Home › Blog › Give Claude a memory of your work

How to give Claude a memory of your work

Akash Gupta Founder, AI Gurukul Last reviewed 24 October 2026 8 minute read

If you use AI for work, you know this feeling. You type the same paragraph again and again. Who you are. What your company does. How you like things written. Every new chat, from zero.

An AI that knows you beats a smart AI that does not. A clever new hire who knows nothing about your work is less useful than an average one who has read your notes. This article shows how to give Claude those notes, once, so every chat starts already knowing your work.

Why does Claude forget what I told it?

Each new chat starts fresh. Claude only knows what is in that chat, plus anything you have set up to carry over. If you never set anything up, you start from zero every time.

There are three ways to carry your work over. A Project, which is what this article is about. Claude's own memory feature, which Anthropic lists on every plan as "memory across conversations". And, if you use Claude Code, an instruction file in your folder. For building an AI Employee, a Project is the one I start with, because you can see and edit exactly what it knows.

What is a Claude Project?

A Project is a space in Claude for one kind of work. It has two parts that matter:

• Instructions. How Claude should behave and do the job. Anthropic says Claude uses these "for all the chats within the project."
• Project knowledge. Files you upload, like documents or notes. Anthropic says anything you upload there "will be used across all of your chats within that project."

One thing that catches people: Claude does not see the Project's name or description. Anything it needs to know goes into the instructions or the files.

How do you make a Project, step by step?

1. Open Claude and click Projects in the left menu, or go to https://claude.ai/projects.
2. Click "New Project". Give it a name you will recognise, like "Morning Brief".
3. Open the instructions box and paste the template below, filled in with your details.
4. Add two or three files Claude needs every time.
5. Start a chat inside the Project and give it the job. Check whether it now does the job your way.

Menu names change often. If a button is not where I said, ask Claude: "Where do I make a new Project?"

What should you write in the instructions?

Write it the way you would brief a new person on their first day. Short and plain. Here is the shape I use:

• Who I am. "I run a small accounting firm in Pune with four staff."
• The job. "Every morning, read my email and list what needs me today."
• What good looks like. "No more than seven items. Most urgent first. One line each, with who it is from."
• What matters to me. "Anything from these three clients is always urgent. Anything about tax deadlines is urgent."
• How to write. "Plain English. Short sentences. No greetings."
• Never do. "Never send or delete anything. Never guess an amount. If unsure, say so."

The fastest way to write this: ask Claude to interview you. "I want you to write project instructions for a helper that does [job]. Ask me questions one at a time until you have enough, then write the instructions." Then read what it wrote and fix anything that is wrong.

Which files should you give it?

Only the files it needs for this job. More is not better. A short, clear file beats a long one it has to dig through.

• A list of your important clients or contacts, and why they matter.
• Your prices, or your standard answers to common questions.
• One or two good examples of the result you want.
• A "lessons" note where you write down what it got wrong last time.

Do not upload passwords, bank details or private details about other people unless you are sure it is allowed and safe. If you work in a clinic or with client data, check your own rules first.

How do you make the memory better over time?

Every time Claude gets something wrong, ask why. Most of the time the answer is that the instructions did not say. So add one line. "Emails from the bank about statements are not urgent." Next time it gets it right.

After two weeks of this, your instructions hold everything a new person would learn in their first month. That is the real value. The AI is the same for everyone. Your instructions are yours.

What if you use Claude Code instead?

Claude Code reads an instruction file kept in your work folder, called CLAUDE.md. It does the same job as Project instructions: who you are, the job, your rules. You can ask Claude Code to write it for you from an interview. This is Day 2, give it a brain, in the 7 day challenge, where the file is called the "about me" instruction file. More in Claude Code for people who do not code.

What are the limits of Claude's memory?

• It only knows what you wrote down. If the rule is in your head, it does not know it.
• Old information stays until you change it. If your prices change, update the file, or it will use the old ones.
• Long, messy files confuse it. Keep each file about one thing.
• It can still make mistakes. Memory makes it better at your work. It does not make it perfect. Keep checking.

What comes after memory?

Memory tells Claude about your work. The next step is to teach it the job itself, so one line starts the whole thing. That is a skill: read your first Claude skill in 20 minutes. The whole path is in how to build an AI Employee in 7 days with Claude.

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