|
Happy Wednesday,
Today I'm showing you how to set up your own OS folder so AI has useful background before it starts working, then I've got a free guide from HubSpot with over 100 ChatGPT prompts for work, then what Claude's new Opus 5.5 is actually good at, and three more image codes that turn an ordinary photo into something completely different.
Let's get into it.
|
IN THIS EMAIL
» Set up your own OS folder
» 100+ ChatGPT prompts for work, with HubSpot
» Claude just made its comeback
» 3 more image codes worth saving
|
📁 DO THIS TODAY
This One Setup Will Put You Ahead of Most People Using AI
Everyone always asks me how to make their AI better. The trick is setting up a file system on your computer. I think it's the single thing you can do to get ahead of most people using AI, and I've realized I've never properly walked you through it in this newsletter. So I want to get everyone set up.
You know when ChatGPT or Claude forgets something you've already told it? We expect the built-in memory to keep all the important bits, but it picks what to remember, and that doesn't always include the thing you need. So you're back to explaining who you are, how you work, and what you meant.
This gives those details somewhere better to live. You build a second brain stored in files on your computer, with notes the AI can read and update. Instead of relying on whatever it remembered, you can point it to your actual preferences, project notes, and previous work.
If 'file system' already sounds confusing, don't worry. I've put together a full guide and done-for-you setup prompt that walks you through it and builds the files for you. Here's what you're setting up and why it matters.
This setup is for the desktop apps on your computer. If you only use ChatGPT or Claude on your phone, skip this one for now. But if you're ready to take AI a step further, this might be the single most important resource I've put together. You only need one of the two apps to start.
When I tell you to try ChatGPT Work or Claude Cowork, this is a big part of why. These tools can work inside folders on your computer. You give the AI access to one folder, and it can read the files, create new ones, and save your finished work back there. Keep access limited to that folder unless you choose to approve more.
Mine is called Maverick OS. Yours can be your name followed by OS. It's just a folder with notes about you, smaller folders for the things you're working on, and somewhere to keep the results. Here's an example of what your files will look like:
|
📁 Your Name OS
📄 START-HERE.md Where the AI should begin
📄 AGENTS.md How the AI should work
📄 CURRENT.md What you're working on now
📄 memory.md Useful changes across your whole system
📁 resources
📄 about-me.md 📄 writing-style.md
📁 workspaces
📁 work
📁 travel
📄 AGENTS.md 📄 CURRENT.md 📄 memory.md 📁 resources 📁 outputs
📁 personal-admin
Each workspace has the same supporting files. Only travel is expanded here. The guide calls workspace instructions guide.md, serving the same role as AGENTS.md.
|
Those .md files are just files of text. Think of a PDF as a printed page and an .md file as the editable note behind it. AI can read PDFs, but scans and complicated layouts can add extra work. Plain text is simpler to read and update.
Say you tell ChatGPT, 'I'm going to Japan in November. I prefer direct flights, and check with me before booking.' Your travel files might look like this:
AGENTS.md · Instructions Prioritize direct flights. Ask before booking anything.
memory.md · What to remember Sept 23: Planning Japan for November. Nothing booked yet.
Think of AGENTS.md as your assistant's job description and memory.md as their notebook. The setup prompt creates these files and tells it to save useful updates after every task. Next time, it reads the relevant notes first. You can open and correct that memory yourself.
Paste the setup prompt, then answer its questions in voice notes. You can also supply your old chats using the export guide, or start without them.
In ChatGPT: open the desktop app, choose Work, then use the project or folder chooser to connect your OS folder. Return to that same local project for future work. You can pin the setup chat in Recents to find it again.
In Claude: open the desktop app and choose Project or folder (shown below), or Work in a folder in the newer interface, then select your existing OS folder. If your app still has a Cowork tab, start there. Star the project from its three-dot menu. I confirmed this on mine: the starred project loads automatically when I open Claude, so my OS folder is ready to go.
If yours opens somewhere else, select your starred project in the sidebar. Either way, check that your OS folder is connected before starting.
Set aside a couple of hours to do this properly. A big history import can take longer, and you can pause and resume. I think it's worth that time to stop spending the next six months repeating yourself.
Pro tip: for the setup, choose Opus 5.5 in Claude or GPT-6 Astra in ChatGPT Work. Set thinking to Extra High, or the highest thinking option your app offers. Give it the strongest model and thinking setting available on your plan for this first pass.
Get the full setup guide and done-for-you prompt →
SPONSORED
100+ ChatGPT Prompts You Can Use at Work
If you open ChatGPT for work and end up asking it the same three things every day, this one's for you. HubSpot put together a free guide with over 100 ready-to-use prompts, covering writing emails, creating content, customer support, data analysis and more.
It also has 15+ industry-specific use cases and 21 best practices to help you get better answers, plus advice on the common problems people run into. Pick the part of your week that takes longer than it should, find a prompt for it, and start there.
Get the free guide →
🚨 AI UPDATE
Claude Just Made Its Comeback
A couple of weeks ago I told you Astra was the one to beat and that I'd be surprised if Anthropic didn't come back with something. Well, here it is. Claude Opus 5.5 launched yesterday, and the part I'm interested in isn't what it scored on an exam. It's whether you actually enjoy working with it again.
The writing is the part I'd pay attention to first. Anthropic says it gets to the point faster, uses less jargon and follows your writing rules better. A smarter answer isn't much use if you spend the next ten minutes making it sound like a person. Getting a useful draft without three rounds of 'please stop writing like that' is the kind of improvement I care about.
It also looks better suited to the jobs that take more than one answer. Lovable's early testing found it could finish app builds with fewer rounds of edits, and Anthropic says it has improved at digging through sources for research. So if you use Claude to write something in your voice, build an app or work through a pile of information, those are the things I'd try first.
It's now the default on Pro, Max and Team, including Cowork. They're also increasing usage limits and giving subscribers a reset they can save for when they need it, which is much more useful than another chart telling you how clever the model is.
I like that these companies keep forcing each other to get better. What I wouldn't do is rebuild my entire setup every time one of them wins a round. Put your work, your preferences and your instructions in that AI brain from the first section, then use whichever one does the job best. Let them fight over the model. You keep the memory.
Read the full Opus 5.5 announcement →
💡 SAVE THIS
3 More Image Codes Worth Saving
A few weeks ago I gave you the billboard, time-machine, and magazine-cover codes. Here are three more that give you a completely different way to use a photo. Same idea as last time: attach an image in ChatGPT, paste the code and the line under it, and let it make the new version.
The line underneath is the actual instruction, so copy that part too. Start with a clear photo where you can see the subject properly, especially if you're using something you sell.
/360-view
/360-view
Create one image showing this product from six different angles, like frames from a 360-degree turntable. Keep the product consistent on a clean white background. |
Upload a product, a pair of shoes, or something sitting on your desk, and ask for the views you'd normally have to walk around it to photograph. This gives you a still contact sheet, so you can compare the angles side by side. It has to imagine the parts your photo doesn't show, so check those details before using it in a real listing.
/impossible-angle
/impossible-angle
Recreate this photo from an impossible camera angle, as though a tiny camera could fit inside the scene. Keep the subject recognizable and make it look like a real photograph. |
This puts the camera somewhere you could never normally get it. Try a sneaker shot from underneath the laces, or a coffee cup photographed from the inside looking out. The subject stays familiar, but the perspective is completely different. If the first one isn't strange enough, tell it exactly where the tiny camera should be.
/meta-ads
/meta-ads
Turn this product photo into a polished square social ad. Keep the product recognizable, add a short headline and a clear call to action, and do not invent prices, discounts, reviews or product claims. |
This is the useful one if you sell anything. Give it a candle, a skincare bottle, or your own product photo and have it build the ad around it. Add what the product actually does and who it's for, then ask for a few different headlines. You still choose and check the final creative before putting it into an ad campaign.
I put all 200 image and video shortcuts in the full image-codes resource here. Save it for the next time you have a photo and no idea what to do with it.
Set aside a proper block of time for the AI brain this week. And if you've only got a minute tonight, try the impossible-angle code on something sitting in front of you.
See you Sunday.
Maverick
P.S. Want your business to actually show up inside AI? The brands that rank are seeing up to 3x the conversions. Get your free AI visibility score and see exactly where you stand inside AI.
P.P.S. If you're new here or you ever skip one, I've got a newsletter archive of all my past emails you can catch up on anytime.
|