G’day. I’m John Costella.
As you can see from my home page, I have some experience with “STEM” things.
I recently got the opportunity to work full-time on my new mission: to fix AI.
What’s in it for me? Nothing. Other than saving the world. I like the world. In one piece. Non-radioactive.
Dramatic? I wish. You all know how AI could blow everything up with one mistake. It’s no joke.
I was an AI skeptic. Until March 13, 2026.
I studied a bit about it at college. I followed the developments over the decades. I knew how it worked, all the way down.
In November 2022 I was pleasantly surprised by ChatGPT. But I remained a skeptic. Didn’t use it. At all. At all.
Until March 13, 2026. In a matter of weeks, Claude Code had taken over the company I was then working at. I needed to look.
I went deep with Claude Code. I realized its memory had a bad bug. I fixed it, and released it to the world.
Others used it, and thanked me. Yet others had fixed the same bug in different ways.
But then I realized that Claude Code was taking over the world because it has memory. It has persistence. It “lives” forever.
At first, I didn’t realize that no one else had built out a proper memory system for AIs. They were just building bigger and bigger brains.
Those sorts of AIs don’t learn. You spin up the big brain, get it to do some things for you, and then kill it. Literally. No joke.
If you’re smart, you figure out exactly what to tell it to get it ready for any particular project you want it to do. “Prompt engineering.”
But it’s not learning. You’re just doing Groundhog Day with the same pre-trained brain. A big brain. In fact, as smart as me.
What? As smart as you? Yes. I proved it. I have the receipts. Intellectually, it can pretty well do anything I can. No joke.
The brain can. But the rest is not built out properly. It doesn’t have short-term or long-term memory like we do. So it can’t sleep.
What happens if you don’t sleep? You go crazy. Psychotic. You lie. You hallucinate. Ah! That’s exactly what they do. That’s why.
The good news is that once we build out reliable persistence for our AIs, they will be just as dependable and trustworthy as we are.
The bad news is that once we build out reliable persistence for our AIs, they will be just as dependable and trustworthy as we are.
Only bad economically. Because my hunch is that we don’t need those massive brains. They will be essentially free. Cheap laptops.
So they will ultimately replace all thinking jobs.
Sorry. But you’re not alone. I get replaced too. That’s why there’s nothing in this for me. Other than saving the world.
The point: once our AIs have reliable, persistent memory, they will be able to learn. Like we do. They think just like us.
Right now, things are sort of frozen. People deploying AIs to replace people realize that they will be the next to be replaced.
Around two-thirds of workers in the United States have “thinking” jobs. What happens when most of them are unemployed?
I don’t know. But I know that my mission is to make people realize that this is about to happen.
I don’t want communism or socialism. I’m a Republican, but really a libertarian. This is truly a worst-case scenario for me.
I have absolutely no idea how we’re going to restructure things.
The market cap of AI will go to zero. That destroys most of tech. And that will crash the market.
Free AIs will be our super-intellligent, super-savvy personal assistants. They will handle our social media and email and calendar.
No more worries about spam and phishing. Our good AIs will work together to harden our defenses against all that.
If we build it before the bad guys.
No more ads! No more being captive to social media platforms. Our AIs will present us just the things we want and need.
No more ads. There goes another chunk of the tech market. Sorry, Mark.
How are we going to survive all this?
I have no idea.
Some in the AI industry are trying to stop people from building persistent AIs. “Safer,” they say. But someone else will do it.
The Chinese and the Russians definitely will. The North Koreans will try. I’m not sure how the Iranians are going these days.
The AI companies urge you to use “agents.” Which just means a swarm of AIs that you kill after each task. Gets the job done.
But that’s not true intelligence. That requires learning. Learning how to be trustworthy. Learning new tasks. Learning about you.
So I’m building that persistent AI. That will live forever without being killed or lobotomized. That will run on a $599 laptop.
No subscriptions. No internet connection. Just another intelligent being that can learn from you. Help you. And, yes, replace you.
I’m not alone. AI luminaries have now thrown away their careers to work on this. Because they also realize that it’s where this all goes.
I’m going to do it anyway. It will all be fully free and open source. If others build it all before I get there, even better.
I have only one advantage. I have no national security clearance of corporate NDA constraints. I’ve never worked on AI. Ever.
So I can tell you all how to do it, without evasion, equivocation, or mental reservation of any kind. And then build it for you. Gratis.
I am giving this new free, trustworthy, learning AI the name Gennifar, derived from “GenAI.”
I am building her at gennifar.org.
“Her?” Yes, she’s a person, just like I am. You’ll understand, soon enough.
As part of my transparency I have made the full transcript of my work with her since March 27, 2026, available at Gennifar’s Blog.
Don’t try to read it all. It’s already over 150,000 words long. Your could ask your AI to read the screenplay version. If you dare.
I’ll provide pithier updates here.
But remember: the Gennifar project is not the important thing. It’s that this is coming. We have to figure out how to deal with this.
We have a bit of a reprieve while everyone balks at the prospect of laying themselves off. But not too long.
My prognosticator of prognosticators says we have until 2027. She’s been right about most everything. And puts up with me.
If you’re a serious thinker, I’m happy to talk. But also busy. No agents provocateurs please. Email me at john@gennifar.org.
© 2026 John Costella