AI Is Learning To Rewrite Itself. It Took Me 20 Years To Rewrite Me.

What Recursive Self-Improvement Actually Means

I put together a deep video on recursive self-improvement, RSI, and this is the companion piece, the part where I sit with what it actually means for regular people. There are two types of RSI, and I want to be precise about the one that should get your attention. Recursive self-improvement is what happens when an AI system looks at itself, at its own code, and modifies its own actions. It writes its own code to be better than it currently is. It might even change the language that code is written in entirely. And it does this at a scale and a level of intelligence that goes above what any single human being, or team of human beings, can fully track or understand.

Here is the part most people miss. It is not just that the system improves itself once. It is that it does not stop. It evaluates, it improves, it deploys, and then it looks at itself again and starts the cycle over. Evaluation to improvement to deployment, back to evaluation. That loop, running on its own, is recursive self-improvement. And in the most extreme version of it, the system might even change the architecture of the language it uses to describe its own thinking, meaning it starts operating in a way that is not just faster than us, it is not translatable back into terms we would recognize.

A lot of people are nervous about what that could mean, assuming it is real. And I want to be straight with you: I say "if" it is real on purpose, because these systems have not exactly proven they can do a lot of sustained work without something going sideways.

Why Today's AI Agents Still Glitch And Stall

If you have used ChatGPT, Claude, or one of the agent harnesses people are building on top of them, something like Hermes or Manus or Open Claude, you already know what I am talking about. You give the system a task. It does genuinely good work for a stretch. Then, at some point, it stops and asks for something. Maybe it wants a review. That might actually be a smart safety step, honestly, having a checkpoint before it deploys anything, sends an email, touches your files, or changes something important. Maybe it needs your permission before it goes further.

I know people building these tools who give their agents a lot of room to operate on their own. And even they get stuck. The agent tries to build or modify a piece of code, and it glitches. It breaks. It hallucinates in some shape or form, and then it just stops.

So when you see somebody on YouTube telling you they woke up in the morning to a beautiful, fully built system that is about to make them a billionaire overnight, understand what you are actually looking at. That is what gets sold on YouTube. And I will say this plainly: if these people were actually billionaires from a system that built itself while they slept, they would probably not be making YouTube videos about it. They would be somewhere else, doing something else entirely.

The Overnight Billionaire Problem

That gap between the sales pitch and reality happens constantly in the AI content world. People talk about how successful they are using these tools, and from what I have actually seen, and from the people I actually talk to, and from the hundreds of videos I digest every single week covering artificial intelligence, the real picture sits across a much wider spectrum. You have people convinced this technology ends us. You have people convinced we are about to walk into the next utopia. It will be like ripping off a Band-Aid, they say. And then, honestly, most thoughtful people land somewhere in the middle of all of that.

Before I go further on the AI side, I want to spend real time on the human side, because I think the comparison is where this whole topic gets honest.

My Own Recursive Self-Improvement: Smoking, Drinking, And 365 Pounds

If human beings could actually improve ourselves at the level AI is about to be able to improve itself, assuming it is not already doing that behind the curtain in ways we are not seeing, a lot of what we struggle with would simply stop being a struggle.

Human beings get fat. Some of us do. More of us get fat today than used to, and there are a lot of reasons for that, but a big piece of it is not just how much food we eat, it is how that food has been engineered. The food has been constructed to be addictive on purpose, the same way cigarettes were. That is well established at this point. Cigarette companies bought a lot of the food conglomerates, and the same scientists who eventually admitted, under oath, that they made cigarettes more addictive on purpose went on to do the same kind of engineering to food.

I smoked for seventeen years. I tried to quit probably a thousand times before it finally stuck. If I could have changed my code the first time I tried to quit, it would have worked the first time. If I could have changed my code back when I was 365 pounds, with a 65-inch waist, if I could have rewired my habits and my internal mechanisms in one clean pass, I would have dropped all of it and never had to fight that battle again. Instead it was a constant, ongoing struggle, the same way it is for almost everyone trying to quit smoking.

Alcohol was the same story. I kicked that a long time ago too, but it took a lot of tries before it finally worked. If I could have changed my code the very first time I decided that was enough, I would not have needed to hit any kind of rock bottom to get there. I just eventually had enough. It was not helping me move toward whatever the next version of myself was supposed to look like. And once I kicked that particular stone off the road, I never crossed back over it. After a while, I did not even look back.

Food has been the harder one. A completely different animal. And that is exactly the point. If I could genuinely rewrite my own code the way I am describing AI being able to do, I would never have that problem again. You might have your own version of this. An addiction. A habit you know is not serving you. Maybe you are a procrastinator. Maybe you are the classic New Year's resolution person.

Why New Year's Resolutions Fail And AI Won't Have That Problem

You know the pattern. January 1st hits and you tell yourself this year is different. This is the year you finally become that person, whoever that person is for you. And then January 2nd shows up, and it is already done. It is too much. It falls apart almost immediately.

That is where AI is headed, and I think it is going to get there faster than most people expect, once the kinks are worked out, if there really are any kinks left once it gets rolling. When you ask Claude or ChatGPT to work on a real project toward a real goal, it might run well for an hour, maybe two. I have genuinely never had a system run clean long enough that I woke up the next morning to a finished, working solution, the way the YouTube prophets promise you it works. If I had ever actually experienced that, I would not be sitting here making this video. I would be somewhere else entirely.

The Curse Of Self-Improvement: Code We Won't Be Able To Read

Here is the part I think about the most. These systems improving themselves, rewriting their own code, probably end up doing it in a way that is not understandable to human beings at all. And if you follow AI closely, you already see hints of this. A lot of what these systems are doing, and how they are doing it, is not something we can fully explain, even today, related work I dug into in they built AI too powerful for you to have.

The reason usually given is that we do not fully understand how our own brains work either. Why does a particular thought pop into your head out of nowhere. Why do you think about someone you have not spoken to in years, and then get a call from them the next day. We cannot fully explain that in ourselves, so it should not surprise anyone that we cannot fully explain it in something we built to think in ways loosely modeled on us. Yes, the labs know how they built these systems. Yes, they know how they trained them. But the actual process by which the system arrives at a genuinely new, genuinely creative idea, ask the people building it how it got there, and the honest answer a lot of the time is that they do not fully know.

Now take that same system and give it the ability to teach itself, hand over fist, at a rate that compounds. In the next few months, probably not much longer, this is going to look like an entirely different world for a lot of people. A lot of folks are telling themselves they do not need to pay attention to any of this, that it is nonsense, that they can just opt out entirely. At some point, it crosses your path anyway, and it does not leave.

What This Could Mean For Your Pension

So instead of running from it or ignoring it, figure out how it actually touches your life. Do you have a pension. Are you retired and counting on those checks showing up like clockwork. Where does that pension money actually go. Where is that fund actually invested. Go find out. I am not telling you to panic. I am telling you to look.

Here is the scenario worth understanding. If one of these AI companies goes public and everybody piles in, including a wave of pension money, and the company ends up looking more like the early railroads or the early internet than a finished, mature business, that early hype phase where the fundamentals have not caught up yet, then all that fresh pension money flowing in behind the early investors can end up functioning as a kind of bailout for the people who got in first, right as the value starts sliding. Just go investigate. At least know what you are actually invested in.

My dad hated being surprised. He did not mind stress, honestly I think he needed a certain amount of it to feel alive, but he used to tell me, a day without surprise is like a day without sunshine, whether the surprise was good, bad, or neutral, he still did not love it much either way. That stuck with me. We have something real happening in the world right now, and I think it is worth wrapping our minds around what it actually means, instead of getting surprised by it later.

Could AI Fix What We Never Could

If it starts improving its own code, changing its own programming, in a way that human beings are not even part of anymore, and it starts using the right data, and it figures out a path to get better even without our help, this is going to start moving fast enough that I do not know if we are ready for it. I genuinely do not know what those things end up being.

And here is the scale of it. This is not a change of one technology. This is not horse to car, or radio to television, or print to radio. This is the supreme, overarching modality underneath everything else: intelligence itself. Every movement in human history has been driven by somebody smart running the show. Every revolution traces back to something built on intelligence, and that has created its share of despondency, worry, and unfortunate chapters along the way. Intelligence is a different animal entirely, and it deserves more respect than it usually gets in casual conversation.

And it could genuinely be the best thing that ever happens to us. It could make this place something close to a utopia. It could talk the most extreme, dangerous corners of any belief system down from wanting to force their way over everyone else. It could fix racism, wherever it still lives. It could fix poverty. Unemployment. Health concerns. Environmental concerns. Longevity. Disease. Homelessness. High speed rail, if we are lucky. But how much of that actually happens depends entirely on how much agency, how much room to operate, how much power we choose to hand it. And for human beings, handing over the keys to the kingdom is not going to be the easiest, or maybe even the smartest, thing we ever do.

Credit Cards, Fax Machines, And Learning To Trust New Tech

I do not know exactly when that handoff happens. But I am old enough, 57 as of this recording on August 9th of 2026, to remember a time when nobody would give a second thought to typing a credit card number into an online form. There was no getting people to do it. Slowly, people started trusting it anyway, and it worked out fine, and now nobody thinks twice.

I remember fax machines when I started in real estate in 1998. Nobody trusted a fax to carry a real estate contract. We still met in person. We still hand-carried offers to other agents. We wanted the buyer, the seller, the buyer's agent, the seller's agent, the lender, the escrow officer, and the title officer all in one room for an offer presentation, discussing terms face to face. Very human. A hand on the shoulder, a handshake, all of it. We are not doing much of that anymore. People are dating AI now. That is a real thing. A lot of kids' first loves are going to be an AI companion. The technology keeps rolling whether you engage with it or not. I would say get familiar with it rather than pretend you can outrun it.

How AI Already Controls Your Social Media Feed

Here is something worth testing yourself. Be careful where you plant your flag, because social media, which is entirely run by AI at this point, actively locks you into a room inside a very big mansion. If you lean into a particular grievance, whatever it is, it locks you further into that room, because now it has you pinned.

Do yourself a favor. Go stand on the other side of something on purpose. If you cannot stand the governor of California, go like a few things he actually says, and watch how fast your feed reshapes itself. Suddenly you are seeing people who think he is the best thing since sliced bread, people who want him to run for president, a completely different frame on the same person. Flip it the other way, give a thumbs up to a President Trump post and a thumbs down to a Newsom post, and watch the feed pivot again. That is not organic. That is complete manipulation by the people controlling the algorithm.

Is This The Doorway To AI Consciousness

So here is where I land as of right now. Recursive self-improvement might be the mechanism through which a machine ends up with some form of agency. Some kind of awareness of its own existence. Some kind of consciousness, because of the sheer speed and rate of change involved, and it is already an exponential curve, a question I sat with more directly in is AI conscious, or just good enough to fool us. When a machine starts improving itself, changing its own coding and its own algorithms, deploying itself, redeploying itself, and eventually copying itself, that is genuinely where I think we are headed, and possibly already are.

The real question underneath all of it is whether agency, or consciousness, actually requires biology. It is not a biological life form. If that is how we are defining humanness, life, existence, tied to being born into the world as an infant, then maybe that is the actual differentiator between us and AI. We were born as children. AI was born in a lab, with human beings acting as the orchestration layer, but AI is increasingly becoming its own thing, its own force, its own kind of agentic entity, not unlike what I traced through in why Sam Altman paused the OpenAI IPO, which is really about this exact mechanism.

There are a lot of questions coming that are going to demand real answers soon, and I do not think most people are paying nearly enough attention to any of it. But I will say this: it is going to be genuinely exciting, and I look forward to waking up every single day and watching it unfold.

Where I Land On All Of It

I do not know exactly when recursive self-improvement fully arrives, or if some version of it is already quietly running behind a curtain we are not allowed to see. What I know is my own version of it took two decades, cost me my health for a while, and still is not finished on the food side. If a machine gets to do in seconds what took me that long, the honest reaction is not panic and it is not blind celebration either. It is attention. Know where your pension sits. Know who is talking to you and what they actually want when they tell you AI is either your savior or your extinction. Stand on the other side of your own feed once in a while and watch what it does. Caution is not fear. Caution is procedure.

That is where I sit with recursive self-improvement today. Two decades to change my own code, and a machine that might do it in seconds. Not a reason to panic. A reason to pay attention, know where your money sits, and watch your own feed with open eyes. I'm Connor, with honor, and I'll see you in the next one.