40 min read
Three ways to see it
Anand Kumar, the man behind Super 30 in Patna, used to tell his students that the only honest measure of whether you had learned something was whether you could do it without him in the room. This capstone lesson takes the same view. We have spent nine lessons in this Prompt Engineering track learning what a prompt is, how to shape it with personas and roles and examples and chain-of-thought, how to control its output, how to iterate, how to defend it, and how to keep it from rotting in production. The only thing left is for you to do it. This lesson is a structured exercise. By the end of it, you will own a personal prompt library for the five tasks you do most often at work. Not a video you watched, not notes you took, but a working library you actually use.
Step one: list five tasks. Open a blank sheet. Write down five tasks that you do at least once a week in your professional life. Be specific. Not "write emails" but "reply to vendor follow-ups asking about payment status". Not "draft documents" but "draft monthly board memo for the CEO on operations". Not "summarise meetings" but "summarise weekly tax-team meeting into action items by owner". Specificity is everything. A prompt library is useful in proportion to the specificity of the tasks it covers. Spend ten minutes on this step alone. Most professionals find it harder than they expect, and that difficulty is itself revealing.
Step two: for each task, write a system prompt. Use the structure from lesson two: identity, tone, must-dos, must-not-dos, fallback. For the vendor follow-up reply, identity might be "a polite, firm accounts-payable executive at Rupafab, replying to a vendor in formal English with optional Urdu courtesy". Must-dos: confirm the invoice number, give a specific date when payment will be made, sign with your name. Must-not-dos: never commit to a payment date you cannot meet, never share internal cash-flow reasons. Fallback: if the vendor's question is outside your authority, escalate to the finance manager by name. Five tasks, five system prompts. Do not skip the discipline; the structure is the value.
Quick check
Quick check: what makes modern AI different from a rule-based program?
The why-tree
Anand-style opener for your library. Before you build the library, write one paragraph in your own words: which of the five tasks bothers you most every week and why. Save this paragraph as the first page of your library document. When you forget why you did this work in three months, this page will remind you. The library is a tool; the paragraph is the reason the tool exists.
Sources
Sources that cover the whole track for further study. Anthropic's complete prompt engineering documentation at docs.anthropic.com. OpenAI's prompt engineering guide at platform.openai.com. DeepLearning.AI's free course "ChatGPT Prompt Engineering for Developers" by Andrew Ng and Isa Fulford. Lilian Weng's blog post "Prompt Engineering" for the theoretical grounding. The original GPT-3 few-shot paper, Brown et al. 2020. The chain-of-thought paper, Wei et al. 2022. The self-consistency paper, Wang et al. 2022. Yao et al. 2023 on Tree of Thoughts. Simon Willison's blog for the security side. LangChain and LlamaIndex documentation for production patterns. Read at least three of these in the next month.