> ## Documentation Index
> Fetch the complete documentation index at: https://anthropic-training.neuroscale.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Lab: Start using Claude

> Open Claude, learn the chat controls, and run your first real prompts.

This is your first hands-on pass in Claude itself. You are not diagnosing failures yet. You are learning the room: where chats live, how modes and models change the reply, how files and connectors feed context, and how research reaches the open web.

Plan for about 25 minutes. You need a Claude account signed in at [claude.ai](https://claude.ai) or in the desktop app. Either surface is fine for this lab.

<div className="lab-brief" aria-label="Lab overview: three parts">
  <div className="lab-brief__part"><span>Part 1</span><p>Open Claude and find the chat controls</p><i>6 min</i></div>
  <div className="lab-brief__part"><span>Part 2</span><p>Modes, models, files, connectors, research</p><i>10 min</i></div>
  <div className="lab-brief__part"><span>Part 3</span><p>Run sample prompts and compare</p><i>9 min</i></div>
</div>

## Part 1: Walk in

1. Open [claude.ai/new](https://claude.ai/new) or launch the desktop app and start a **new chat**.
2. Look left. That is the chat selector: recent conversations, Projects if you have any, and the control to start fresh. Click an old chat, then click **New chat** again so you feel the switch.
3. Look at the empty composer. Before you type anything, locate these controls. Do not skip any of them:

| Control | What you are looking for |
| - | - |
| Mode / style | Write, Learn, Code, and related options that change how Claude responds |
| Model selector | Which Claude model will answer this chat |
| Attach / plus | Files, screenshots, and other inputs you can add to the message |
| Connectors | Linked tools and data sources Claude can call |
| Research / web | Tools that let Claude search and read from the open web |

<Frame>
  <img src="https://mintcdn.com/anthropic-training/j8TXbg7sp94GkIsZ/images/landscape/claude-web-ui.png?fit=max&auto=format&n=j8TXbg7sp94GkIsZ&q=85&s=696db26c9c0440183f3370a38136e2f6" alt="The Claude chat interface with the composer, chat list, and main controls" width="1468" height="1108" data-path="images/landscape/claude-web-ui.png" />
</Frame>

Say the five control names out loud once. The rest of this lab is pressing each one on purpose.

## Part 2: Touch every lever

Stay in the same new chat. Work top to bottom.

### Modes: Write, Learn, Code

Open the mode or style control in the composer. Switch through **Write**, **Learn**, and **Code** (and any siblings your plan shows). Notice what the UI promises for each:

* **Write** shapes Claude for drafting and production: emails, docs, outlines, polish.
* **Learn** slows Claude into a teacher: questions, explanations, less “here is the finished answer.”
* **Code** tilts Claude toward software: snippets, debugging, implementation steps.

Leave it on **Write** for the next few steps. You will change modes again in Part 3.

### Model selector

Open the model picker. Read the names without choosing based on brand loyalty. Note which option is currently selected. For this lab, pick a balanced default if you have one (often Sonnet). You will change models later in the course on purpose; right now you only need to know where the switch lives.

### Attach a file

Click the attach or **+** control and upload one harmless file from your machine: a PDF, a short `.txt`, a screenshot, or a spreadsheet with nothing confidential. Then send:

> Summarize this in five bullets. Tell me what kind of file it is and who the likely audience is.

Confirm Claude used the attachment. If the reply could have been written without the file, ask a follow-up that only the file can answer.

### Connectors

Open the same attach / tools menu and find **connectors** (or open **Customize → Connectors**). Browse what is already available on your account. Do not connect a high-stakes system yet unless your facilitator says otherwise. For this lab:

1. Identify one connector you might use this week (Drive, Calendar, Slack, Notion, or similar).
2. If it is already connected, enable it for this chat and ask a read-only question, such as “What are my next three calendar events?” or “List the three docs I touched most recently.”
3. If nothing is connected, note the one you will wire later in [Connectors](/connectors/connectors). Still open the directory so you have seen the shelf.

### Research and web tools

In the tools area, find **Research**, web search, or the equivalent research toggle for your plan. Turn research or web access **on** for one turn and send:

> What changed in the last 30 days about Anthropic’s Claude product that would matter to a first-time enterprise user? Cite sources.

Watch for citations, tool calls, or a research progress state. Then turn research **off** and ask a follow-up that should stay inside the chat only. You are learning the difference between “answer from training” and “go look.”

## Part 3: Sample prompts

Run these three prompts in order. Change one control before each run so you feel the levers, not just the wording.

### Prompt A — Write mode, no research

Mode: **Write**. Research: off. Keep your attached file if it is still in the chat, or start a clean chat.

> Draft a 120-word kickoff note to my team about a 1-on-1 Claude training day. Tone: clear, calm, no hype. End with three things people should bring.

Edit two sentences in the reply yourself. Getting a draft is step one. Owning the voice is the job.

### Prompt B — Learn mode

Switch to **Learn**. Same topic, different ask:

> I am new to Claude. Teach me the difference between attaching a file and connecting Drive. Quiz me with three questions before you summarize.

Answer the quiz in the chat. Notice whether Claude holds back the full explanation until you engage.

### Prompt C — Code mode, then research

Switch to **Code**. Send:

> Write a tiny Python function that takes a list of meeting titles and returns only the ones containing "1:1" or "one-on-one", case-insensitive. Include a three-line docstring and one example call.

Then turn research or web tools **on** and ask:

> Is there a standard library feature I should prefer instead? Check current Python docs and say yes or no with a link.

You now have one drafting run, one teaching run, and one coding-plus-research run. That is enough to prove the controls are not decoration.

## Flight card

Check items off as you land them. All ten is a completed lab.

<div className="lab-card" aria-label="Lab completion checklist">
  <label><input type="checkbox" /> <span>Opened a new chat and used the chat selector to move between conversations</span></label>
  <label><input type="checkbox" /> <span>Found Write, Learn, and Code (or the equivalent mode control)</span></label>
  <label><input type="checkbox" /> <span>Opened the model selector and noted the active model</span></label>
  <label><input type="checkbox" /> <span>Attached a harmless file and got a file-aware answer</span></label>
  <label><input type="checkbox" /> <span>Opened connectors and identified one you might use this week</span></label>
  <label><input type="checkbox" /> <span>Ran at least one turn with research or web tools on</span></label>
  <label><input type="checkbox" /> <span>Completed Prompt A in Write mode</span></label>
  <label><input type="checkbox" /> <span>Completed Prompt B in Learn mode</span></label>
  <label><input type="checkbox" /> <span>Completed Prompt C in Code mode with a research follow-up</span></label>
  <label><input type="checkbox" /> <span>Edited one Claude draft so the final words are yours</span></label>
</div>

## Debrief with your facilitator

1. Which control felt most new: mode, model, attach, connectors, or research?
2. How did Learn feel different from Write on the same topic?
3. Name one real task from this week you would start in Write, one in Learn, and one with research on.
4. What will you attach or connect the next time you open Claude for actual work?


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.