Learn prompt engineering to get reliable work out of an LLM in three weeks
Three weeks of focused practice — about 40 minutes a day on one free foundations course, one free developer-oriented course, and the reference docs from the people who build the models — gets you from copy-pasting tricks off social media to writing prompts that do real work on purpose. Roughly 15 hours total. Be honest with yourself: this is a practical skill, not magic incantations, and the models shift under your feet every few months. What lasts is the habit of being specific, giving examples, and testing your prompts like code — not any one "ultimate prompt" you memorise.
3 weeks · ~15 hours · write and test prompts that reliably do a real task
1.Learn Prompting (foundations course)
Start with Learn Prompting, the largest free, open-source course on the subject and the most neutral one — it teaches prompting across ChatGPT, Claude, and other models rather than selling you one vendor's tricks. Work the "Introduction to Prompt Engineering" track: clear instructions, role and system prompts, few-shot examples, and chain-of-thought. Do not just read; open a chat window in a second tab and try every technique on a prompt you actually care about. The goal of week one is to stop guessing and start understanding why a prompt works.
Free; open-source
Learn Prompting →2.ChatGPT Prompt Engineering for Developers
DeepLearning.AI's short course, taught by Isa Fulford of OpenAI and Andrew Ng, is about ninety minutes of video with a live coding notebook beside it. It reframes prompting as something you build and iterate on — summarising, extracting, transforming, and writing with structured prompts — rather than a chat parlour trick. Run every notebook cell yourself and change the prompts until they break, then fix them. This is where prompting starts to feel like engineering: you state what you want, see what you get, and tighten the loop.
Free
DeepLearning.AI · Prompt Engineering for Developers →3.Anthropic's prompt engineering docs, then a real task
Read Anthropic's prompt engineering overview — the clearest vendor reference there is, covering being clear and direct, multishot examples, giving the model room to think, XML tags, and prompt chaining. Then put it to work: pick one task you do every week — triaging emails, turning notes into a draft, classifying support tickets, cleaning a spreadsheet — and engineer a prompt that does it reliably across ten different inputs, not just the one you tested. When it holds up on inputs you did not design for, you have actually learned the skill.
Free
Anthropic · Prompt engineering →If this doesn't fit you
If you learn best by doing rather than reading, skip straight to Anthropic's free interactive prompt engineering tutorial (github.com/anthropics/prompt-eng-interactive-tutorial). It is a hands-on, chapter-by-chapter course you run in a notebook or even a spreadsheet, with exercises and answer keys, so you write and grade prompts the entire way through. Pick it if lectures bore you and you want to be typing prompts from minute one.
Why this path
Most people "learning prompts" collect screenshots of clever one-liners and wonder why they don't work next month. The bottleneck is not a missing magic phrase — it is that they never learned to be specific, to show examples, and to test a prompt against inputs they didn't cherry-pick. This sequence fixes exactly that: Learn Prompting gives a vendor-neutral mental model, the DeepLearning.AI course turns it into an iterative build-and-test habit, and Anthropic's docs plus your own real task force the habit to survive contact with messy reality. Because the models change fast, the durable skill is the method, not any single prompt — and that is what this path drills.