Prompt Workshop: Write and Test

The concept

This lesson is a hands-on workshop, not new theory — it asks you to apply everything from this module (prompt anatomy, few-shot examples, system prompts, and iteration) to one real task from your own job, building a complete, reusable prompt from scratch instead of just studying someone else's worked example. The goal isn't to understand the concepts abstractly; it's to walk away from this lesson with one actual, saved prompt you'll genuinely use again this week.

The workshop pulls together four skills you now have, applied in order. Start with the four-part anatomy (context, task, format, constraints) to make the task unambiguous — this is the foundation, and skipping it is the single most common reason a "built" prompt still under-performs. Add at least one example input/output pair if the task involves tone, structure, or judgment calls that are easier to show than describe; a single good example often does more work than an extra paragraph of description. Consider whether this is a task you'll do repeatedly enough to warrant a standing system prompt or persona, rather than rebuilding the instructions every time — weekly or more-often tasks almost always justify it. And finally, treat your first output as a draft to debug — expect at least one iteration round before the prompt is genuinely reusable, using the diagnose-one-gap-at-a-time method from the previous lesson.

The habit this builds is arguably the single most valuable practical skill in the course. Most people who use AI tools write a new, improvised prompt every single time they need something, re-explaining context and format from scratch — which is slow and produces inconsistent results from one day to the next. People who get consistently better, faster results instead build a small library of tested, saved prompts for their recurring tasks — status reports, meeting notes, customer replies, content drafts, meeting agendas — that they reuse and only lightly adjust each time. The time investment to build one good, reusable prompt pays back every single time you use it afterward, which for a weekly task could be dozens of times a year.

A practical note on where to save a working prompt: a dedicated notes app, a shared team doc, or a purpose-built prompt-library tool all work — what matters is that it's somewhere you'll actually open again next week, labeled clearly enough that future-you (or a teammate) can find and reuse it without re-reading this whole course.

Worked example

Starting point: a vague request like "help me write a status update." Applying the four-part anatomy: context — "I'm a product manager reporting to my VP of Engineering"; task — "summarize this week's progress on the mobile app redesign, including one blocker"; format — "4 bullet points: status, blocker, impact, next step, each under 20 words"; constraints — "no jargon, confident tone, no apologizing for the blocker."

Adding a one-shot example of a previous week's bullet-point status update anchors the exact phrasing style expected, rather than leaving "confident tone" open to interpretation. Setting this up as a reusable system prompt — "You are my status-update drafting assistant; always use the 4-bullet format above; here's last week's example for style" — means next week's update only requires pasting in new raw notes, not rebuilding the whole prompt from memory.

After the first run, one iteration round typically gets it to final, reusable form — for example, trimming a bullet that ran to 28 words down to under 20, or swapping out a buzzword the VP dislikes. The final saved version might read: "[System prompt as above, with example]. This week's raw notes: [paste]. Produce the 4-bullet update." That final line is now a template — the learner just swaps in new raw notes each week and runs it, turning a 15-minute writing task into a 2-minute paste-and-check task.

Exercise

Pick a task you do at least weekly in your actual job: status reports, meeting notes, customer replies, or social captions are common choices, but use whatever's real for you. Build a complete prompt using the four-part structure (context, task, format, constraints), add at least one example of what "good" looks like for this specific task, and run it.

Do at least one documented iteration — identify one specific gap in the first output (using the diagnostic categories from the previous lesson) and fix it, then re-run. Decide explicitly whether this task is frequent enough to deserve a standing system prompt; if yes, set one up in your AI tool rather than just keeping the prompt text in a notes file.

Save the final, working version of this prompt somewhere you'll actually find it again. Write one line next to it describing what it's for and when to use it, so that three months from now, you (or a teammate) can find and reuse it without needing to remember the reasoning behind it. This saved, working prompt is the deliverable for this lesson — not just the one output it happened to produce today.

Recap

• A prompt worth reusing is deliberately built using anatomy, examples, and iteration — not improvised fresh each time.
• The workshop's output is a saved, working prompt template for a real recurring task, not just a single good answer, plus a one-line note on when to use it.
• Building a small personal library of tested prompts for recurring tasks is the single highest-leverage habit from this module — keep adding to it after the course ends.

Resources

• Anthropic: "Prompt library" — https://docs.anthropic.com/en/prompt-library/library
• OpenAI Cookbook: example prompts and patterns — https://cookbook.openai.com/
• Notion / Google Docs (or any shared doc tool) as a simple home for a saved prompt library — https://www.notion.so/