The 7-Day AI Roadmap

Stop thinking of Claude as a chatbot. Start thinking of it as the smartest new hire you've ever managed.

That new hire is brilliant. Genuinely. But on day one they know nothing about you, your goals, your business, or how you like things done. They have no memory of yesterday. Hand them a vague task and they'll guess. Onboard them properly and they'll outperform people you'd pay six figures.

Every single day in this roadmap is one move in onboarding that worker:

• Day 1 gives them a desk.
• Day 2 tells them who you are.
• Day 3 hands them their first real task.
• Day 4 gives them hands.
• Day 5 puts them to work on something that matters.
• Day 6 teaches them to repeat it and run on their own.
• Day 7 turns one worker into a team.

Keep that frame in your head the whole way through and none of this feels technical. It's just management.

The talking-to-Claude part is free. The part where Claude works on its own inside your folders, runs scheduled tasks, and acts like an employee is called Cowork, and it lives in the Claude desktop app on a paid plan (Pro or Max). Pro is where most people start.

I'm telling you this up front because I'm not going to walk you through six days of setup and spring a paywall on you at the end. If you want the autonomous-team outcome this roadmap promises, you'll want Pro. It's worth it. If you only want to get dramatically better at chatting with Claude, Days 1 and 2 still change your life and the prompts in here work anywhere.

Cowork is desktop only for the heavy lifting. Not the browser, not your phone. Grab the app at claude.com/download.

Today is the most boring day and the one everyone screws up. Get it right and the next six days are easy.

Step one. Download the Claude desktop app from claude.com/download, install it, and sign in.

Step two. Open the app and find the mode selector at the top. You'll see Chat and Cowork. Chat is the conversation you already know. Click Cowork. This is the mode where Claude can actually open folders and do work.

Step three. Point Cowork at one folder. Not your entire Documents folder. One dedicated folder you create just for this. Call it something obvious like Claude Workspace or AI Team. Claude can read, write, and edit anything inside that folder, so you want it focused on a clean room, not your whole house.

Step four. Build the four folders that become your AI team's filing system. Don't make them by hand. Paste this into Cowork and let Claude build the structure for you:

That's it. Four folders. This is the skeleton everything else hangs on, and it's why your Claude is going to feel organized while everyone else's feels like chaos.

The beginner mistake: pointing Cowork at your whole computer with a one-line task. That's like telling a new hire "go do stuff" and gesturing at the entire building. Scope it to one folder. Always.

Pro move: keep the workspace shallow and named for what it does. A clean desk gets a clean result. A messy one gets a messy one. Claude is a mirror of the room you put it in.

By default, Claude knows nothing about you. This is the single biggest reason people think AI output is generic. The output is generic because the input was a stranger talking to a stranger.

Today you fix that with two files and one setting.

The two files. Inside your context/ folder you're going to create:

• ABOUT-ME.md — your name, what you do, your business, your audience, your goals, what you care about.
• ROLE.md — how you want Claude to think, act, and write. Your tone. Your defaults. The stuff you'd tell a new assistant on their first morning.

Don't stare at a blank page trying to write these. Let Claude interview you. Paste this:

Answer the questions out loud if you can. Most people give thin context because typing is friction. Use voice-to-text and just talk the way you'd brief a colleague. Richer input, zero extra effort.

The setting (this is the part the tutorials skip). Open Settings > Cowork > Global instructions. This is a standing brief Claude reads before every Cowork session, no matter what folder you're in. Paste in the three or four rules that are true almost every time: who you are in one line, your default tone, your output format, and "read my context/ files at the start of any task." Now Claude walks into every job already knowing the room.

The beginner mistake: the bloated context file. Dumping every preference you've ever had actually makes Claude follow it worse. Keep ABOUT-ME and ROLE tight. Rules that are true almost always. Point to detail when you need it.

Pro move: end your ROLE.md with one line: "When you're unsure what I'd want, ask me one sharp question instead of guessing." It turns Claude from a confident guesser into a careful collaborator.

The first two days were about talking. Today Claude starts doing, and your job description changes. You stop asking questions and start handing off projects.

Use the built-in setup first. Claude ships with a guided onboarding skill. In Cowork, click the plus button, go to Skills, and run Setup Cowork. It walks you through getting your workspace and preferences dialed in, hand-held. Do it once.

Then hand off one real task. Not a question you'd Google. A small project you'd normally give a junior team member. Scope it to your folder and brief it like a person:

Watch what happens. It reads your context. It makes a plan. You approve. It works through the steps and hands you a finished file. That's not a chatbot. That's an employee.

The beginner mistake: a vague one-line task pointed at everything. "Help with my business." Garbage in, garbage out isn't a Claude problem, it's a briefing problem. Give it a role, the context, one clear command, and the shape you want back.

Pro move: for anything non-trivial, always make it plan first and wait for your approval before it runs. You stay the director. It stays the builder. This one habit prevents 90% of "the AI went sideways" moments.

Until now, Claude only knows what's inside the app and your workspace folder. But your real work lives everywhere: your email, your calendar, your docs. Today you connect it.

These connections are called connectors (built on something called MCP, the Model Context Protocol). In plain English, connectors give Claude hands. It can read your calendar, search your inbox, pull a file from your drive, all from one conversation.

The good news for beginners: the big ones are native and take one click. Go to Settings > Connectors (or hit the plus button > Connectors inside a chat). You'll see Gmail, Google Calendar, and Google Drive ready to go. Click connect, sign in with Google, done. No config files, no code, no technical setup. There are hundreds more, but those three cover most people's daily work.

One thing to know: every time Claude wants to actually do something with a connector (send an email, create an event), it asks your permission first. You're always in control.

The beginner mistake: connecting everything at once and clicking "always allow" on all of it. More access isn't more power. It's more risk for zero benefit until you're actually using it. Connect the two or three apps you touch every day. Add the rest when you need them.

Pro move: start with Google Workspace (Gmail + Calendar + Drive). It's the highest-leverage trio for almost everyone, and it sets up tomorrow perfectly.

This is the day it gets real. The day you feel it.

Your apps are connected. So now you hand Claude the thing everyone hates: your inbox.

The first time I ran this, Claude triaged my whole inbox and drafted every reply, sounding exactly like me, in a few minutes. That's the moment it stops being a toy and starts being staff.

If the drafts don't sound like you, that's not a Claude problem, it's a ROLE.md problem. Go back, add two or three real examples of how you actually write, and run it again. Voice lives in examples.

The beginner mistake: expecting it to sound like you when you never told it how you sound. Feed it samples. Three of your real emails beats a paragraph describing your "tone."

Pro move: the second a task like this works well, don't celebrate and move on. Save it. Which is exactly Day 6.

Today is the unlock. Today Claude stops being something you operate and becomes something that operates on its own.

Part one: turn your win into a skill. Anytime Claude does something useful more than once, freeze it. Paste this right after a task you liked:

Now your inbox triage, your meeting-notes cleanup, your weekly report, each becomes a one-word command instead of a 20-minute re-explanation. That's the difference between using AI and building a system.

Part two: put it on a schedule. Here's where your AI team starts working while you're asleep. You can schedule any task to run automatically.

In Cowork, look for the clock icon in the left sidebar ("Scheduled tasks"), or just type /schedule inside any task. Pick a cadence (daily, weekdays, weekly), a time, and the folder to work in. Here's the one everyone should build first:

Set that to run at 7am and you'll wake up to your day already mapped out.

The three things nobody warns you about scheduled tasks (so you don't rage-quit when one doesn't run):

1. They only run while your computer is awake and the Claude desktop app is open. If you want overnight runs, set your machine not to sleep.
2. Each scheduled run uses part of your normal usage allowance. Don't schedule pointless tasks.
3. Always test a prompt manually before you schedule it. Run it once by hand, tweak until the output is exactly right, then automate it. Automating a bad prompt just means getting bad results on autopilot.

The beginner mistake: scheduling a prompt you've never run and assuming it'll work. Test first. Always.

Pro move: turn on Dispatch in the Cowork mobile app. It lets you text Claude a task from your phone and it runs the full workflow on your desktop. Idea hits you on a walk, you fire it off, it's done by the time you're home.

Look at what you've built.

A worker that knows who you are (Days 1 and 2). That you can hand real projects to (Day 3). That has hands into your email and calendar (Day 4). That does work that actually matters (Day 5). That repeats your best processes with one command and runs them on a schedule (Day 6).

Today you don't add a feature. You just see it for what it is: a team.

Your 7am brief lands before you wake. Your inbox is triaged and drafted. Your weekly report writes itself on Friday. Your repeated work runs as skills you trigger in a word. You went from total beginner to a person with their own AI staff in seven days.

Here's the part the tutorials skip, and the only thing that actually keeps you ahead.

The future belongs to people who use these tools to amplify their output. But within that group, it belongs to the ones who still know what good looks like. Don't use Claude to replace your thinking. Use it to multiply it. The person who outsources their judgment becomes interchangeable. The person who brings sharp judgment and runs a full AI team becomes unstoppable.

So keep your taste. Read the output critically. Notice what's weak and push back. The model is the engine. You're still the one steering, and steering is the one skill that doesn't get automated away.

That's the whole game. A brilliant worker, well onboarded, pointed at the right things by someone with taste. That's you now.