Day 1
Where Claude is different
Learn what Claude is genuinely better at, so you stop using it like a search engine.
Concepts
- · Long context: it can hold whole documents, transcripts and codebases in one conversation
- · Strong at nuance, careful writing, structured reasoning and following detailed instructions
- · It does not browse by default the way an answer engine does — it reasons over what you give it
- · Free vs paid: usage limits, model choice, Projects and larger file handling
Exercise
Paste a long real document — a contract, report or transcript — and ask for the three things that would change a decision. Compare that to skimming it yourself.
Try this prompt
Here's a [document type]. I'm a [role] deciding whether to [decision]. Pull out only what affects that decision, flag anything ambiguous, and tell me what's missing from this document.
Mastery signal: You can name one task where Claude clearly beat your usual tool, and one where it didn't.
Day 2
Context, roles and instructions
Front-load context so you stop iterating three times to get something usable.
Concepts
- · Tell it who you are, who the output is for, and what 'good' looks like
- · Give examples of the style you want — one good sample beats a paragraph of adjectives
- · Ask it to think through the problem before answering when the task is analytical
- · Tell it explicitly to ask clarifying questions instead of guessing
Exercise
Take a task you did badly with AI last month and rewrite the prompt with role, audience, constraints, format and an example. Compare results.
Try this prompt
You're helping a [role] at a [context]. The audience is [audience]. Produce [output] in [format], no longer than [length]. Here's an example of the tone I want: [paste]. Ask me anything you need before starting.
Mastery signal: Your first response is usually good enough to edit rather than redo.
Day 3
Files, data and analysis
Work over your own material rather than general knowledge.
Concepts
- · Upload PDFs, spreadsheets, images and documents and ask questions across them
- · For data files, ask for the method as well as the answer so you can check the working
- · Multi-document comparison: contradictions, gaps, changes between versions
- · Privacy: check your organisation's policy before uploading confidential material
Exercise
Upload two versions of the same document and ask for a change log with the material differences only — then verify three of them yourself.
Try this prompt
Compare these two documents. List only material differences that change obligations, numbers or timelines. Ignore formatting and wording changes. Cite the section for each.
Mastery signal: You caught a real difference or error you'd have missed reading manually.
Day 4
Projects: context that persists
Stop re-explaining yourself. Set up a space where your standards and material live.
Concepts
- · A Project bundles files, custom instructions and related conversations
- · Custom instructions carry your role, tone, standards and formatting rules into every chat
- · Good Project knowledge: style guides, past examples, terminology, key reference documents
- · One Project per recurring workstream, not one giant Project for everything
Exercise
Create a Project for a recurring responsibility. Add instructions and three reference files, then run a real task inside it and note the quality difference.
Try this prompt
Project instructions: I'm a [role] at [org type]. Always write in [tone], use [terminology conventions], format as [format], and flag assumptions explicitly. Prefer questions over guesses.
Mastery signal: Your prompts got shorter because the Project already knows your context.
Day 5
Artifacts and building things
Get deliverables, not chat messages.
Concepts
- · Artifacts render documents, tables, diagrams and small interactive tools in a side panel
- · Iterate on an artifact in place rather than regenerating the whole thing
- · Good uses: calculators, checklists, templates, dashboards, one-page briefs
- · Anything you'll reuse should become an artifact you keep, not a message you scroll back for
Exercise
Turn a recurring manual task into a small artifact — a calculator, template or checklist — and use it for real once.
Try this prompt
Build an interactive [calculator/checklist/template] for [task]. Inputs: [list]. It should output [output]. Keep it simple enough for a non-technical colleague to use unaided.
Mastery signal: You used something Claude built for you on real work, not just as a demo.
Day 6
Advanced surfaces: Claude Code, connectors and MCP
Understand what the wider ecosystem adds — and whether you need any of it.
Concepts
- · Claude Code works in your terminal and repository for real software work
- · Connectors and MCP let Claude reach tools and data sources you approve
- · Desktop and mobile apps for capture and quick access; browser extension for in-context work
- · Honest assessment: most non-developers get 90% of the value from Projects and Artifacts alone
Exercise
Pick the one advanced surface relevant to your job, try it for a single real task, and decide honestly whether to keep it.
Try this prompt
Given that I'm a [role] doing [main tasks], which parts of Claude's ecosystem would actually change my week, and which would just be novelty? Be blunt.
Mastery signal: You made a keep-or-drop decision about one advanced feature based on a real trial.
Day 7
Verification and a weekly rhythm
Build the habits that make AI output safe to put your name on.
Concepts
- · Always verify: figures, names, dates, quotes, citations, legal, medical and financial claims
- · Claude can be confidently wrong — fluency is not accuracy, and it doesn't browse by default
- · Ask it to argue against its own conclusion to surface weak assumptions
- · Set a rhythm: drafting slot, analysis slot, weekly review — rather than random usage
Exercise
Write a five-line personal policy: what you always use it for, what you never use it for, and what you always check before sending.
Try this prompt
Here's your previous answer. Argue the strongest case against it, list every assumption you made, and tell me which claim is most likely to be wrong.
Mastery signal: You've used Claude on three real tasks this week and caught at least one error before it mattered.