Prompting for every job: 59 days
Fifty-nine one-hour sessions, one a day, from 3 November to 31 December, in four modules. The goal: write prompts that get a usable answer the first time, edit the ones that do not, and prove with tests that they keep working, in whatever IT job you do or are aiming for.
Who this path is for, and where it leads
For anyone who works, or wants to work, in IT and already talks to a model every day: technicians, developers, testers, designers, analysts, project leads, and the self-taught learner who follows one of the other paths of this site. It assumes no diploma and no paid tool: a free local model is enough. Prompting is rarely a job title on its own; it is a skill that every job on this site now expects. The roles that ask for it explicitly are often called AI engineer, LLM application developer or AI automation specialist, and they always ask for the evaluation side too: test cases, regressions, measured results.
What you show a recruiter on 31/12: a public Git repository holding your personal prompt library, each prompt with its purpose, its inputs, its expected output and its tests, a test report that runs on two models, and a short write-up of three prompts you improved, with the before, the after and the measured difference.
The method, the same every day
| Time | Phase | What you do |
|---|---|---|
| 20 min | Theory | The session's lesson (on the module page), the exercises after each part. |
| 30 min | Practice | Write, run and edit prompts on a real task, note what changed and why. |
| 10 min | Recap | Three lines in the notebook, then redo the day's exercise without notes. |
The recap is the phase people skip, and it is the one that makes things stick: redoing from memory forces you to retrieve the information, and retrieval is what anchors it.
The tools, installed once (day 1)
- A local model: Ollama and one small instruction model your machine runs comfortably. The site's article Your first local model with llama.cpp shows the other main way to run one; the commands of this path use Ollama, and the prompts work the same on both. A free hosted chat (Claude, ChatGPT, Gemini, Le Chat) is a useful second model to compare against; never paste anything private into it.
- Git and a free account on a code forge, for the prompt library you build from day 1.
- A text editor and a digital notebook: Obsidian or a plain folder of Markdown files, one file per day.
- From module 4: promptfoo, an open-source tool that runs a set of prompts against test cases (it needs Node.js).
Learning with an AI as a tutor, without letting it do the work
This path is about talking to models, so you will use one all the time. The rule stays the one of every path on this site: the model explains, quizzes you and reviews what you wrote; it does not write your exercises. Ask it to quiz you on the day's notions, to criticise your prompt against the checklist of module 1, to explain why two prompts give different answers. Never paste its answer as yours, and check what it says about itself against the vendors' documentation linked below: a model describing how it works is a model producing plausible text, not a measurement.
Prompting, in one card
Every prompt of this path is built from the same seven parts, each used only when it earns its place: role and context (who is asking, for whom, in what situation), task (one verb, one deliverable), constraints (what must and must not happen), inputs (the data, clearly fenced off from the instructions), output format (exactly what comes back), examples (one or two, when the format or the tone is hard to describe), and success criteria (how you will check the answer). Then you test it, and you keep the version that passes.
The golden rule
One hour every day beats seven hours on Sunday. If you miss a day, you do not catch up the next day: you shift everything by one day, and you trim module 4. The dates shown assume a start on 3 November; only the order and the regularity matter.
The four modules
Module 1 lays the foundations: how a model reads, the anatomy of a prompt, editing, structured output, self-checks and prompt injection. Modules 2 and 3 take the jobs one by one, from the server room to the product: for each, a weak prompt, its diagnosis, the edited prompt, and exercises. Module 4 finishes the jobs, then moves to agents (system prompts, tool descriptions), to testing prompts, and to the final project. Each module ends with a deliverable that goes into your portfolio.
Recommended free resources
- Anthropic — Prompt engineering overview and its interactive tutorial
- OpenAI — Prompt engineering guide
- Google — Prompt design strategies (Gemini)
- Mistral — Prompting
- OWASP — LLM01 Prompt Injection
- Prompt Engineering Guide (DAIR.AI), a free index of techniques with links to the papers
The end-of-path checklist, to tick on 31/12
- I explain context window, tokens and temperature in two sentences each.
- I write a prompt with the seven parts and cut the ones it does not need.
- I fence data off from instructions and I explain why that does not fully stop prompt injection.
- I get JSON that a program validates, and I reject what does not validate.
- I have a tested prompt for at least five jobs of this path.
- I write a system prompt and a tool description an agent uses correctly.
- I measure a prompt with test cases and compare two versions on numbers.
- My prompt library is public, documented and tested.
A box that will not tick is not a failure: it is the list of what you will work on first.
- 1Prompts module 1: how a model reads, and how to write for itDays 1 to 14: tokens, context window and sampling, instructions versus data, the seven parts of a prompt, editing, structured output, self-checks and prompt injection.
- 2Prompts module 2: operations jobs, from support to securityDays 15 to 28: prompts for the system administrator, the support technician, the network engineer, the DevOps engineer and the security analyst, each with a weak prompt, its diagnosis, the edited prompt and exercises.
- 3Prompts module 3: building jobs, from code to architectureDays 29 to 43: prompts for developers (review, tests, debugging, refactoring), web designers and UX, data and SQL, machine learning, QA and architecture, each with a weak prompt, its diagnosis, the edited prompt and exercises.
- 4Prompts module 4: project jobs, agents, evaluation and the final libraryDays 44 to 59: prompts for project management, technical writing and the job search, then system prompts for agents, tool descriptions, testing prompts with test cases, and the final project: a documented, tested personal prompt library.