100 short challenges that move you from pasting questions into a chatbot to running repeatable jobs you'd stake real work on.
You type a request, the answer comes back generic, and you paste it again with a few more words — five times, until it's close enough. That loop is what this course fixes. Across 100 short challenges you rewrite real prompts and watch the output change: adding a word budget, naming your reader, showing two examples instead of describing a rule, asking the model to plan before it writes. Then you go past single questions. You pin answers to text you supply so the model stops inventing facts (retrieval, sometimes called RAG — retrieval-augmented generation), force replies into a fixed shape a spreadsheet or another tool can read, split one bloated prompt into a chain of small steps, and write test cases that catch a prompt the day it quietly breaks. No coding, and any chat model works — ChatGPT, Claude, Gemini. Every challenge gives you a worked demo, specific things to try yourself, and a prompt you can paste straight into your own work. You finish by automating one task you really do every week.
Built by Lakshya Kumar
We grant free access case-by-case — students, career-switchers, builders on a tight budget. Sign in to send us a note.
Sign in to applyFinished the tasks? Take the prompt to your AI and get tested on it. We copy the prompt and open the app — just paste it in.
Add a word budget and name your reader, and the same request comes back tight enough to send without editing it first.
Show, don't tell: how 2–3 examples beat a paragraph of instructions.
Set who is speaking and who is reading, and answers arrive at the right depth instead of landing in generic middle ground.
Ask for a plan, a draft, and a critique as separate passes, and the sloppy first answer stops being the one you receive.
Pin every reply to the same fields so the next tool in your workflow never chokes on a surprise sentence.
When one prompt does five things badly, splitting it into five steps makes each one testable and cheap to fix.
Answer only from documents you supply, with citations, so you can check every claim in seconds instead of trusting it.
Write test cases, including hostile ones, so a prompt that quietly degrades gets caught by you and not by your users.
Hand the model a tool and it can act on its own — learn the limits that keep a helpful loop from becoming a costly one.
Complete all modules, then submit the required number of capstone projects. Each must earn a passing rating from an admin reviewer.
Pick one multi-step task in your actual work and automate it end-to-end with prompts. It can be a research brief, a report summary, a content pipeline — anything you do repeatedly. Use a chain if needed.
Paste this into any AI chat. Fill in the bracketed parts with your context — you'll get back a straight answer on whether this belongs on your plate.
I'm considering taking an "AI Prompting" course that teaches people to go from asking AI for answers to designing prompt-driven workflows. It covers: instruction clarity, constraints, few-shot, personas, chain-of-thought reasoning, structured outputs (JSON), multi-step chains, retrieval basics, guardrails, and light agents.
Here's the context you need:
1. My day-to-day work involves: [describe your job / the kinds of tasks you repeat weekly]
2. My current AI usage looks like: [e.g. "I paste things into ChatGPT once a week" OR "I already script with APIs"]
3. The result I want from learning this is: [e.g. "stop re-writing the same emails", "automate a research routine", "build internal tools at work"]
Given that, answer these specifically:
- Name 3 concrete things in my work that would get meaningfully faster or better if I got good at prompting. Be specific about the task, not generic ("faster writing" is not specific).
- Name 1 thing this course will NOT help me with, so I don't have wrong expectations.
- On a scale of 1–10, how high-leverage is this course for someone in my situation, and why?
- If I only had 2 hours this week, which single skill from the syllabus above would give me the biggest payoff, and how would I know it worked?The official OpenAI take — overlaps with Anthropic's but frames things in tokens-and-tools terms.