's famous developer course teaches exactly this as its core method: nobody writes the right prompt first time, and nobody needs to.
Dana's second try. After the generic complaints essay, Dana pasted 40 actual complaint tickets in and asked for "the top 5 themes, each with a count and one verbatim quote". Ninety seconds later she had a slide she actually used. Same tool, same morning. The difference was all in the ask.
Self-studyWhat belongs in a strong ask - the anatomy3 min read▶
You will meet this anatomy in every session of both tracks, so here it is once, plainly. A strong ask usually carries five things:
| Ingredient | Answers | Example fragment |
|---|---|---|
| Goal | What does done look like? | "Draft a one-page memo recommending one option" |
| Context | What would a new hire need to know? | "We are a mid-market retailer; the audience is the board" |
| Constraints | What are the rules? | "Under 400 words, no jargon, GBP figures" |
| Format | What shape should it take? | "Three headed sections, each ending in one action" |
| Example | What does good look like? | "Match the tone of the memo below: ..." |
Not every ask needs all five - a quick question needs none of them. The skill is noticing which missing ingredient caused a weak answer. That diagnosis habit is Session 2's whole subject.
"Isn't prompt engineering dead?" 7 min live
You may have seen the headlines: the six-figure prompt engineer job has largely disappeared, and analysts now talk about context engineering instead. Both true - and neither means what the headlines imply. Here is the honest read for a leadership audience.
LiveWhat died, what lives4 min▶
| Died | Lives - and grew |
|---|---|
| The standalone "prompt engineer" job title | Prompting as a baseline skill in every role, like spreadsheets |
| Secret-phrase folklore ("act as a genius...") | Clear goal, right context, defined output, real examples |
| Tricks to force step-by-step thinking | Models that reason by themselves - you state the goal and the bar |
| One perfect mega-prompt | Context engineering: giving AI the right documents, data, and tools |
The executive translation: the craft moved up a level, closer to what leaders already do. Modern models need less coaching on how to think and more clarity on what you actually want and what information they should use. Asking well became more valuable, not less.
Self-studyWhy "context engineering" matters to you anyway2 min read▶
Context engineering is the practitioner discipline of curating everything the model sees: documents, data, tools, history. Your builders will live it in their track. It reaches your desk in two ways:
- Personally: the highest-leverage upgrade to your own AI use is rarely fancier wording - it is pasting in the actual document, the actual numbers, the actual example. Context beats phrasing, almost every time.
- Organisationally: when teams propose AI projects, the real question is rarely "which model?" It is "what context will it have?" - which data, which systems, kept current how. That is a strategy and budget question, and it is yours. Session 5 turns it into a checklist.
The makeover: one real ask, upgraded live ★ 15 min · work it live
Take one real request from your own week - an email to summarise, a memo to draft, a plan to critique. We upgrade it together using the anatomy from Part 2. Here is the worked example with Dana's complaints ask, before and after.
Pick your real ask. Something you genuinely need this week. Toy examples teach nothing - this is a build-along on your own work.
Run it raw first. Send the one-line version. Keep the answer open in a tab - it is your before photo.
Rebuild it with the anatomy. Goal, context, constraints, format, example. Steal Prompt B's shape wholesale.
Compare, out loud. What changed? Usually: specificity, relevance, and a format you can actually use. Name which ingredient did the work.
Golden-rule it. Show your upgraded ask to the person next to you. If they can follow it cold, ship it. If they frown, you found the gap before the model did.
Three things before Session 2 ◐ 20 min total
- Upgrade three real asks this week using the anatomy. Keep the before and after answers side by side - bring your best contrast to Session 2.
- Apply the golden rule once: show one prompt to a colleague before sending it. Notice what they ask you - that question was the hole in your brief.
- Catch one confident guess. Find one place where the AI filled a gap you left (a number, a name, an assumption about your company). That instinct - spotting the fill - is Session 4's foundation.
What this session covers
Leader-level coverage of the sources below - working ideas in plain language, honestly flagged where the depth lives in the builder track. Official certificates and videos stay with the vendors; links are in the course map.
Three questions before you go 🎯 ◐ 90 seconds
1 · The golden rule of clear prompting says to test your prompt by...
Prompting well is delegation, and the test is human: could a smart person with no background follow this brief? No technology required.
2 · When your ask leaves a gap - a missing number, audience, or definition - the model will usually...
Left to itself, the model fills gaps confidently rather than asking. You can invite questions ("ask me anything unclear first") - but the default is the guess.
3 · With the newest reasoning models, the best habit shift for leaders is...
Reasoning models plan internally. Micro-managing their method can make results worse - the leverage moved to clear goals and success criteria. Delegation, again.