Vault / course/courses/claude-101.md

updated 2026-06-25

Course: Claude 101

Mirrors: Anthropic Academy — Claude 101 · https://anthropic.skilljar.com/claude-101 Audience: Everyone. No technical background needed. · Time: ~60 min + project Prereqs: A claude.ai account (setup-checklist). · Backing notes: claude-overview, model-family, claude-ai-features Project: p01-first-conversations

This is the front door. By the end you can hold a productive conversation with Claude, pick the right model, use Projects and Artifacts, and know what Claude should and shouldn't be trusted with.

Learning objectives

After this course you can:

Explain what Claude is and where it lives (web, desktop, mobile, Code, API).
Start a conversation that gets a useful answer on the first or second try.
Choose between Opus, Sonnet, and Haiku for a task.
Use file uploads, Projects, and Artifacts in claude.ai.
Recognize Claude's limits (hallucination, knowledge cutoff) and verify accordingly.

Module 1 — Meet Claude

🎞 Frame 1 · What Claude actually is · ⏱ ~2 min

🎬 Scene — A chat window opens. You type "Explain photosynthesis to a 7-year-old," and a friendly, structured answer streams back.

🧠 Concept — Claude is a family of large language models from Anthropic, trained to be helpful, harmless, and honest. It predicts text, but it's tuned to follow instructions, reason carefully, and stay safe.

🖼 On screen

Claude is good at	Claude is not
Writing, summarizing, explaining	A search engine (no live web unless a tool gives it one)
Reasoning over text & images	A calculator / database of facts
Drafting & editing code	A source of guaranteed-true facts
Following multi-step instructions	Aware of events after its knowledge cutoff

✅ Checkpoint — Name one task Claude is great at and one it should not be trusted to do unaided.

🎞 Frame 2 · Where Claude lives · ⏱ ~2 min

🎬 Scene — The video pans across a phone app, a browser tab, a terminal, and a code editor — all Claude.

🧠 Concept — The same model family powers many surfaces. Pick the surface that fits the job.

🖼 On screen

claude.ai / desktop / mobile  → chat, Projects, Artifacts (everyone)
Claude Code                   → coding agent in your terminal/IDE (developers)
Claude API                    → build Claude into your own apps (developers)
Cowork                        → Claude works alongside you on tasks (everyone)
Bedrock / Vertex AI           → Claude inside AWS / Google Cloud (enterprises)

✅ Checkpoint — Which surface would a non-coder use to summarize a 40-page PDF? (claude.ai with a file upload.)

🎞 Frame 3 · A conversation, not a search box · ⏱ ~2 min

🎬 Scene — Side by side: a one-word Google search vs. a full-sentence request to Claude that gets a tailored answer.

🧠 Concept — Claude rewards context and intent. Tell it the goal, the audience, and the format, not just keywords.

🖼 On screen

❌ "marketing email"
✅ "Write a 120-word marketing email announcing our new
    yoga studio's free trial week. Friendly, no jargon,
    end with a clear call to action."

⚠️ Gotcha — Vague in, vague out. The single biggest beginner win is adding context.

✅ Checkpoint — Rewrite "fix my resume" into a request that gives Claude goal + audience + format.

Module 2 — Talking to Claude well

🎞 Frame 4 · The four levers of a good prompt · ⏱ ~3 min

🎬 Scene — A prompt gets visibly better as four labels are added: Role, Task, Context, Format.

🧠 Concept — Most good prompts set four things. You don't always need all four, but reaching for them fixes most bad answers. (Deeper: prompt-engineering-basics.)

🖼 On screen

Lever	Question it answers	Example
Role	Who should Claude be?	"You are a patient math tutor."
Task	What exactly to do?	"Explain long division."
Context	What does it need to know?	"The student is 9 and hates fractions."
Format	What should output look like?	"Three short steps, then one practice problem."

✅ Checkpoint — Identify the missing lever: "You are a chef. Give me a recipe." (Context — for whom? what ingredients? — and Format.)

🎞 Frame 5 · Iterate, don't restart · ⏱ ~2 min

🎬 Scene — Instead of rewriting from scratch, the user replies "shorter, and add a vegetarian option" and the answer improves.

🧠 Concept — Claude remembers the current conversation. Refine in follow-ups: "make it shorter," "more formal," "show your reasoning." This is faster than re-prompting.

🖼 On screen

You: Draft a cover letter for a barista job.
Claude: <draft>
You: Tighten it to 150 words and sound less generic.
Claude: <better draft>

✅ Checkpoint — Give two follow-up phrases you'd use to steer a too-formal answer.

🎞 Frame 6 · Give Claude your documents · ⏱ ~2 min

🎬 Scene — A PDF is dragged into the chat; Claude answers questions grounded in it.

🧠 Concept — Uploading files (PDF, images, spreadsheets, docs) lets Claude work from your material instead of its memory — which reduces hallucination.

🛠 Try it now — Upload any PDF and ask: "What are the three most important points, with the page they're on?"

✅ Checkpoint — Why does uploading a document make answers more trustworthy than asking from memory?

Module 3 — Models, Projects, and Artifacts

🎞 Frame 7 · Opus, Sonnet, Haiku · ⏱ ~3 min

🎬 Scene — Three lanes on a track: a heavy-lifter, an all-rounder, and a sprinter.

🧠 Concept — Claude comes in tiers. Trade capability vs. speed/cost. (Deeper: model-family.)

🖼 On screen

Model	Best for	Feel
Opus	Hardest reasoning, deep analysis, agents	Smartest, slower, priciest
Sonnet	Most day-to-day work	Balanced default
Haiku	High-volume, simple, latency-sensitive	Fastest, cheapest

⚠️ Gotcha — Don't reflexively pick the biggest model. Sonnet handles most tasks; reach for Opus when it actually struggles.

✅ Checkpoint — Which tier for "classify 10,000 support tickets by topic"? (Haiku — simple + high volume.)

🎞 Frame 8 · Projects: give Claude lasting context · ⏱ ~2 min

🎬 Scene — A "Project" sidebar holds documents and custom instructions that every chat in it can see.

🧠 Concept — A Project is a workspace with shared knowledge (files) and custom instructions, so you don't re-explain yourself each chat. (Deeper: claude-ai-features.)

🖼 On screen

Project: "Q3 Marketing"
  ├─ Knowledge: brand guide.pdf, past campaigns.csv
  ├─ Instructions: "Always match our friendly brand voice."
  └─ Chats: launch email · blog post · ad copy …  (all share the above)

✅ Checkpoint — What two things does a Project remember across chats? (Knowledge files + custom instructions.)

🎞 Frame 9 · Artifacts: living output · ⏱ ~2 min

🎬 Scene — Claude generates a document/app in a side panel you can edit and iterate on, separate from the chat.

🧠 Concept — Artifacts are standalone outputs (a doc, a table, a small web app) Claude builds in a side panel and revises in place — great for anything you'll keep using.

✅ Checkpoint — Name something better as an Artifact than a chat reply. (E.g., a reusable checklist, a small interactive tool.)

Module 4 — Working safely and well

🎞 Frame 10 · Trust but verify · ⏱ ~2 min

🎬 Scene — Claude gives a confident answer; the user double-checks a key fact before using it.

🧠 Concept — Claude can be confidently wrong (hallucinate) and has a knowledge cutoff. For facts, dates, numbers, and anything high-stakes: verify, or give Claude the source.

🖼 On screen

High-stakes? → Provide the source OR verify the output.
Recent event? → Claude may not know it; give it the info.
Numbers/quotes? → Spot-check.

✅ Checkpoint — Two situations where you must verify Claude's output before acting on it.

🎞 Frame 11 · Privacy & good inputs · ⏱ ~2 min

🎬 Scene — A reminder card: think before pasting secrets; check your org's data policy.

🧠 Concept — Don't paste secrets/credentials or regulated data unless your plan and policy allow it. Better inputs (clear, complete, well-scoped) produce better, safer outputs.

✅ Checkpoint — What kind of data should you not paste into a chat without checking policy first?

🎞 Frame 12 · You've got the basics · ⏱ ~1 min

🎬 Scene — A recap slide: converse with context, iterate, upload, pick a model, use Projects/Artifacts, verify.

🧠 Concept — You can now get real work done with Claude. Next: make your prompts reliably great in track-2-effective-prompting.

✅ Checkpoint — Without looking, list the four prompt levers from Frame 4.

🛠 Project

Complete p01-first-conversations — Your First 10 Conversations. You'll run a set of real tasks (summarize, draft, plan, analyze a file), apply the four levers, and keep a before/after of one prompt you improved.

🧪 Self-check quiz

What do the letters in "helpful, harmless, honest" describe?
Name the four prompt levers.
You need to summarize 50 long PDFs cheaply and fast. Which model tier?
What does a Project remember that a one-off chat doesn't?
Give one reason to upload a document instead of asking from memory.
What two limits make verification necessary?
True/false: the biggest model is always the right choice.

<details><summary>Answers</summary>

Claude's design goals / training objectives. 2. Role, Task, Context, Format. 3. Haiku. 4. Shared knowledge files + custom instructions. 5. Grounds the answer in your material, reducing hallucination. 6. Hallucination and knowledge cutoff. 7. False — match the model to the task.

</details>

🎓 Certificate criteria

You've "passed" Claude 101 when you can:

Hold a multi-turn conversation and improve the output with follow-ups.
State when you'd use Opus vs. Sonnet vs. Haiku.
Set up one Project with knowledge + instructions.
Complete p01-first-conversations and journal one prompt you improved.

Tick this course off in progress and record the date you earned Anthropic's official Claude 101 certificate.

🔗 Sources & deeper notes

Official course: https://anthropic.skilljar.com/claude-101
Vault notes: claude-overview, model-family, claude-ai-features, prompt-engineering-basics
Next course: ai-fluency-framework-foundations · Next track: track-2-effective-prompting