Home › Blog › Build an AI Employee in 7 days

How to build an AI Employee in 7 days with Claude

Akash Gupta Founder, AI Gurukul Last reviewed 24 October 2026 10 minute read

Most people who try to build an AI helper quit early. Not because it is hard. Because they try to build everything at once, get stuck, and never see it work.

So I teach it the other way round. You build one assistant, and you add one part a day to it. You see it work on the first day. By the seventh day it runs every morning without you, and you have watched every part of it get built. This is that plan, day by day. I call it: learn how to build your first AI Employee, your AI Personal Assistant.

What will you have at the end of 7 days?

One AI Employee: your AI Personal Assistant. Every morning it reads your Gmail and Google Calendar, tells you what matters today, drafts replies for you to check, and keeps your to-do list. It only reads your email. Any reply it drafts is for you to check and send yourself.

You will not have an AI company. You will have one assistant, and you will know how to build the second AI Employee. If you want the idea first, read what is an AI Employee?.

Get Ready: what do you set up before Day 1?

• A laptop. Mac or Windows. The AI runs on Anthropic's servers, so it does not need to be powerful.
• Claude Pro. Claude Code is in the paid Claude plans, not the free one. Claude Pro was USD 20 a month billed monthly, or USD 17 a month paid yearly, on claude.com/pricing on 24 October 2026. Indian users are billed in rupees, so check the price in your account.
• VS Code. A free app from Microsoft for working with files. You will not write code in it. It is simply the window where you talk to Claude Code and see the files it makes.
• Claude Code, added to VS Code and signed in with your Claude account.

If this part feels new, read Claude Code for people who do not code. Once it is set up, you check it with one simple request to Claude, and you are ready.

Day 1: meet your assistant

Today you see it work. Claude interviews you about your work and your day, one question at a time. Then it plans your day once, from what you told it.

You can try this now. Open Claude Code in an empty folder and type: "Interview me about my work, one question at a time. Ask about my day, the people who matter and what I keep forgetting. Then plan my day today." Note how long your own planning usually takes. That number is your reason to keep going.

Day 2: give it a brain

A new conversation forgets everything. So on Day 2 your assistant gets an "about me" instruction file: who you are, your work, the people who matter, and your rules. Claude Code reads it every time you start, so you never explain yourself again.

You do not write it from a blank page. Claude interviews you and writes the file, and you fix anything wrong. How to write a good one, including the same idea in the Claude app with Projects, is in how to give Claude a memory of your work.

Day 3: teach it a job

Now it learns its first real job: the morning brief. You save the job as a skill. A skill is a saved set of steps that Claude follows whenever the job comes up. You teach it once, and from then on one line starts the whole job.

Claude helps you write it from a conversation, so again there is no blank page. Step by step in your first Claude skill in 20 minutes.

Day 4: give it hands

Today it can see your real day. You connect Gmail and Google Calendar to your Claude account: in Claude, go to Customize, then Connectors, press Connect, sign in with Google and approve access.

Reading only, always. Anthropic's Gmail documentation, checked on 24 October 2026, says the connector can search and read email, and that Claude "can't create, send, or modify emails." So the worst it can do on Day 4 is give you a wrong list. The safe way to start with email is covered in how to build an AI email assistant for Gmail that only reads.

Day 5: rules and testing

Two things. First, a never do list: the rules it must never break, written into its instructions. Never send. Never delete. Never share private details. Never guess a number.

Second, the 10 run test. Run the morning brief 10 times and score each result: right or wrong. If 8 or more out of 10 are right, it is ready for daily use with you still checking. If not, fix the instructions and test again. Full method in how to test an AI before you trust it.

Day 6: put it on a timer

This is the day it stops being a chat and becomes an employee. The morning brief is set as a scheduled task, so it arrives every morning on its own. Anthropic's help page says scheduled tasks are on all paid plans.

The question everyone asks is about the laptop. Anthropic says scheduled tasks run "even when your computer is asleep or the Claude Desktop app is closed", but a task that needs files on your laptop runs only on your laptop. Details, checked on 24 October 2026, in how to make Claude work every morning without you.

Day 7: birth day

Your AI Personal Assistant gets its birth certificate: a card with its name, its job and your name, that you can share. You add your name to the Graduates wall with the others who finished. And you write down the job for your second AI Employee, because the second one takes a fraction of the time.

That is also the moment to look back at your Day 1 number. How long did planning your day take before? How long does reading the brief take now?

How much time does it take each day?

Plan for 20 to 30 minutes a day. Get Ready and Day 4 can take longer, because setting up and connecting apps are where most people hit small problems. Missing a day is fine. The parts build on each other, so do them in order, not fast.

Can you do it in the Claude app instead of Claude Code?

Mostly, yes. The Claude app has the same parts under different names: a Project instead of the instruction file, Skills, Connectors and scheduled tasks. The memory, skill, testing and timer articles linked above explain the app way too. I teach the challenge in Claude Code because the files sit in a folder you can see, and the same setup grows into a small AI team later. For the other route, read how to build an AI agent with Claude, without coding.

Where do people get stuck?

• The job is too big. "Run my inbox" is too big. "Tell me the five things that matter today" is right.
• Skipping the test. One good run proves nothing. Run it 10 times.
• Giving too much access too early. Read first. Any reply goes out only after a person checks it and presses send.
• Menus that moved. Claude changes often. If a button is not where I said, ask Claude itself: "Where do I find Connectors?"

Who should not do this yet?

• If nothing in your day repeats, there is nothing steady to hand over yet.
• If you will not check its work, do not build it.
• If the job is medical, legal or financial advice to someone, keep a qualified person in charge.
• If you cannot pay for a plan right now, practise writing clear instructions in the free Claude chat. Anthropic says Projects are on the free plan, up to five, so you can practise the Day 2 idea there.
• Claude accounts are for adults 18+. Under 18? Learn it together: the parent builds the assistant, and the child learns alongside.

Can you do this with others?

You can do every step above alone, using this article. If you want step by step lessons for each day, the starter files, and a place to ask when you are stuck, that is what the free 7 day challenge in AI Gurukul is. What the community includes is in Join AI Gurukul: building with AI, made easy.

Akash Gupta, Founder, AI Gurukul

Akash studied at IIM Ahmedabad and built Peepoye, a Hindi YouTube channel. He does not write code. He runs his own company, Peepoye Network, with AI Employees he built himself with Claude, and teaches the same method in AI Gurukul. His story.

28.2 croreviews on his own channel

12.8 lakhsubscribers

1.4 crorehours watched