Your First AI Hire

Your first AI hire.

This is your workbook for today. Six short labs, each one giving Claude something you'd give any new hire. Every prompt has a copy button, and every lab has a fix for the usual snags.

Before Lab 1

  1. Sign in to Claude with your Pro account. Use the desktop app if you have it. If not, claude.ai in your browser works for Labs 1 to 4.
  2. Turn on file creation. Open Settings, then Capabilities, and switch on Code execution and file creation. It lets Claude make real files, like spreadsheets, and skills need it.
  3. Choose Sonnet 5.5. Open the model menu in the message box and pick Sonnet 5.5. Avoid anything named Fable; on the Pro plan it costs extra from the first message.
  4. Unzip the starter kit. On Windows, right-click the downloaded file, choose Extract All, and extract it to your Desktop. On a Mac, double-click it in Downloads, then drag the first-ai-hire-main folder to your Desktop. You need the kit for Labs 2, 3, 5, and 6.
About your data. Your own email and calendar are fine for today. Keep customer and confidential files out until you've checked your company's AI policy. On a personal Pro plan you choose whether your chats help train Anthropic's models: Settings, Privacy, "Help improve our AI models."

Your screen may look a little different from your neighbor's. Anthropic is switching accounts to a combined version of Claude a few at a time. The steps work either way, and where they differ, both are shown.

Lab 1 of 6, about 8 minutes. Works in any browser.

Give it a briefing.

You'll have: a Project where Claude already knows who you are, what you do, and how you like to work.

  1. In Claude's sidebar, click Projects, then New project. Name it My AI hire.
  2. Start a chat inside the project and paste this:
    Interview me so you can write a one-page briefing about me that any AI assistant could use to help me at work.
    
    Ask one question at a time, six in total. Cover my role and company, who I serve, my biggest recurring tasks, how I like to communicate, and what I'm focused on this quarter. Push for specifics and real examples.
    
    When we're done, write the briefing in the second person ("You are helping a...") under clear headings. Keep it under 400 words.
  3. Answer the six questions. Typing is fine. If you'd rather talk, the Claude desktop app has a microphone button in the message box, and you speak about three times faster than you type.
  4. Copy the briefing Claude writes. In the project, open Instructions (sometimes "Set project instructions"), paste it in, and save.
  5. Start a new chat in the project and test it:
    Based on what you know about me, what's one piece of work I could hand to you every week? Be specific to my job, not generic.
It worked if the answer uses details from your interview without you repeating them. Keep the briefing open in a tab; you'll paste it again in Lab 4.
Running out of time mid-interview?

Type: "Write the briefing now with what you have, and mark the gaps." You can fill the gaps later.

I can't find Instructions in my project

Look on the right side of the project page, or in the project's menu. If you can't find it, save the briefing as a file and add it to the project. Claude reads project files in every chat.

Bonus: Ask it to draft a reply to the hardest email you got this week. Paste the email in. Notice how much less you have to explain.

Lab 2 of 6, about 7 minutes. Works in any browser.

Give it the keys.

You'll have: Claude connected to your Google Calendar, Gmail, and Drive, so it can see your actual week.

  1. In the sidebar, open Customize, then Connectors, then click +.
  2. Add Google Calendar. Sign in with Google and approve. Repeat for Gmail and Google Drive.
  3. Go back to your My AI hire project and start a new chat. Click + in the message box, hover over Connectors, and make sure all three are on.
  4. Paste this:
    Look at my calendar for the next five days and my email from the last three days. Using what you know about me, tell me:
    
    1. What am I forgetting or about to drop?
    2. Which meeting needs the most prep, and what should I bring?
    3. Which one email should I answer first, and why?
    
    Read only. Don't send, change, or delete anything.
It worked if it names real meetings and real emails from your week.
Google blocked your work account? Use the pretend week

Many work Google accounts block new apps until a Google admin marks Claude as trusted. You can't fix that in the room, so use a pretend week instead:

  1. Open the starter kit, then practice-files, then sample-week.
  2. Add calendar-this-week.md and inbox-last-3-days.md to your project, or drag them into the chat.
  3. Type "Pretend this calendar and inbox are mine," then paste the prompt above.

Later, send your IT team the email on the pre-work page. A personal Google account also works.

My company uses Outlook and Microsoft 365

The Microsoft 365 connector is in the same Connectors list, but it needs a work Microsoft account, and a Microsoft admin has to approve it once for your company. Personal Outlook.com and Hotmail accounts won't work. For today, use the pretend week above.

Is this safe?

Claude uses your permissions and nothing more. By default it asks before it sends, shares, moves, or deletes anything, and the prompt above tells it to read only. You can disconnect any connector in Customize, Connectors.

Bonus: "Find the most recent document in my Drive about [a project]. Summarize where it stands in five bullets."

Lab 3 of 6, about 10 minutes. Works in any browser.

Give it playbooks.

You'll have: a ready-made morning-brief skill, and one skill you wrote for your own job.

A skill is a recipe card: a text file with a name, a one-line description of when to use it, and the steps. Claude scans the descriptions of all your skills and pulls out the full recipe only when your request matches, so you don't have to remember it exists.

  1. Check that Code execution and file creation is on in Settings, Capabilities. Skills don't work without it.
  2. In Claude, open Customize, then Skills. Click +, then Create skill, then Upload a skill. Choose morning-brief.zip from the starter kit's skills/zips folder. Upload the ZIP file, not the folder.
  3. Now write your own. Pick something you do every week, then paste this and fill in the blank:
    Help me create a skill for ______ (something I do every week, like writing my weekly update, prepping for one-on-ones, or comparing vendor quotes).
    
    Ask me three questions, one at a time: how I do it today, what a great result looks like, and the mistakes to avoid. Then write the skill and package it as a ZIP file I can upload.
  4. Download the ZIP Claude makes, upload it the same way as step 2, and try it once.
It worked if morning-brief and your own skill both show in Customize, Skills, and your skill does its job when you ask in plain words. You'll see morning-brief in action in Lab 4.
No connectors? Try the decision-memo skill

Upload decision-memo.zip the same way, then type: "Help me decide whether we should [a real decision you're facing]." It interviews you and writes a one-page memo. No email or calendar needed.

The upload failed

Upload the ZIP file, not the folder or the SKILL.md file by itself. The folder inside the ZIP has to have the same name as the skill. If you edited a skill, don't add new lines between the two --- lines at the top.

Where do other people's skills come from?

You can browse skills inside Claude: Customize, Skills, +, Browse skills. Anthropic also publishes skills on GitHub (linked at the bottom of this page). This kit is the same idea: someone shares the recipe in public, and you take it and make it yours. That's open source.

Bonus: Upload meeting-prep.zip and try "Prep me for my next meeting." Or open skills/write-like-me/SKILL.md in any text editor, fill it in with three short things you've actually written, zip the folder, and upload it. Now everything Claude drafts sounds more like you.

Lab 4 of 6, about 5 minutes. Works in any browser.

Give it a schedule.

You'll have: a morning brief that writes itself every weekday at 7, before you sit down.

  1. Click Scheduled in the sidebar, then New task, then Set up manually. Name it Morning brief.

    On the new combined Claude, there may be no Scheduled button. Describe the task in any chat instead ("Every weekday at 7 a.m., run my morning brief"), paste the prompt below, and click Schedule when it offers.

  2. Paste this as the prompt, then paste your briefing from Lab 1 at the bottom:
    Use my morning-brief skill to write today's brief from my Google Calendar and Gmail.
    
    Read only. Don't send, archive, or delete anything.
    
    Here's who I am:
    [paste your briefing here]
  3. Set it for weekdays at 7:00 a.m. If there's a model choice, pick Sonnet 5.5. Leave the folder empty, so it runs in the cloud even when your laptop is closed.
  4. Save it, then click run now. If it asks for approval to read your calendar or email, approve it.
It worked if you get a brief in four parts. Tomorrow's will be waiting under Scheduled. For the first week, set a 7:05 a.m. phone reminder that says "Read my AI brief," until it's a habit.
No connectors? Schedule an industry brief instead
Every weekday, search the web for news from the last 24 hours about my industry, my company, and my biggest competitors. Give me the five items that matter most to someone in my role, one line each, with why it matters and a link.

Here's who I am:
[paste your briefing here]
Why paste the briefing again?

A scheduled task you create from the sidebar may not see your Project's instructions. Pasting the briefing in guarantees it knows who you are.

Bonus: Upload weekly-update.zip, then add a Friday 3 p.m. task: "Use my weekly-update skill to draft this week's update from my sent email and calendar. Show it to me so I can review it before I send it."

Lab 5 of 6, about 8 minutes. Needs the Claude desktop app, with a chat version below.

Give it a desk.

You'll have: a messy folder cleaned up, an expense spreadsheet, and a to-do list, made by Claude working directly on your files.

The practice-files/messy-folder in the starter kit holds 16 files from a made-up business trip to Denver: receipts, two versions of a vendor quote, phone notes, and a duplicate. Nothing in it is real.

  1. Open the Claude desktop app and check the model is Sonnet 5.5. If you see a Cowork option, choose it. On the new combined Claude, just start a new task in the main box.
  2. Give it the folder: click Add folder in the row under the message box (or drag the folder in), and choose practice-files/messy-folder.
  3. Paste this:
    This folder is a mess from a business trip. Please:
    
    1. Find duplicate files and move them into a folder called "Duplicates."
    2. Sort everything else into folders: Receipts, Quotes, Notes, Other.
    3. Rename each file so the name says what it is and the date, like "2026-09-15 Lyft airport to hotel.txt."
    4. Build a spreadsheet of what the traveler personally paid for the Denver trip, September 15 to 17, with a total.
    5. Write a short to-do list from the notes, most urgent first.
    
    Show me your plan before you start. Don't delete anything.
  4. Read the plan. If it looks right, tell it to go.
  5. Open the spreadsheet and check the total. Then ask:
    Compare the two strap quotes. Which should I accept, how much does it save, and are they still valid?
It worked if the travel total (flight, hotel, one Lyft ride, client dinner, parking) is $1,480.49.
If yours says $1,548.69, it counted the duplicate Lyft receipt twice.
If it's about $4,065, it included the expo booth invoice, which the company paid in August.
Either way, tell it what's wrong and watch it fix the sheet. And did it notice that both strap quotes have expired? That's the habit: hand it off, then check one thing.
No desktop app, or no Cowork? Do it in a chat

Drag all 16 files from the folder into a chat, paste the same prompt, and ask for the spreadsheet as a download. You won't get renamed files, but you'll get the report, the check, and the to-do list.

It keeps asking me for permission

That's on purpose. Claude asks for access to the folder and before anything it can't undo. Read each request and approve the ones that match its plan.

Lab 6 of 6, about 15 minutes. Best in the desktop app, with a browser-only path below.

Give it a workbench.

You'll have: a working piece of software you described in plain English, open in your browser. See a finished example.

Pick one project

FolderWhat you'll build
build-projects/personal-siteA one-page website for you, a side project, or an event. Claude interviews you first.
build-projects/dashboardAn interactive dashboard from a spreadsheet. Includes made-up sales data, or use your own non-confidential CSV.
build-projects/team-toolA small tool for your team: a meeting cost calculator, an onboarding checklist, a quote builder, or your own idea.

Each folder has a file called CLAUDE.md. It's the briefing for your coding hire: it tells Claude you're new, to keep things simple, to work offline, and to interview you first.

Path A: the Code tab in the desktop app

  1. In the Claude desktop app, switch to Code, with Sonnet 5.5 as the model.
  2. Choose your project folder from the starter kit as the folder to work in. The folder picker is in the row under the message box.
  3. Paste the starter prompt for your project (below). Answer Claude's questions. Say yes when it asks to create files.
  4. When it's done, double-click the file it made (index.html, dashboard.html, or tool.html) to open it in your browser. If it opens in a text editor, right-click it and choose Open With, then your browser.
  5. Ask for two changes. Refresh your browser to see each one.
Path B: Claude Code in the terminal (if you installed it)
  1. Open Terminal (Mac) or PowerShell (Windows).
  2. Type cd and a space, drag your project folder onto the window, and press Enter.
  3. Type claude and press Enter. Then paste the starter prompt.
Path C: the Code tab won't open

First, update the desktop app; older Windows versions asked for Git. If it still won't open, do the same build in Cowork: start a Cowork task, add your project folder, and paste the same starter prompt. The prompts tell Claude to read CLAUDE.md, so you get the same result.

Path D: browser only, no desktop app
  1. Open a new chat at claude.ai.
  2. Drag in your project's CLAUDE.md file. For the dashboard, drag in sample-sales.csv too.
  3. Paste the starter prompt and add: "Build it as an artifact I can see here."
  4. Ask for your two changes in the same chat.

Starter prompts

One-page website

Read CLAUDE.md, then help me build a one-page website. Interview me first.

Dashboard

Read CLAUDE.md and sample-sales.csv, then build me a dashboard. Start by telling me what's in the data.

Team tool

Read CLAUDE.md, then help me build a small tool for my team. Interview me first.
It worked if you can open it and use it. For the dashboard, Claude states the total revenue. With the sample data it should be $1,707,576.80.

Publish it (optional)

  1. Open a chat in Claude and drag in the file you built.
  2. Type: "Show this as an artifact."
  3. Click Share on the artifact and turn on the public link. Anyone with the link can open it.

Don't publish anything with customer or company data. The sample data is fine to share. Prefer a website host? Netlify Drop takes a dragged folder, but the link stays password-protected until you claim it with a free account.

Bonus: "Look at what you built and suggest three improvements a real user would want. Then make the best one."

Your next 30 days

One habit carries all of it: if you do it twice, make it a skill.

Week 1Read your morning brief every day. When it gets something wrong, fix the skill, not the output. Anything blocked today? Send the IT email.
Week 2Write two skills for work you repeat.
Week 3Build one small tool your team actually uses.
Week 4Teach someone what you did today.

If you lead a team

  1. Pick three people, each with one task they repeat, like quote comparisons, weekly reports, or meeting prep.
  2. Ask IT about a company Claude plan, so work data lives under your company's terms instead of personal accounts.
  3. Give those three people one hour and this workbook. Every file in the starter kit is free to share.
  4. After 30 days, count the hours they got back. That number is your business case.

Where to keep learning

Claude 101Anthropic's beginner course. 13 lessons, about two and a half hours.
AI FluencyAnthropic's course on working well with AI: delegation, description, discernment, and diligence.
Prompt design guideAnthropic's short guide. Its core advice matches today: treat Claude like a new contractor with no context.
Guide to building skillsAnthropic's PDF on writing skills that work.
Anthropic's skills on GitHubPublic skills you can read, copy, and change.
Claude help centerStep-by-step guides for every feature in today's labs.
One Useful ThingEthan Mollick's newsletter. Plain-English writing on AI at work, from a co-author of the BCG study.
s33kThe open-source analytics tool from today's story, if you want to see what a non-engineer built with AI.

Common snags

"You've reached your usage limit"

Pro has a limit that resets every five hours, plus a weekly limit. Check both in Settings, Usage. Sonnet 5.5 stretches it further than bigger models. If you hit it today, pair up with a neighbor for the rest of the labs.

My screen doesn't match the steps

Anthropic is moving accounts to a combined Claude a few at a time, and the desktop app updated on October 15. If you don't see a Cowork option, start the task in the main box. If a menu name is different, look for the closest match in Settings or Customize.

My work laptop won't let me install anything

Labs 1 to 4 work at claude.ai in a browser. Lab 5 has a chat version, and Lab 6 has Path D. Send the IT email afterward.

I can't find the starter kit folder

Check your Downloads folder first. On Windows, unzipping can nest the folder twice (first-ai-hire-main inside first-ai-hire-main), and your Desktop may live inside OneDrive.

Words you'll hear

Large language model (LLM)
An AI trained on a huge amount of writing to predict what comes next. Claude is one.
Context
Everything the AI knows going into a task: your briefing, your files, your instructions. Better context, better work.
Project
A folder of chats in Claude that share the same instructions and files.
Connector
A link that lets Claude read (and, with your approval, act in) another app, like Gmail.
MCP
Model Context Protocol. The shared standard connectors are built on. Think USB-C for AI.
Skill
A text file of instructions Claude pulls out when a task matches it. A recipe card.
Scheduled task
Work Claude does on a clock, without being asked each time.
Cowork
Claude doing multi-step work on files and apps on your computer, and handing back finished files.
Claude Code
Claude building and running software for you, from a plain-English description.
CLAUDE.md
A briefing file in a project folder that Claude Code reads automatically.
Artifact
A page or app Claude builds right in the chat, which you can share with a link.
Open source
Software or files shared in public so anyone can use, read, and change them.
GitHub
The website where most open-source work is shared. You can download from it without an account.