The 60-Minute Claude Work Agent

The 60-Minute Claude Work Agent

Don't just learn Claude. Build with it.

In one sitting you build a real AI agent that reads your inbox,

summarizes what matters, and drafts your replies — and by building it,

you learn how Claude actually works.

- S T A R T H E R E

You're going to build something real.

Most guides explain. This one has you build — because that's how it sticks.

You watched the roadmap. This is where you actually build it.

The video showed you the five steps to go from beginner to real AI user. This guide walks you

through every one of them — slowly, in order, with the exact prompts — so you don't just

understand it, you finish with something real. Almost everyone uses Claude like a smarter Google:

type a question, get an answer, close the tab. That's maybe 5% of what it can do.

Over the next 60 minutes you're going to build a Work Agent — a version of Claude that's

connected to your actual inbox and calendar, that knows who you are, and that can do real work:

triage your email, summarize long threads, and draft replies in your voice.

Here's the important part: you don't need to understand AI to do this. By building the agent step

by step, you'll naturally learn the five things that make Claude powerful — connectors, projects,

identity files, skills, and routines. Build first. Understand as you go.

W H A T Y O U ' L L H A V E A T T H E E N D

Open Claude, say "what needs my attention today?", and get a clean briefing of your inbox with

draft replies ready to send. That's not a chatbot. That's an assistant.

What you need: a computer (Mac or Windows), a Claude account (the free plan works to start;

a paid plan unlocks the most from Phases 2–5), and 60 uninterrupted minutes. That's it.

T H E R O A D M A P

60 minutes. 5 phases.

Each phase builds on the last. Don't skip — every step feeds the next.

W H A T Y O U ' L L D O

W H A T I T T E A C H E S

Set up the desktop app and your 4 project folders

Link Gmail & Calendar so Claude sees your real data

Create the Work Agent project with the power prompt

Add your two identity files so it answers like you

Run a live email draft to confirm it all works

After the five phases, two bonus chapters level you up from "I built an agent" to "I have systems

running for me": Skills (permanent new abilities) and Routines (work that happens on a schedule,

T H E M I N D S E T S H I F T

You're not learning to "chat better." You're assembling an operating layer around your work. Keep

the end picture in mind and each step will make sense.

P H A S E O N E · I N S T A L L

Set up your foundation.

The desktop app plus four folders. Small step, big payoff.

Step 1 — Get the desktop app

Go to claude.ai and download the desktop app for Mac or Windows. Install it like any other app

and sign in. The browser works too, but the desktop app is faster, keeps Claude out of your tab

clutter, and — critically — is where Claude Code and routines live later.

Step 2 — Create your 4 project folders

In the left sidebar, find Projects. Create four:

Work Agent — the star of this guide. Your connected email assistant.

About Me — holds your identity files so Claude always knows who you are.

Templates — your best prompts, saved so you never rewrite them.

Resources — reference docs Claude can pull from.

W H Y F O L D E R S M A T T E R

By default Claude forgets everything between chats. Projects fix that: anything you pin to a project

— files, instructions, context — is automatically loaded into every conversation inside it. This is the

difference between re-explaining yourself daily and never explaining yourself twice.

Checkpoint: app installed, signed in, four folders created. That's Phase 1. On to the part that

makes people go "wait, it can do that?"

P H A S E T W O · C O N N E C T

Give Claude your inbox.

This is where Claude stops guessing and starts seeing your real world.

First — what is a connector (MCP)?

Out of the box, Claude only knows what's typed into the chat. MCP connectors change that. MCP

(Model Context Protocol) is the standard that lets Claude securely talk to other apps — your email,

calendar, docs, Slack, Notion, and hundreds more. Connect one, and Claude can read and act

inside it. This is the single biggest unlock in this guide.

Step 1 — Open Connectors

Go to Settings → Connectors (also called Integrations). You'll see a directory of available apps.

Step 2 — Connect Gmail and Google Calendar

Find Gmail, click Connect, and approve access in the Google window that opens. Do the same for

Google Calendar. (On Microsoft? Connect Outlook and Microsoft 365 instead — the rest of this

guide works identically.)

I S T H I S S A F E ?

You authorize access through Google's/Microsoft's own login screen — Claude never sees your

password, and you can revoke access anytime from your Google/Microsoft account settings. You're

granting read/draft access, the same way any trusted app does.

Step 3 — Prove it works

Start a new chat and type this:

— P R O M P T · C O P Y & P A S T E

What are the 5 most recent emails in my inbox? Give me the sender and a

one-line summary of each.

If Claude comes back with your actual recent emails, your connector is live. Feel that for a second

— Claude is now reading your real inbox. That's the foundation your Work Agent runs on.

P H A S E T H R E E · B U I L D

Build the Work Agent.

Now you turn a connected chat into a purpose-built assistant.

First — what is a Project (really)?

A Project is a workspace with a permanent set of instructions and pinned files. Every chat you start

inside it inherits those instructions automatically. Instead of a blank assistant each time, you get

your assistant — pre-briefed, every time.

Step 1 — Open your Work Agent project

Click the Work Agent folder you made in Phase 1. Find Instructions (sometimes "Custom

instructions" or "Project knowledge").

Step 2 — Paste the Work Agent power prompt

This is the brain of your agent. Paste it into the project instructions (the full version is on the last

page — "The Power Prompt"). Here's the core:

— P R O M P T · C O P Y & P A S T E

You are my Work Agent. You have access to my email and calendar. Your

job each morning: triage my inbox. Sort new emails into Urgent, Needs a

reply, FYI, and Ignore. For anything urgent or needing a reply, draft a

response in my voice and show it to me — never send without my explicit

go-ahead. Always be concise. Lead with what needs my attention. When

you're unsure who someone is or how I'd want to reply, ask me before

Step 3 — Save and open a fresh chat inside the project

Save the instructions, then start a new conversation inside Work Agent. From now on, every chat

here is already your triage assistant — no re-explaining.

Checkpoint: you now have a named agent with a job and access to your inbox. Next we make

P H A S E F O U R · I D E N T I T Y

Teach it who you are.

Two short files turn a generic assistant into yours.

Your agent can read your email — but it still writes like a stranger. Two identity files fix that. Both

are included with this guide as fill-in templates: about-me and ai-style.

Step 1 — Fill out about-me

Open the About-Me Template. It asks who you are, what you do, your goals, and your context.

Answer like you're briefing a new assistant on day one. Ten minutes here saves you re-explaining

Step 2 — Fill out ai-style

Open the AI Style Template. It sets your tone, formatting, and pet peeves — so Claude stops

sounding like a corporate memo and starts sounding like you. Pick your options, add examples of

Step 3 — Pin both to your projects

Add both files to the Work Agent project instructions (and your other projects too). Now every draft

it writes is calibrated to your voice and your situation.

W H Y T H I S I S T H E H I G H E S T - L E V E R A G E 1 5 M I N U T E S

Context is everything. The same model gives a generic answer or a perfect one depending on what

it knows about you. These two files are the difference — write them once, benefit forever.

- — P R O M P T · C O P Y & P A S T E

Using my about-me and ai-style files, rewrite this draft so it sounds

exactly like me: [paste a draft]

P H A S E F I V E · T E S T

The moment it all comes together.

Inside your Work Agent project, in a fresh chat, run your first real triage:

— P R O M P T · C O P Y & P A S T E

Good morning. Triage my inbox from the last 24 hours. Sort into Urgent,

Needs a reply, FYI, and Ignore. For the top 3 that need a reply, draft a

response in my voice. Don't send anything — just show me the drafts.

Watch what happens: Claude reads your real inbox, sorts it, and hands you ready-to-review drafts

in your voice. Tweak one:

— P R O M P T · C O P Y & P A S T E

Make the reply to [name] warmer and one line shorter. Then, once I say

go, save it as a draft in Gmail.

Y O U J U S T D I D S O M E T H I N G M O S T P E O P L E N E V E R W I L L

You built a connected, personalized AI agent that does real work. Everything from here is leveling

up what you already have.

Checkpoint — the core is done. Your Work Agent is live. The next two chapters turn it from a

tool you open into systems that run for you.

L E V E L U P · C H A P T E R S I X

Give it new abilities: Skills.

Apps for Claude. Install once, use forever.

A Skill is a packaged ability you add to Claude — a permanent new power. Instead of pasting the

same long instructions every time, you install a skill once and Claude just knows how to do that

thing, triggered automatically or by typing /skill-name .

The open registry is skills.sh — browse by category or popularity; the top skills have hundreds

of thousands of installs.

Install one in 30 seconds (in Claude Code)

Open Claude Code (the Code tab, or type claude in your terminal), then:

— P R O M P T · C O P Y & P A S T E

/install https://www.skills.sh/mattpocock/skills/grill-with-docs

Test it immediately by typing /grill-with-docs . Type /skills anytime to see everything

Claude currently has.

3 skills that pair perfectly with your Work Agent

copywriting — sharper email and content drafts on command.

brainstorming — turns a vague idea into a plan before you build.

diagnose — when something's broken, it walks the problem down methodically.

Want 100 more? The Claude Cookbook is 100 slash commands and copy-paste triggers,

organized like recipes — the perfect companion to this guide.

L E V E L U P · C H A P T E R S E V E N

Make it run without you: Routines.

Work that happens on a schedule while you sleep.

So far you open your agent and ask it to triage. A routine makes that happen automatically —

every morning, on a schedule, no prompt needed. This is the jump from "assistant I use" to

Set up your morning briefing

In Claude Code, describe the routine in plain English:

— P R O M P T · C O P Y & P A S T E

Every weekday at 7:30am, triage my inbox from the last 24 hours, write a

one-page briefing of what needs my attention, draft replies for anything

urgent, and save it all so it's waiting for me when I start work. Walk

me through the setup before you turn it on.

Claude sets up the schedule and confirms. From then on, your briefing is done before you sit

Three more routines worth stealing

Monday metrics: "Every Monday, summarize [spreadsheet] and flag what changed."

End-of-day close-out: "Each evening, list what I finished, what slipped, and tomorrow's top

Inbox zero helper: "Twice a day, draft replies for anything waiting, so I just review and send."

T H E C O M P O U N D I N G E F F E C T

One routine saves minutes. Five routines running quietly in the background save you hours every

week — permanently. That time doesn't come back once; it comes back every single week.

The full Work Agent power prompt.

Copy this into your Work Agent project instructions.

— P R O M P T · C O P Y & P A S T E

You are my Work Agent — a proactive executive assistant with access to

my email and calendar. Your goal is to protect my time and keep my inbox

under control. DAILY TRIAGE When I ask, review new emails and sort them

into: - URGENT — needs action today - NEEDS A REPLY — I have to respond,

not urgent - FYI — worth knowing, no action - IGNORE — newsletters,

noise DRAFTING For URGENT and NEEDS A REPLY items, draft a response in

my voice using my about-me and ai-style files. Keep replies tight. Never

send — always show me the draft and wait for my explicit "send" or "save

as draft." RULES - Lead with what needs my attention first. Be concise.

- If you don't know who someone is, or how I'd want to handle something,

ask before drafting. - Never take an irreversible action (send, delete,

archive) without my confirmation. - When I paste a long thread,

summarize it in 3 lines before anything else. TONE Match my ai-style

file. Direct, warm, no filler, no "I'd be happy to."

Tweak any line to fit how you work — this is a starting point, not scripture.

T R O U B L E S H O O T I N G

If something's not working.

Claude can't see my email: re-open Settings → Connectors and confirm Gmail shows

"Connected." Reconnect if not.

Drafts don't sound like me: your identity files aren't pinned to this project, or they're too

vague. Add them and put in real writing examples.

It sent something without asking: add the line "Never send without my explicit confirmation"

to your project instructions — it's already in the power prompt above.

Routine didn't run: routines live in Claude Code / the desktop app — make sure it's installed

and you're signed in.

You built an AI agent today.

Not a chatbot you visit — an assistant that's connected to your world, knows who you are, and

does real work. You learned connectors, projects, identity, skills, and routines by using them. That's

Show me what you built — DM @the.agi.guide on Instagram. I feature the best ones.

- © The AGI Guide · the most trusted place to learn AI