Short on time? Get the key takeaways in ChatGPT →
Over the last few months, I built myself an AI business assistant that runs inside Claude Code or ChatGPT’s Codex.
It remembers everything about my business, so I never have to re-explain myself again, and it’s taken so much repetitive busy work off my plate.
In this post, I’m walking you through how mine is set up and what it does for my business every week. I kept it simple, so even if you’ve barely touched ChatGPT, it’ll still make sense.
Would you like to save this?
*This post contains affiliate links, clearly marked with an asterisk (*). If you click on these links and make a purchase or sign up, I may earn a commission or credit, at no extra cost to you.
So what is an AI business assistant, anyway?
An AI business assistant is just a set of folders on your computer that an AI tool reads before it does any work for you. The folders hold plain text files about your business: who you serve, what you sell, how you write, and what you’re working on.
I know that sounds almost too simple… but that’s really all it is.
Mine runs inside Claude Code* (made by Anthropic) and/or Codex (ChatGPT’s version of the same kind of tool, made by OpenAI).
I simply open either one, point it to work inside my AI business assistant folder, and it reads my instructions automatically. So once it’s set up, it already knows my business before I type a single word.
That ongoing context is the biggest difference between Code/Codex and the regular chat window you see when you log into ChatGPT or Claude:
| Regular AI chat app | AI business assistant | |
|---|---|---|
| Where your business info lives | Inside that app’s chats, projects, and memory | In folders on your own computer |
| How it remembers you | App memory and any files you upload to a project | Reads its instruction file every session, then automatically pulls in the business files each task needs |
| Saves your corrections for next time | Sometimes, in the app’s memory | Writes them into your files, where you can see them |
If you’ve ever felt like ChatGPT keeps “forgetting” you, there’s a reason for that.
Every AI app has a context window, which is basically like the AI’s working desk. Everything in your conversation has to fit on that desk at once.
When a chat gets really long, the oldest stuff slides off the edge (or gets summarized and potentially loses some details).
When you start a new chat, you get a fresh desk, but many things from the old chat stay behind.
Building a folder system gets around that, because your AI tool will always write the important stuff down in files inside your folders.
Then it pulls any relevant information back onto its “desk” whenever a task needs it, so that it doesn’t lose track of anything you’ve already told it.
Why I built mine in a folder I own
Like most people, I started dabbling with AI by using the regular chat features in ChatGPT and Claude.
It was great for brainstorming, but I was constantly re-explaining my business and how I like to write.
(I’m sure you can relate to that frustration!)
When I was finally ready to take AI seriously as business tool, I joined Authority Hacker‘s AI membership to learn Claude Code. (My workspace’s general structure comes from what I learned there, so they deserve the credit!)
Once I realized I could have Claude Code store all of my work and preferences inside one workspace folder that lives on my computer…
… and that Claude Code itself would be the one responsible for maintaining the content of those folders, based on the conversations we had…
… and that by storing everything locally on my own computer, I would maintain full ownership and control over everything…
… I was hooked.
Let me show you how it’s set up
My AI assistant’s workspace contains one main instruction file and a handful of folders (with more files inside each).
The instruction file is called AGENTS.md. (AI tools love files that end in .md, which stands for markdown. It’s just a plain text file.) Both Claude Code and Codex read it first, every single time I open a new session.
I think of AGENTS.md like the onboarding binder you’d hand an assistant on their first day. Mine covers things like:
- How to talk to me. For example, “don’t dump 20 questions on me at once; ask one and wait for my answer.”
- Where everything should be saved, so every new file lands in the right folder.
- What to do when I correct it, so it saves the fix and never repeats that mistake.
- How to protect my work, like saving a backup before changing anything.
- What it should never do without asking me first.
I keep AGENTS.md short on purpose, because it gets read at the start of every session and a long doc would just waste tokens.
The specific details about my business live in separate files, and the AI only opens them when a task actually needs them.
Those files about my business live in folders, and each folder has its own job:
| Folder | What it holds |
|---|---|
| Brain | Everything the AI knows about my business: my business, my people, my offers, my writing voice, plus my to-do list, ideas, and wins |
| Inbox | A drop zone for docs and screenshots I’ll deal with later |
| Ongoing | Work that never ends, like blog posts, emails, and podcast episodes |
| Projects | Work with a finish line, like a launch or a website refresh |
| Skills | Your saved step-by-step instructions for tasks you do over and over |
| Archive | Finished work and older versions |
(If you’ve heard of Tiago Forte’s PARA method for organizing, it’s the same general idea. I just renamed a couple of folders so they made more sense to me.)
And YOU don’t have to create those business detail files yourself.
When I was learning all this, I assumed I’d have to sit down and write a voice doc and an ideal customer profile from scratch. Nope!
Your AI can interview you, one question at a time, and fill everything out for you. You can even just ramble out loud, since these tools connect to your microphone.
From there, the files keep filling in details as you work.
If I mention “Ugh, I keep forgetting to update my About page,” it offers to add that task to my to-do list. I never have to open the “to-do list” file or make any updates myself!!
Skills are how it stops you from repeating yourself
Once your assistant knows your business, the next step is teaching it how you like things done. That’s where skills come in.
A skill is a saved set of step-by-step instructions your AI follows the same way every time. It’s how you stop copying and pasting the same prompts over and over.
Think about the tasks you do every week: writing your email, turning a recording into show notes, pulling your numbers at the end of the month.
You describe the task to your AI assistant once (or even screen-share yourself doing it), and it writes the skill for you. If it misses a step, you give it feedback, and it updates the skill.
Each skill has a short name and a one-line description of when to use it, kind of like a recipe box where you can read every label without pulling out all the cards.
Your AI only reads the full instructions when it’s about to do the task, so you can save dozens of skills without crowding its desk.
You can even put skills on a schedule. A “weekly wins” skill could run every Sunday morning, check your stats across your tools, and have a summary waiting when you sit down.
Connecting it to your business tools is where it gets really fun
AI tools connect to your other software through something called MCPs (short for Model Context Protocol). If you run a WordPress site, think of them like plugins, but for your AI. Claude has a whole connector directory, and ChatGPT has its own version, too (they call them plugins, instead of connectors).
Once your assistant is connected, you just ask it questions about your data, instead of visiting each tool’s website directly:
- “What were my top three posts last month?”
- “Are any posts slipping in the rankings? Which keywords should I watch?”
- “Which of my emails actually led to sales?”
I connected mine to Google Search Console and Google Analytics so that it can understand my traffic data. (As of October 2026, there’s no easy one-click connector for those, but there’s a more involved way to do it.)
If you’re not sure which numbers are even worth asking about, I broke that down in the website metrics that matter now.
And it doesn’t just read your data. It can do things inside your tools, too.
I connected Kit*, my email platform, so once I’m happy with an email I’ve been working on, I can say “load this as a draft using my template,” and it loads the whole thing inside Kit and picks the right audience.
I still double-check everything, of course, but oh my gosh, it saves me so much time and mental effort!
It gets better every time you correct it
Even better… every correction and decision you make gets saved inside the workspace files, so you never have to repeat yourself.
For example, when I relaunched my Search Savvy program this year, I started calling it a “program” instead of a “membership.”
I told my AI assistant once, it saved that in my business brain, and now every email and sales page it helps me draft says “program” automatically.
The same thing happens with your voice. Whenever it writes something you’d never say, you just reply “I would never say that,” and it adds it to a running list of banned phrases.
It even catches mix-ups. Like, if I tell it a fact that doesn’t match what’s saved in its brain, it points that out and asks which one is right.
Here’s what writing an email looks like now
So what does all of this look like in real life? Let me walk you through writing an email with my AI assistant, start to finish:
- I open Claude Code and point it to work inside my business workspace folder.
- It automatically reads my instructions before I type anything and knows the current context of my business.
- I brain-dump to Claude what I want to say in my email: maybe a story from my week, or the main point I want people to walk away with. (This could even happen while I’m out on a walk with the baby, since I can voice chat with Claude!)
- It pulls in the files it needs from my workspace, like my voice file, my offers file if I mention a program, and my launch notes if the email is part of a launch.
- It drafts the email in my voice, checks it against my never-say list, and I edit it right there back and forth in the chat, just by giving Claude feedback.
- When I’m happy with it, I say “sync this to Kit,” and it loads into Kit’s editor with my template, ready to go.
It handles the difficult techy stuff, too!
For example, Kit lets you show different content to different subscribers based on their tags, but you have to use a coding language called Liquid to do it, and it’s super easy to make a small typo that breaks the whole thing.
Because it was so tedious to implement, I rarely used liquid personalization before. But now, I just ask Claude to add a section for subscribers with a certain tag, and it writes the code for me. IT IS SO AMAZING!!
One important note: I don’t ask AI to “write an email about X” from scratch. The stories, firsthand experience, and personal details are what make content sound like YOU. I use my AI assistant to structure my ideas, not to come up with them.
And here’s what it does for my business every week
These days I work about 20 hours a week, with a 5-year-old and a 1-year-old at home, so every hour counts. My AI assistant takes the tedious and techy work off my plate:
- It keeps an eye on my numbers. I built a dashboard that pulls search traffic, website visitors, email list growth, revenue, and funnel performance into one view.
- It analyzes my launches. After I reopened Search Savvy, it pulled my sales data from Stripe and my email stats from Kit. It found that about 80% of sales came from my email list, about two-thirds of buyers paid in full, and about 20% of my past SEO Made Simple students (out of roughly 1,000) weren’t on my email list anymore, so they never even heard about it.
- It turns my calls into content. It reads my monthly Search Savvy office hours transcripts and pulls out post ideas, mostly for Threads right now. It can even automatically set them up as drafts inside my Threads scheduling tool so that I can quickly edit and schedule them out when I have time.
- It supports my courses. It writes the notes under my course videos and designs lesson graphics in my brand fonts and colors.
- It handles the fiddly tech stuff. When I wanted a free bonus for my waitlist, it set up a private page with a few program lessons on my course site in about two minutes.
Why handing work to AI finally feels easy
All of that is work I used to do myself… and handing it off has been a big deal for me, because I’ve always struggled to delegate.
Some of that was money, but a lot of it was the people-pleaser in me. I don’t love managing people, and I really don’t love asking someone to redo their work.
With an AI assistant, you can ask for as many revisions as you want. You can say “nope, try again” five times in a row, and nobody’s feelings get hurt. You can change your mind at 11 PM on a Sunday.
And since every correction gets saved, you don’t have to give the same feedback twice.
It’s not perfect, of course. Like any new assistant, it makes mistakes, especially at first, so I still check everything before it goes out. But it gets a little better every time.
For perspective: a few years ago I paid a part-time assistant $3k+ per month… and most of what I used to hand off to them, my AI assistant handles now for $200/mo.
This isn’t meant to discourage people from hiring (of course not)…
…but to think about the possibility of knocking out more of the “busy” work with tech so that you can expand the human-side of your business more effectively.
Want to build your own?
If you’ve been reading this thinking “I want one of these”… good news, you can absolutely build your own.
To me, this whole thing feels like the early days of blogging (I started my first food blog in 2010, for reference).
It’s new and a little messy… and the people who make time to understand it now are going to have a huge head start. I want that to be us!
That’s exactly why I built this into my Search Savvy program, for online business owners and content creators (mostly in food and wellness) who want an AI assistant to help them work more efficiently, plus the know-how to get found in Google and AI search.
The first phase of the program walks you through setting up this exact AI assistant, and it works with both Claude and ChatGPT. As of October 2026, that whole section is live and ready for you to implement!
I’m opening the doors again soon. If you want to hear first, join the waitlist and I’ll send you my free Get Your AI Bearings videos right away.
They’ll help you understand how all of this works and figure out whether Claude or ChatGPT is the better fit for you.
Want a head start before the doors open?
Put this post to work
Not sure where to start? Click a button and let AI turn this post into your to-do list.
Erica Julson is a registered dietitian turned digital marketing pro.
She has over 15 years of experience blogging and building online businesses and has taught over 1,000 wellness professionals inside her programs.
Since starting out online, Erica has earned over $1 million dollars in revenue and built 5 different blogs. She now uses her knowledge to help others succeed in the online space, via her YouTube channel, podcast, blog, Facebook group, and online course.
Transparency and integrity are amongst her core values and you'll love her personable, down to earth teaching style.