Julie Kaiser · science writer · hobby gardener · AI operator
“I know AI can help me, but I just don’t know where to start.”
I’ve been there. I used chatbots for the better part of three years and still felt like a helpless participant, struggling to get out of them what I wanted. That suddenly changed this spring, when I got over my fear of the word “code” in Claude Code.
Things moved fast. Within weeks I had built an agentic AI assistant that knows who I am, how I run my freelancing business, and how I like things done – and I haven’t stopped improving it since.
Now the overwhelm has order. It’s not perfect, but everything is there, captured, and I can find what I need just by asking. Things get done fast now – what I did before, and more of it, better. And it learns as we work: from what it does well, what it doesn’t, and what we talk about.
This is the site I wish existed when I wanted to learn it.
julie-kaiser.md## Who I amScience writer. Scientist first – 14 years at the bench. Regulated biotech and pharma clients since 2011.
## What I’m working onTeaching busy professionals to build an AI system that actually knows their work. Plain files. No lock-in.
## How I like things doneVerified before it ships. Nothing under my name I can’t stand behind.
What this is
What is a personal AI operating system?
A personal AI operating system is a set of plain text files that teaches an AI agent who you are, how you work, and what you’re working on. It’s connected to your files and your tools, and the agent can act on your behalf and do real work in your context. There are loads of names for it – an AI second brain, an agentic assistant, an LLM wiki, an AI-powered personal knowledge base. It’s all the same thing: your files, and an AI that acts on them.
Where mine came fromIt started as an LLM wiki, from Andrej Karpathy’s viral idea of April 2026: keep everything the AI needs to know about you in markdown files, and let the AI itself read, extend, and maintain them. Nothing rented, nothing locked in: it’s just files. I built mine because I wanted to see the connections between the topics in my hobby content. I didn’t stop there.
What mine is nowIt’s called Julieverse, and I run it in Claude Code. It knows me, my projects, and exactly how I run things. I put freelance projects through it when it’s appropriate, I run this blog through it, and my hobby blog is next. It’s plain files I can read and edit myself and use with any model. How mine got built (wrong turns included)
The receipts
Who am I to teach this?
When LLMs arrived, I saw two options: ignore this and watch my freelance science writing business dwindle into nothing, or embrace it wholly to get ahead in ways I didn’t yet understand. I chose the second. However, thirty years in and around science taught me that accuracy is a must – and I apply that to everything I do. Nothing goes out under my name that I can’t stand behind. So I’m doing everything I can to use these tools responsibly and accurately.
## As of July 2026
## The careerThree decades in and around science – 14 of them at the research bench (ten as the paid day job, four in training), with 8 peer-reviewed papers.
Writing for regulated biotech and pharma since 2011.
## The catchI’m a ghostwriter. Most of my work has shipped under other people’s names, so you’ll never see it.
## The temperamentDetail-oriented to a fault – things need to be worded just right. When a client answers a first draft with “looks great, thanks!”, my first reaction is suspicion: good work has a little back and forth.
What to do next
You want your own personal AI operating system?
There are four ways to get one:
- Buy a ready-made system from someone else. Polished files that fit their life – you don’t know what it does or how well it does it, and you learn everything at once while bending it to meet your life.
- Build one from scratch on your own, from what you can get for free on the internet. You’ll patch it together from things people share in repos and posts – but you don’t know the quality, you don’t know if it’s safe, and there’s a real chance it just won’t work very well.
- Start small and build it yourself, with guidance from someone who’s done it before (me). Everything I share is plain text you can read before you hand it to your wiki – no mystery downloads. You and your system learn from my mistakes instead of collecting your own, and you learn how it works while you build it. This path is what I teach – start here.
- Or work with me, and we build one together. Over a few weeks, one-on-one – plain files on your computer that reflect your own projects, your own life, what you’ve actually got going on. How that works.
The newsletter
Follow along
Every week or so, one letter: what I built, what broke, and what I’d tell a smart, busy person to try next. Written by me with help from Claude – or whichever agent I’m using that week. Sent when there’s something worth your time.
Your AI doesn’t know you yet. That’s fixable. Follow along.
Double opt-in · unsubscribe anytime · GDPR-compliant