Capstone — three production prompts
The capstone — three production prompts of your own
Hagar shipped the Bayt Coffee assistant. The owner is happy, customers stopped phoning the shop for tracking numbers, and the side gig is paying for her espresso habit. Now her main job — a SaaS company where she writes frontend code — needs the same treatment. Two of her teammates also want assistants. She is suddenly the in-house prompt engineer, and the thing standing between her and shipping three more prompts is one weekend of work.
That is the capstone. Pick three real scenarios from your own life or work and ship a production-grade system prompt for each. Not toy prompts. Not tutorial prompts. Three prompts you would feel comfortable handing to a colleague tomorrow.
The capstone workflow — for each of the 3 prompts you ship
What "real" means
A real scenario has four properties:
- A specific user. Not "developers" — "the senior backend engineers on my team", or "first-time customers landing on our pricing page". The narrower, the better.
- A repeating task. Something done at least once a week. One-shot prompts do not need this much polish.
- A measurable outcome. "Was the email rude or fine?" is measurable. "Did it sound creative?" is not.
- An owner. Someone (probably you) who will iterate on the prompt when it misbehaves in week three.
If a scenario does not have all four, swap it for one that does.
Three suggested categories
You can pick any three, but most people land on one from each:
| Category | Example |
|---|---|
| Personal automation | A tutor for your weekend reading; a meeting-notes formatter; an inbox triage helper. |
| Job-related tooling | A code-review assistant; a PR description writer; a customer-support drafter for one specific issue type. |
| Creative or hobby | A pitch-feedback partner; a recipe-from-pantry-list generator; a workout planner with hard constraints. |
What to deliver
For each of the three prompts, you will hand in:
- The scenario brief (one paragraph: who is the user, what is the task, what does success look like).
- The system prompt itself (under 400 words, all 5 slots present).
- One real input you fed it.
- One real output the model returned.
- A self-graded score out of 10 against the rubric in the next lesson.
That is it. Not a writeup, not a deck. The four artefacts above for three scenarios. The next lesson is the rubric you will grade yourself against, and the lesson after that is three worked example scenarios you can use as templates if you are stuck.
Next: the 10-point rubric. :::
Sign in to rate