Before you start: which Claude do you have?
Claude has different plans with very different capabilities. Knowing which one you are on decides which features in this course you can actually use today.
FREE
Limited daily messages, smaller model (Haiku), no Cowork or Projects. Good for trying Claude out.
PRO ($20/MO)
5x more usage, access to Sonnet and Opus, Projects, Artifacts, file uploads. The standard plan for serious individual use.
MAX ($100-200/MO)
Much higher usage limits, priority access, advanced Cowork and Computer Use features, longer context. For heavy daily users.
TEAM / ENTERPRISE
Per-seat plans with admin controls, shared Projects, security and compliance features, optional zero data retention. For organisations.
Know Your Claude Surfaces
Claude lives in five different places. Picking the right one is the single biggest factor in whether the output feels like a clever assistant or a chatbot.
If you remember one thing from this module, make it this: the wrong surface cannot be rescued by a better prompt.
When people first try Claude, they usually open claude.ai in a browser tab and ask it to do a task like "analyse my sales spreadsheet" or "fix the bugs in my code." The output is generic, partial, or just guesses based on what you describe rather than what is actually in the file. The conclusion many people reach is "Claude is impressive at chat but not great at real work."
The problem is not Claude. The problem is that Claude in a browser tab can only see what you paste or upload. It cannot see your folder structure, your codebase, your file system, or any of the local context that would make its answer truly useful. That is what other surfaces are for.
There are five different places Claude lives, and each one has different capabilities and context. Picking the right surface for your task is the single biggest factor in whether the output feels like a thinking partner or a generic AI.
Each surface solves a different problem. Same underlying model family, but very different access to your work.
Chat
The classic chatbot. Available in any browser at claude.ai, in the mobile app, and in the desktop app. Stateless by default, meaning each new conversation starts fresh with no memory of past chats unless you turn on the Memory feature.
Projects Pro+
A saved workspace inside claude.ai with files, custom instructions, and shared context. Upload files once and Claude remembers them across every conversation in that Project. Best for recurring topics where the same reference material applies every time.
Cowork Pro+ · Desktop
A desktop tool that lets Claude read and write files on your computer. Point it at a folder and it can analyse spreadsheets, edit documents, refactor codebases, and produce real .docx, .xlsx, .pptx, and .pdf files in the folder you specify. The closest Claude gets to "do this for me on my actual computer."
Claude Code Pro+ · CLI
A command-line tool for developers. Run claude in your terminal, and Claude can read your entire codebase, write new files, run tests, and execute commands. Built for engineers handing off real coding tasks rather than asking trivia.
Artifacts
Not really a separate surface, but worth its own card because most people never use it well. Artifacts are interactive content that renders inline in your conversation: charts, HTML pages, React components, SVG diagrams, working apps. Available in Chat and Projects.
Claude is not one model, it is a family. Picking the right one for the task affects both quality and how quickly you burn through your usage limits.
| Model | What it does | When to use it |
|---|---|---|
| Opus 4.7 | The most capable model. Highest quality reasoning, longest sustained focus, best for nuance and complexity. Slower and uses more of your daily quota. | Strategy work, complex analysis, important emails, anything where quality matters more than speed. |
| Sonnet 4.6 | The balanced workhorse. Strong reasoning at much faster speed than Opus. The default for most paid users and most everyday work. | General drafting, code review, summarisation, most "real work" tasks. The sensible default. |
| Haiku 4.5 | The fast model. Lighter reasoning, much faster output, much lower cost via the API. Good enough for many simple tasks. | Quick lookups, simple drafts, casual chat, automated tasks where speed matters more than depth. |
| Extended thinking | A toggle (not a separate model). Gives Claude time to "think out loud" before answering. Available on Opus and Sonnet. | Hard problems where reasoning matters: maths, multi-step logic, complex tradeoffs, debugging. |
Default to Sonnet, escalate to Opus when needed
For most paid users, Sonnet is the right default. It is fast, capable, and uses far less of your quota than Opus. Save Opus for tasks where the quality difference will matter, such as strategic writing, complex analysis, or anything you would otherwise spend a long time refining manually. Many users burn through their Opus allowance on tasks Sonnet would have handled equally well.
Diagnosing bad output almost always lands on one of these.
Using Chat for work that needs Cowork or Projects
You paste pieces of three documents into Chat and ask Claude to cross-reference them. The output is partial because Claude is working from the snippets you pasted, not the full files. Either upload them to a Project (so Claude has them in full and remembers across conversations) or use Cowork to point Claude at the folder containing them.
Re-explaining yourself every conversation
You start every chat with "I am a [role] working on [topic], my style is [tone]..." That is what Projects custom instructions are for. Set the context once at the project level, and every conversation in that project inherits it. Saves you 30 seconds per chat and makes Claude better-aligned by default.
Asking Opus to do Haiku work
You burn through your Opus allowance on quick lookups and simple rewrites. Then for the actual hard task you needed Opus on, you have already hit your quota. Match the model to the task: Haiku for fast and simple, Sonnet for most work, Opus for hard reasoning.
A simple decision guide for everyday tasks. When in doubt, come back to this.
If your task is...
Before the quiz, remember
- Five surfaces: Chat, Projects, Cowork, Claude Code, Artifacts. Each has different capabilities and context.
- Three main models: Opus (best quality, slowest), Sonnet (balanced default), Haiku (fast and cheap).
- The wrong surface cannot be fixed by a better prompt. Chat cannot see files in your folder. Cowork can.
- Projects = Claude's long-term memory for a topic. Stop re-uploading the same files every conversation.
- Match the model to the task. Sonnet by default, Opus for the hard stuff, Haiku for speed.
Test yourself
Five quick scenarios to check your understanding. Click an answer to see feedback.
Nice work!
You scored 0/5. Mark this module complete below to save your progress.
Write Prompts That Get Usable Output
Most of the difference between "wow" and "useless" comes from how you ask, not what you ask. Here is the pattern that works for Claude.
More than any other single habit, this is what separates Claude power users from everyone else.
Two people on identical Claude plans, with identical use cases, can get wildly different output. Sometimes 10x different in quality. The difference is almost never which model they pick. It is how they ask.
When you type a vague prompt, Claude has to guess at what you actually want. It guesses based on statistical patterns from a vast amount of training data. The "average" answer that fits the average meaning of your vague prompt is, by definition, generic. That is what generic output actually is: the AI defaulting to the most common interpretation because you did not give it enough to work with.
When you write a specific, well-structured prompt, Claude has far less to guess at. It can produce something tailored to your situation, your audience, your constraints. Same model, same plan, same tool. Radically different output.
A small note about Claude specifically: it tends to be more responsive than other AI tools to explicit role-setting and to "ask me clarifying questions before you start" prompts. Lean into both.
Same task, three approaches. The difference is not length, it is structure.
Let Claude do the thinking, not you
The "Better" prompt is good, but Claude still has to invent what your voice sounds like. The "Best" prompt gives it real reference material to ground on, plus permission to ask before writing. The result sounds genuinely like your work because it has been shaped by your actual writing, not by Claude's guess at how you might write. This is the highest-leverage habit to build.
Most great prompts have five ingredients. Missing any one usually explains why output feels off.
Five-ingredient prompt, annotated
Each colour shows what that part of the prompt does.
Role
Who Claude is being in this response. Changes tone and perspective.
Task
What you actually want produced. Be specific.
Context
Who it is for, what came before, the situation around the task.
Constraints
Length, tone, what to avoid, hard rules.
Format
The shape of the output: sections, options, structure.
You do not need all five every time. For quick questions, "task" alone is fine. But for anything you would call "real work," check you have covered at least Task + Context + Constraints. That alone transforms your hit rate.
Toggle between levels to see how the same request transforms as you add ingredients.
Live prompt builder
Same LinkedIn post request, three quality levels.
If your output is consistently disappointing, check if you are falling into one of these patterns.
Not telling Claude who it is
Same content. Completely different feedback. A role directs the lens Claude uses. Without one, you get generic, blandly positive feedback.
Saying "good" and expecting Claude to know what that means
"Better" means nothing on its own. Define what "good" looks like: more direct, warmer, more formal, shorter. Give Claude a direction, not a vibe.
Pasting one paragraph and expecting Claude to know the whole project
Out of context, every paragraph reads the same to Claude. Give it the surrounding material so it can make changes that actually fit your wider piece.
Writing a 500-word prompt when a question would do
You do not need to guess what context matters. Let Claude ask. Clicking answers is faster than writing a perfect prompt from scratch.
Letting Claude write whatever it wants when you have a strong voice
Without a voice reference, Claude defaults to "AI house style" which sounds bland and generic. Give it real examples of your writing and explicit phrases to avoid. The output starts sounding like you.
If you remember one thing from this course, make it this.
Stick it at the end of any meaningful prompt. Claude pauses before writing, identifies what it needs to know, and presents you with specific clarifying questions, often as multi-select options or a clean numbered list.
You answer in under a minute. Then Claude writes the response using your actual input rather than its guess. The quality difference is significant and the effort difference is almost nothing.
Claude is unusually good at identifying what it does not know
Asked to "do X," Claude tends to do X with whatever assumptions seem reasonable. Asked to "ask first," Claude is unusually good at surfacing the assumptions it would otherwise make silently. This is one of Claude's standout traits: when you give it permission to think out loud and ask before executing, it does so well. Most other AI tools just power through. Use this.
Do not start with "build me X." Use Chat to align on what you want. Once it is clear, then build.
Chat turns are cheap. Builds are expensive. A 30-second alignment conversation saves you from a 30-minute rework cycle on a wrong-shaped output. This is the single most important workflow pattern for any task with real consequences.
Eight reusable patterns. Save them, modify them, use them constantly.
The drafter
When you need text written from scratch.
The editor
When you have a draft and need feedback.
The analyst
When you want insights from a file or data.
The summariser
When you want a tight summary.
The comparer
When you need to weigh options.
The explainer
When you want to understand something.
The artifact builder
When you want an interactive deliverable.
The reframer
When you are stuck or output is not working.
Before the quiz, remember
- Bad prompts get generic output, not because Claude is bad, but because "average" is the default when context is missing.
- Five ingredients matter: Role + Task + Context + Constraints + Format. You do not need all five every time, but check you have at least Task + Context + Constraints for real work.
- "Ask me clarifying questions before you start" is the most useful phrase. Claude is especially good at responding to it.
- Plan first, build last. Cheap chat turns, then expensive build. Never the other way around.
- Voice matters. Upload past examples plus a list of phrases to avoid, and Claude starts sounding like you, not like AI.
Test yourself
Five questions on prompt patterns. Pick the best answer in each case.
Well done!
You scored 0/5. Mark this module complete below.
Set Up Your Workspace
Most people re-explain themselves to Claude every conversation. They do not have to. Projects and Cowork solve this exact problem.
The fastest way to make Claude useful is to make it stop forgetting.
By default, every Claude conversation is stateless. New chat, blank slate. You re-paste the same context, re-explain who you are, re-upload the same files. Most users live with this and never realise there is a better way.
There are three ways to give Claude persistent context: Projects (saved workspaces in claude.ai), Cowork folders (a local folder structure on your desktop that Claude reads), and Memory (an opt-in feature that lets Claude remember things across all your chats). Each solves a different version of the same problem.
Setting up your workspace is a one-time job that pays off on every future task. The hour you invest now saves you a few minutes on every prompt for the next year.
Each solves a different scope of "remember this for me."
| Method | What it remembers | Best for |
|---|---|---|
| Projects | Files you upload, custom instructions, the conversations within that project. Each project is its own bubble. | Recurring topics with their own reference material. Weekly reports, ongoing client work, a book you are writing. |
| Cowork folder | Whatever is in the folder you point Claude at. Read-only context plus an output folder Claude can write to. | Real desktop work. Document drafting, spreadsheet analysis, anything where output needs to live in your normal folder structure. |
| Memory (opt-in) | Personal facts, preferences, ongoing context across every chat in your account. Globally available. | Persistent identity: your role, your style preferences, names you reference often. Less for project-specific material. |
Projects beat re-uploading. Always.
If you find yourself uploading the same file to a fresh chat for the third time, stop. Make a Project, upload the file once at the project level, and every conversation in that project will have access to it. This single habit saves more time than almost any other Claude tip.
For Cowork users, the folder structure you choose is the difference between Claude feeling like a colleague and feeling like a confused intern.
The recommended pattern is one parent folder (typical name: Claude-Cowork) with four subfolders. Only one folder is where Claude writes to. Everything else is read-only context.
| Folder | Purpose |
|---|---|
| ABOUT-ME | Your identity. Who you are, what you do, how you write, what matters to you right now. Claude reads this on every session. |
| PROJECTS | One subfolder per live project. Briefs, drafts, references, supporting material. Read-only, Claude does not write here. |
| TEMPLATES | Your best past work, kept as reusable structure (not content). Claude refers to them when asked to match style or format. |
| CLAUDE-OUTPUTS | The only folder Claude writes to. Everything else is read-only. Keeps your real files safe from accidental edits. |
When you start Cowork, point it at the parent folder. Claude reads everything as context, but only writes inside CLAUDE-OUTPUTS. You move things over to your real working files when you are happy with them.
Most of the value of a structured workspace comes from two markdown files in your ABOUT-ME folder.
about-me.md
Who you are, what you do, who you work for, what you are working on right now, how you make decisions, what you care about. Treat it like the briefing you would give a new assistant on day one. Update it monthly.
anti-ai-style.md
Every phrase, word, or pattern Claude must never use in your output. Examples: "delve," "boundaries," "it is worth noting," "navigate the complexities of," any em-dashes, any sentence starting with "Moreover." This is your style filter. The longer this list, the more your output sounds like you.
One great markdown file beats fifty random uploads
Quality of context beats quantity. A well-written about-me.md is worth more than 20 random documents you uploaded "just in case." The signal-to-noise ratio matters: if Claude has to wade through unrelated material to find what is relevant, output suffers. Curate ruthlessly.
They overlap, but each has clear strengths.
| Situation | Use Projects | Use Cowork |
|---|---|---|
| Working on the train, on mobile, off your usual machine | Yes - cloud-based, anywhere | No - desktop only |
| Output needs to land in your existing folder structure | No - manual download | Yes - writes directly to disk |
| Sharing the workspace with a teammate | Yes on Team plans | No - your local machine |
| Files are sensitive and should not leave your computer | No - uploaded to Anthropic | Yes - stays local |
| Recurring topic with stable reference material | Yes - what Projects are designed for | Yes, but more setup overhead |
| One-off complex task on real local files | OK if you upload them | Yes - faster, no upload step |
Before the quiz, remember
- Three persistence options: Projects (cloud, per-topic), Cowork folder (local desktop), Memory (global personal facts).
- Projects beat re-uploading. If you have uploaded the same file twice, make a Project.
- Cowork folder structure matters: parent folder with ABOUT-ME, PROJECTS, TEMPLATES, CLAUDE-OUTPUTS. Only the last is writable.
- Two core files do most of the work:
about-me.mdfor identity,anti-ai-style.mdfor voice filtering. - One great context file beats fifty random uploads. Quality over quantity.
Test yourself
Five questions on workspaces, Projects, Cowork, and persistent context.
Great work!
You scored 0/5. Setting up a workspace once pays off forever.
Build Real Things with Artifacts
Most people use Claude for text. The biggest underused capability is its ability to build real, interactive deliverables you can actually use or share.
It is one of Claude's biggest features and one of its most invisible.
Artifacts are interactive content that renders inline in your conversation. Charts you can click on. HTML pages that work. React components you can interact with. SVG diagrams. Mermaid flowcharts. Working prototype apps. Real .docx, .xlsx, .pptx, and .pdf files you can download.
The reason most people never use Artifacts is that they do not know to ask for them. Claude defaults to text in chat. If you want something interactive, you need to say so. Once you start asking for Artifacts intentionally, the value-per-prompt jumps significantly.
Artifacts are available in Chat and Projects on any paid Claude plan. They are not a separate surface, they are a kind of output Claude can produce when you ask for it.
A guide to what each type is best at, with example prompts.
| Type | What it is | Best for |
|---|---|---|
| Interactive HTML | A working web page rendered in your chat. Buttons, forms, animations all work. | Decision tools, calculators, prototype interfaces, simple games, interactive explainers. |
| React components | Functional React components with hooks, state, charting libraries (recharts, lucide). | Dashboards, data viz with interactivity, mini apps, interactive demos. |
| SVG graphics | Vector graphics inline. Logos, icons, infographics, illustrations. | Quick infographics, custom diagrams, simple illustrations, presentation graphics. |
| Mermaid diagrams | Flowcharts, sequence diagrams, gantt charts written as text. | Process flows, system diagrams, decision trees, project timelines. |
| Markdown documents | Long-form structured text. Easy to copy or download. | Reports, articles, briefing notes, anything text-heavy you want as a clean document. |
| Real .docx / .xlsx / .pptx / .pdf | Native files Claude generates and you download. | Deliverables for stakeholders who need real Office files, not screenshots. |
The phrasing matters. Specific phrases turn vague text output into specific Artifact output.
| What you want | What to say |
|---|---|
| An interactive chart | Create an interactive chart of [data]. Don't code, just the chart. Use recharts. |
| A React mini-app | Build a [thing] as a React component. Tailwind for styling. Make it interactive. |
| A flowchart / decision tree | Show this process as a Mermaid flowchart. Include decision diamonds for branches. |
| An infographic | Create an SVG infographic showing [concept]. Use [colour palette] and clear icons. |
| A real Word doc | Create a .docx file with [content]. Proper headings, professional formatting, ready to send. |
| A real spreadsheet | Build me an .xlsx file with [structure]. Working formulas, conditional formatting, sample data. |
| A real PowerPoint | Make a .pptx deck for [audience and purpose]. Five slides, clean design, speaker notes on each slide. |
Specify the format upfront
Claude responds reliably when you name the output type. "Make me a chart" produces different results to "Create a recharts bar chart as an Artifact." Be specific about the type, the framework or format, and what should be in it. The more precise you are, the better the first output.
Most great Artifacts come from a build, then refine, then refine cycle. Here is how to keep iterations clean.
Start broad, refine narrow
First prompt: "Build me a dashboard for tracking volunteer hours." See what Claude produces. Then refine: "Add a filter for campus." Then: "Switch the bar chart to a stacked bar." Each refinement should change one thing.
Reference the Artifact directly
Instead of "make this better," say "in the dashboard, change the colour of the totals row to teal." Claude knows which Artifact you mean and edits it in place rather than starting over.
Reset rather than over-iterate
If an Artifact has gone wrong after several rounds of edits, do not keep patching. Start a new conversation, paste a screenshot or description, and rebuild fresh. Cleaner result than ten more "actually try this instead" prompts.
Download or share the final version
Artifacts have a download button. Real files (.docx, .xlsx, .pptx) save to your computer. HTML/React Artifacts can be downloaded as standalone files you can host or share. Some Pro features let you share an Artifact via a public link.
Artifacts are powerful, but they are not always the right tool. Knowing when to skip them saves you time.
You need it inside an existing Word or Excel file
An Artifact is a fresh file. If you need to add content to your existing document, use Cowork to point Claude at the file and edit it in place. Or use Copilot inside that Office app. Artifacts are great for creating from scratch, less great for inserting into existing files.
You need a long, polished document
For genuinely long-form text (a 5000-word report, a book chapter), Claude in plain chat with a Project for context tends to be cleaner than building it as an Artifact. Use Artifacts for structured documents, dashboards, and interactive output. Use chat for prose.
You want production-ready code
React Artifacts are great for prototypes and demos. For real shipping code that needs to integrate with your codebase, use Claude Code in your terminal instead. It can read your repo, run tests, and commit changes. Artifacts are isolated demos, not part of your build.
Before the quiz, remember
- Artifacts are interactive output rendered inline. Charts, apps, diagrams, real Office files.
- Most people never use them because they don't think to ask. Specifying the format upfront unlocks them.
- Six main types: Interactive HTML, React components, SVG, Mermaid diagrams, Markdown docs, real Office files.
- Iterate one change at a time. Reference the Artifact directly when refining.
- Reset rather than over-iterate when an Artifact has gone wrong. Fresh conversation, fresh build.
- Use Cowork or Claude Code instead when you need work in your existing files or codebase.
Test yourself
Five scenarios. Pick the right Artifact type and approach for each.
Nicely done!
You scored 0/5. You're halfway through the course.
Automate with Cowork, Claude Code, and MCP
The shift from "Claude answers questions" to "Claude does tasks." This is where real hours start being saved.
Each is built for a different scope of work. Picking the right one matters.
Cowork
The desktop app that reads and writes files in a folder you specify. Best for sustained work on real documents: drafting, editing, analysis, refactoring documents, generating real .docx, .xlsx, .pptx, and .pdf files in your normal folder structure.
- "Read every file in PROJECTS/Q4-report. Build me a summary in CLAUDE-OUTPUTS."
- "Clean up this messy spreadsheet, fix the dates, flag duplicates, save the cleaned version next to it."
- "Take the bullet points in notes.md and produce a polished 5-page Word document."
- "Refine this 20-slide deck across multiple passes until every slide has a clear point and consistent formatting."
Claude Code
A command-line tool for developers. Run claude in your terminal and it can read your codebase, write new files, run tests, execute commands, and commit changes. The closest Claude gets to "pair programmer."
- "Read the codebase. Find every place we handle authentication and check for inconsistencies."
- "Add a feature: users can export their data as JSON. Write the endpoint, the tests, and the docs."
- "This bug report mentions a crash on file upload. Find the cause and propose a fix."
- "Refactor src/utils into smaller files. Keep all tests passing."
MCP servers
The Model Context Protocol lets Claude connect to external tools: your email, calendar, project trackers, databases, search engines, custom APIs. Once connected, Claude can read from and write to those systems as part of a normal conversation.
- Connect Gmail/Calendar: "What is on my calendar this week, and which emails need a response by Friday?"
- Connect Linear/Jira: "Summarise every open ticket assigned to me, grouped by priority."
- Connect Notion: "Find every page tagged 'roadmap' and summarise the changes from the past month."
- Connect a custom DB: "Run a query for active users last 30 days, give me the top 10 by usage."
Automation is intoxicating. Before you commit to building one, check the task is actually a good candidate.
| Automate when... | Do not automate when... |
|---|---|
| The task is repeatable with the same inputs and outputs each time. | Every instance is subtly different and requires judgement. |
| It takes 10+ minutes per occurrence, at least weekly. | It happens occasionally and does not take long. |
| The cost of getting it wrong is low or easily caught. | A mistake would damage relationships, miss compliance, or be unrecoverable. |
| The process is well documented or easily described. | The process lives in someone's head and nobody has written it down yet. |
| You can review the output before it goes live. | The output goes straight to external parties with no review. |
Document the process first, automate second
The single biggest predictor of whether an automation will succeed is whether the process it replaces was clearly documented beforehand. If you cannot describe the workflow in plain English to another human, Claude will not be able to automate it either. Write it out, test it as a checklist a person could follow, then automate.
A real, simple automation you can build today using Cowork. The pattern generalises.
Pick a small, annoying task
Examples: weekly report drafted from standing inputs, email replies based on a stable template, tidying a recurring spreadsheet, generating a board pack from monthly data. Not "run my whole job," just one chunk that recurs.
Describe it in plain English
"Every Friday I take the data in weekly-numbers.xlsx, write a 200-word summary highlighting biggest changes, and email it to the team. Generate the summary as a draft I can review before sending."
Set up the folder structure
Put the standing inputs in a Cowork PROJECTS subfolder. Put your past summaries in TEMPLATES so Claude can match the style. Put your style rules in ABOUT-ME. Run Cowork pointed at the parent folder.
Run it weekly with the same prompt
"Read this week's weekly-numbers.xlsx. Match the tone of past summaries in TEMPLATES. Generate this week's draft summary in CLAUDE-OUTPUTS." Open it, review, send.
Refine the prompt over time
Each week, note what you changed when reviewing. Update about-me.md with the rule that prevents that change next time. After a month, the draft needs almost no editing.
MCP is powerful and a little risky. A short checklist for getting it right.
Start with read-only connectors
Connect things that only read your data first (calendar, email read-only, search). See what value Claude adds before you give it write access. The risk profile is much lower for read-only.
Use OAuth, not API keys, where possible
OAuth means Claude gets scoped access tied to your identity, with revocation in one click. API keys are wider and harder to revoke if something goes wrong.
Approve actions individually at first
Claude asks for confirmation before performing actions on connected services. Keep that on for at least the first month. Loosen only for actions you have seen go right consistently.
Check the connector's permissions before adding
Each MCP connector requests specific permissions. Read them before granting. "Send email on your behalf" is much wider than "read your inbox." Match the scope to what you actually need.
Before the quiz, remember
- Three automation surfaces: Cowork (real files), Claude Code (codebases), MCP (external services).
- Match the surface to the task: docs/spreadsheets to Cowork, code to Claude Code, integrated workflows to MCP.
- Document before you automate. If you cannot write the process down, Claude cannot follow it either.
- Start small. One annoying weekly task, not the whole job. Iterate.
- MCP safely: read-only first, OAuth over API keys, approve actions individually until you trust the pattern.
Test yourself
Five questions on choosing the right automation surface and using it safely.
On a roll!
You scored 0/5. Two modules to go.
Power Tips and Stretching Your Plan
Small habits that separate light users from heavy users. Most take 30 seconds to pick up and pay back forever.
Daily habits that compound
/ to reference an earlier conversation. Useful when you want to pick up where you left off without uploading the same files again.Feature discoveries most people miss
Most "Claude is broken today" moments have a specific, fixable cause. Diagnostic order to work through:
Output is generic or vague
Check: are you in the right surface? If you are in Chat asking about specific files, move them to a Project or open Cowork. Then check: have you given audience, constraints, and format? If not, add them. Then check: are you on Sonnet when you needed Opus + Extended Thinking?
Output sounds like AI, not like you
You have not given Claude a voice reference. Upload three of your past pieces and instruct it to match tone, sentence length, and structure. Build out your anti-ai-style.md with phrases to ban. Both compound over time.
Output references something that never happened
Claude is hallucinating, usually because it was underfed on context. Upload the actual source material. Ask "cite the source" to force it to link claims to real data. If it cannot cite, it was guessing. Add files instead of letting it improvise.
Output cuts off mid-thought
Claude hit a token limit in its response. Either ask "continue" (it will pick up where it left off) or break the task into smaller chunks. Long Artifacts especially: build them in pieces and stitch together rather than asking for one giant build.
Output is technically correct but not useful
You are probably asking the wrong question. Step back: "I have been trying to [goal] and getting stuck. Ask me five questions to understand the real problem, then suggest a different approach." Claude will surface assumptions you are making.
Practical ways to get more from a Pro plan before paying for Max.
| Habit | Why it stretches your usage |
|---|---|
| Default to Sonnet, not Opus | Sonnet uses far less of your quota. Most tasks do not need Opus. Reserve Opus for hard reasoning. |
| One Project per topic, not one per task | Avoids re-uploading the same files. Claude reads project files once, not every conversation. |
| Plan in Chat, build in Cowork | Cheap Chat conversations align the spec, then the build is one clean run rather than ten messy iterations. |
| Use the API for batch jobs | The API has its own usage allowance separate from claude.ai. Useful for repeated automated tasks. |
| Reset stale chats | Long conversations carry old context that confuses Claude and wastes tokens. Start fresh when a thread has gone wrong. |
Before the quiz, remember
- Default to Sonnet, escalate to Opus + Extended Thinking only when needed.
- Projects = Claude's long-term memory. Use them for anything recurring.
- "Ask me clarifying questions" is faster than writing a perfect prompt yourself.
- Voice references plus phrases-to-avoid turn AI output into your output.
- Reset stale chats rather than over-iterating. Fresh context, cleaner output.
- The desktop app is faster than the browser and has Cowork built in. Install it.
Test yourself
Five questions on power-user habits, troubleshooting, and stretching your plan.
Almost there!
You scored 0/5. One module left.
Go Deeper
Official Anthropic learning, the prompt library, and community resources. Bookmark what is relevant to your role.
Before the final quiz
- Anthropic's docs are excellent and updated constantly. Bookmark
docs.claude.comas your first stop. - The prompt library is full of patterns you can adapt without writing from scratch.
- The MCP ecosystem is growing fast. New connectors appear weekly. Worth checking quarterly for new options.
- For coding, Claude Code documentation is essential, not optional.
- Subscribe to Anthropic's newsroom so model and feature changes do not catch you out.
Test yourself
Five questions pulling together everything from the course.
You did it!
You scored 0/5. Mark this module complete to finish the course.
Course complete
You've worked through all seven modules. You now know more about Claude than the vast majority of people who use it. Want help rolling this out across your team?
