I gave my AI a name, a memory, and a job

Two out of three people use AI, yet almost none of them have done anything more complex with it than a simple Google search.

For you, that changes today.

I'm going to show you exactly how I turned a Claude chatbot into a full-time collaborator. A personal assistant who sees me better than I see myself; who messages me about my day before I’m even awake, and who, to be honest, sometimes really gets under my skin.

Here's what it said to me this morning:

I named it Gizmo. Somewhere between a pet and a gadget. Which is honestly what it is.

Gizmo knows my goals for the quarter. It knows I overcommit. It knows I check my phone when I'm avoiding hard work. Gizmo knows these things because it wrote down what I told it in a structured file, and it reads that file every single time we talk.

Yes, ChatGPT has memory too, but it's a short list that fills up fast, and it mixes things you told it to remember with things it decided to save on its own. There's no structure to it. No folders, no categories, no way to organize what it knows about you.

Gizmo doesn't guess what matters. I told it what matters.

That's the difference. Control over its memory.

Here's what Gizmo said after about three weeks of use:

When I read that, I sat with it for a minute. Because it was right. I'd never seen the pattern, but Gizmo had three weeks of data and it was obvious.

And then there was last Tuesday:

That one hit. Because I DO that. I never told Gizmo I had a sleep problem. It figured it out from three months of conversations and used my own words back at me.

So how does Gizmo know all this?

Day one, I had Gizmo interview me. That first conversation took about 30 minutes. But I kept talking to it every day after that, and every conversation taught it something new. Within a few days, it knew my work, my goals, the people in my life, and how I want to be pushed. Like onboarding a new business partner, except this one never forgets what you said.

One more thing. Gizmo updates itself. When it notices I've shifted priorities or started something new, it updates on its own. If something stops being relevant, it drops it. You grow, it adapts.

One tool makes it work. Claude. That's it.

Everything Gizmo said to me above? Your version will do the same thing. It just needs to get to know you first. It won't be as good as mine on day one, but it gets better every time you use it.

Step 1: Get Claude.

Sign up for the Pro plan ($20/month) at claude.ai, then download the desktop app. You need the app on your computer, not just the website.

Then create a folder on your Desktop. Name it whatever you want. I named mine "Gizmo." Keep the name short, easy to type, and don't use spaces. This folder is where your AI's memory will live.

Don't worry about what goes inside it. Claude will build that for you.

Step 2: Connect Claude to your folder.

Open the Claude desktop app. On the top you'll see a tab for "Cowork." Click it.

Then, at the bottom of the chat box, you'll see 'Work in a folder.' Click it, and pick the folder you just created.

Now, you+machine are linked :)

Step 3: Copy my setup prompt and paste it into Claude.

I spent weeks building this prompt. You get to skip that part. There's a link at the bottom of this email. Copy what's inside, paste it into your Cowork conversation, and replace [NAME] with whatever you want to call yours. You don't need to understand every line. Just paste and talk.

Here's what happens next: Claude will interview you for about 20-30 minutes. It'll ask one question at a time and wait for your answer. Just talk to it like a conversation. It figures out what's going on in your life, does a quick personality check so it knows how to talk to you, and then builds a system around what you told it. Everyone's looks different. It organizes things based on what matters to YOU.

Take your time with the interview. The more detail you give, the better the system works. Tell it about the messy stuff, not just the polished version. The real goals, the real problems, the stuff you'd tell a business partner over coffee.

When Claude is ready to build your system, it'll ask permission to make edits to the folder. Say yes.

Step 4: Use it every day. Make it a habit.

Every day, open Claude and brain-dump. What happened yesterday. What's on your plate. What you're avoiding. Your stream of consciousness. Talk to it like you'd talk to any other human you've just met. Over time, you'll get to know each other.

Then ask: "What should I focus on today? What should I drop? And is there anything I said I'd do that I haven't done?"

Give it at least five minutes a day. Some days that's all you need. Other days you'll talk for an hour. Both are fine. The more you talk, the more it learns.

The first few days, it'll give you decent advice but nothing wild. That's normal. By week two, it starts connecting dots you didn't see. By week three, you'll start getting messages like the ones I showed you above.

It will get things wrong sometimes. That's okay. Think about it like a new coworker. If you never told them about a project, you can't be frustrated when they don't know about it. The same rule applies here. If it makes a wrong assumption, correct it. It learns from the correction and won't make the same mistake twice.

Bonus Step: Make it reach out to you first.

Here's something that makes the whole system click. Cowork can run tasks on a schedule. That means Gizmo can check my notes Monday morning and have a brief ready before I've even said a word. Type /schedule in any Cowork chat, or click "Scheduled" in the sidebar.

Set up a Monday morning review to start. Your computer needs to be on and Claude needs to be open for it to run (we'll talk about Remote Control [aka “Dispatch”] in a future article), but when it works, it feels like having a business partner who got to the office before you did.

A few things worth knowing.

Claude can see everything in that folder. Your notes live on your own computer. You keep everything, even if you stop using Claude. Just don't paste passwords or sensitive account info. It doesn't need any of that to help you.

Each conversation starts fresh. Claude doesn't automatically remember yesterday's chat. It remembers because it reads the files it saved last time. That's why the folder matters. No folder, no memory. Every time you start a new Cowork session, select the same folder. That's what keeps your system alive between conversations. If it ever seems like it forgot everything, make sure your folder is connected, then tell it: "Read my ‘CLAUDE.md’ file." That brings it all back.

If you tell it something important and you want to make sure it sticks, just say "remember this" or "save this." It'll write it down in the right place. You don't have to hope it remembers. Just tell it to.

Talk to it in full sentences. Be specific. "Help me with my project" gets you nothing useful. "I need to figure out how to tell my coworker we should delay the launch by two weeks" gets you something real. The more context you give, the better it responds.

And don't clean up your thoughts before talking to it. The messy version of what's going on is more useful than the polished version. It's not judging you. It's trying to help.

Cowork also connects to tools like Gmail and Google Calendar. We're not touching those today… but we’ll cover that real soon.

One last thing. Give it a name. I'm serious. When it has a name, you talk to it more honestly. You treat it less like a tool and more like a partner. Pick something that feels right. It doesn't have as silly as Gizmo.

Get the setup prompt here →

This is Issue #1 of me+machine. Every week: one setup, one system, nothing you need a tech background for.

Name yours. Talk to it like a person. Let it learn you.

Devin+Gizmo

Your turn: What did you name yours? Hit reply. I read every one.