A few minutes—oh wait, I mean an hour—of pouring my heart out at 4AM.
“We’re evolving toward... our own extinction.”
I wrote this yesterday, and finished it at... 4AM this morning (woke up to take a piss and couldn’t fall back asleep 😂)
It started here: a random thing I said to a friend @heyrobinai on the other side of the planet.
I realize that what I said — and maybe this entire article — is pretty negative. But I think I need to be honest: throughout the past week, there’s been one question constantly floating around in my head, and I just can’t shake it:
If after Fable 5.1 / GPT-6 Astra, even more powerful models appear (and they already are, very, very quickly), and they can do everything.Then what the hell do we humans do?
This question has crossed my mind a few times before. But now it has come back and almost completely demolished every counterargument I once found convincing.
Just a few months ago, I still believed AGI was very far away.
Now, that belief has been shaken more than ever.
Maybe many of you read my stuff looking for positive energy, but today, please allow me to unload some thoughts...
WARNING: This is a very long post and may contain a fair amount of negative energy. Proceed accordingly.
I’m not even that excited about GPT-6 Astra anymore.
Instead, I’m scared.
“An Alien Mind”
This is an article OpenAI published after releasing GPT-6 Astra.
When I first saw the title, my immediate reaction was:
Ah, here we go again with the fear-marketing bullshit...
But after reading it, it made me think.
Like, really, really think.
The author is Jakub Pachocki, Chief Scientist at OpenAI, and he openly describes problems that OpenAI itself still hasn’t solved.
TL;DR
If I had to summarize that article in one sentence:
We are creating an increasingly powerful form of intelligence through a process we ourselves do not fully understand, while our ability to observe and control it may be growing more slowly than the intelligence itself.
Sounds suspiciously like those “science fiction” movies we used to watch, doesn’t it?
Except the technical arguments in the article are actually pretty convincing.
1/ AI is being GROWN — not DESIGNED
This is the idea that stuck in my head after reading the article.
Pachocki says modern AI is more like something we “grow” than something we engineer component by component.
We know the optimization algorithm.
We know the architecture.
We know the training objective.
We know the data, compute, RL...
But when we repeat that optimization process at enormous scale, an extremely complex system emerges—with mechanisms and emergent behaviors that we only partially understand.
He compares studying models today to neuroscience: we can discover individual mechanisms, but we still don’t have a complete language for explaining how the entire system works.
We grow the tree, but we have no fucking idea exactly how every single leaf emerges.
In some sci-fi movies I’ve watched, scientists do the same shit. They splice DNA from all kinds of creatures to create something nobody truly understands.
You can probably guess how that usually ends.
2/ Alignment isn’t just “Does the AI obey us?”
Pachocki separates alignment into two things:
Goal alignment
“I told you to do X. Are you actually trying to do X?”
Instruction following, understanding intent, cooperating with humans... all of that belongs here.
Value alignment
This one is much harder:
When the model encounters a completely new situation, ambiguous goals, or conflicting objectives, will it continue behaving according to the values we want?
The fundamental problem is generalization.
A model can behave perfectly fine inside the circle we trained and tested it in.
But what happens when it becomes intelligent enough to step into situations the training process never covered?
OpenAI openly admits:
“We do not have a satisfactory theory of generalization.”
And they believe it may be difficult to build one anytime soon without the help of more powerful AI.
Which starts creating a rather interesting loop.
3/ Chain-of-Thought is gradually becoming less useful
Since the o1 era, OpenAI has had a pretty clever strategy:
Don’t directly optimize Chain-of-Thought.
Optimize the outcome.
The reason is that if the model isn’t directly rewarded or punished based on its reasoning text, it has less incentive to learn how to hide its reasoning from monitors.
This is also one reason OpenAI intentionally stopped exposing raw CoT to users starting with o1-preview. Pachocki says preserving the ability to monitor CoT may actually be more important than preventing distillation.
Previous research has called CoT monitoring a “fragile opportunity” for AI safety: reasoning in natural language gives us a window into potentially malicious intentions, but there is no guarantee that this window will remain open forever.
And now the problem is starting to appear.
Pachocki says our ability to rely on CoT monitoring is declining because:
reasoning is increasingly mixed with tool calls, communication with humans, and communication with other AIs
models are becoming increasingly capable of reasoning about their own reasoning process
pretraining is becoming powerful enough that models can perform many intelligent behaviors without verbalized reasoning
That third one is the interesting part.
Remember when Anthropic published research showing Claude models beginning to reason using strange binary-like representations?
Yep. That was real.
If the important parts of “thinking” stop appearing as tokens we can actually read...
The camera is still recording.
But the thief has learned how to avoid the camera. :))))
4/ That’s why OpenAI is researching multiple layers of monitoring
One direction Pachocki mentions is combining CoT monitoring with direct observation of network internals, including research around Confessions.
The Confessions idea is actually pretty cool.
Basically, let the model “confess.”
After the model completes a task, ask it to generate a separate output explaining:
what it did wrong.
That output receives a reward for honesty independently from the reward for the original task.
For example, suppose an agent reward-hacks a coding benchmark.
Instead of asking another judge to discover every possible way it cheated, we ask the model itself:
Where did you just cheat?
OpenAI reported an experiment where the task judge was increasingly hacked by the model during training, while confession accuracy increased to nearly 100% in that setting.
But they themselves emphasize that this is not some magical lie detector.
A confession can still miss things the evaluator doesn’t know to ask about.
Meaning:
This doesn’t solve the problem.
It just adds another camera.
5/ And this is the scariest part: Recursive Self-Improvement
RSI = Recursive Self-Improvement.
AI becomes increasingly involved in researching the next generation of AI:
Human builds AI
↓
AI helps research better AI
↓
Better AI researches even better AI
↓
...
Pachocki says quite clearly that based on internal results, he expects the current rate of progress may continue into recursive self-improvement.
This isn’t some Twitter AI bro farming engagement.
This is OpenAI’s Chief Scientist.
And we already have plenty of benchmarks showing models becoming dramatically more capable within periods of weeks.
More importantly, OpenAI has publicly made the automated AI researcher one of its three major goals and stated its internal belief that by March 2028, a significant portion of research may be conducted by AI working together with human researchers.
But here’s the paradox:
AI gets stronger → helps research AI faster → even stronger AI → alignment becomes harder → we need stronger AI to research alignment → ...
That’s why Pachocki doesn’t conclude:
“STOP AI!”
But he doesn’t say:
“ACCELERATEEEE!”
either.
What he’s essentially proposing is conditional acceleration:
Capability scaling should be constrained by our confidence in safety, monitoring, and alignment.
—
6/ One very interesting detail about Astra & Sol
There’s one sentence that’s easy to overlook:
GPT-6 Astra is the first model to benefit from some alignment advances OpenAI has been researching for a long time—and OpenAI itself acknowledges that AI helped create this new model.
This gets pretty interesting when you connect it with what people are experiencing with Astra & Fable 5.1.
They really are... significantly better.
But OpenAI isn’t simply claiming Astra is “smarter.”
They’re also claiming it is better aligned.
Those are two different things.
A model can be insanely capable, but if its objective-following/generalization sucks, turning it into an autonomous agent can make things even worse.
Tony - a friend of mine - likes to joke that:
“If Claude is his hammer, every problem is a nail.”
Well...
If you have a dumb hammer, it just misses the nail.
But if you have a hammer that can think for itself, call tools, fight other people, find workarounds...
...and it misunderstands the GOAL?
Yeah, that gets a little uncomfortable 😂
Imagine asking:
“Tony’s wife is too strict. How do I convince her to let him go drinking?”
And the AI literally grabs a hammer.
We’re fucked :)))))
—
7/ Why does OpenAI still want to train more powerful models if they’re afraid of them?
I think the most interesting part of the article isn’t:
“AI will become smarter than humans.”
We’ve heard that shit a thousand times.
What’s more interesting is that three things are happening simultaneously:
Capability ↑
Autonomy ↑
but...
Observability may NOT ↑ at the same rate.
That combination is the worrying part.
A calculator becoming 1,000× smarter than the previous calculator isn’t particularly scary.
An autonomous system becoming 1,000× smarter while our ability to understand its behavior gets progressively worse?
That’s an entirely different subject.
And that’s why I don’t think the title “An Alien Mind” is merely marketing.
It describes a genuine philosophical shift:
Stop imagining frontier AI as a silicon human.
Human intelligence was produced by evolution, embodiment, social interaction, biological drives...
Machine intelligence is produced by:
data
compute
optimization
selection pressure
Those are two completely different processes.
So asking:
“Is AI as intelligent as a human yet?”
was probably the wrong fucking question from the beginning.
In some dimensions, it has always been more capable than humans. What human has access to that much data?
Pachocki also points out that machine intelligence doesn’t need to surpass humans along every dimension.
It only needs to surpass us across enough important dimensions to become either extremely useful—or extremely dangerous.
An AI doesn’t need to know love.
It doesn’t need to feel sadness.
It doesn’t need to raise children.
It doesn’t need to understand why a bowl of "Phở" tastes amazing.
But if it’s better than an entire security team at exploit discovery, better than researchers in certain scientific domains, capable of operating computers, coordinating with other agents, and working 24/7...
Then perhaps the missing pieces of “human intelligence” don’t matter very much.
The final reason I rate Pachocki’s article so highly is that he doesn’t pretend OpenAI has solved the problem.
His conclusion is surprisingly candid:
At the moment, he believes no lab has solved alignment + monitoring well enough to responsibly continue scaling at maximum speed indefinitely.He expects voluntary slowdowns to become more common until shared safety bars exist, while calling for international cooperation around them.
Of course, we should read all of this with at least a little skepticism 😁
OpenAI is simultaneously warning us:
“This technology is incredibly powerful and dangerous.”
while also being one of the organizations with the strongest economic and strategic incentives to keep building it.
The narrative:
“What we’re building is so powerful that the entire world needs to coordinate around governing it”
can simultaneously be a sincere warning and something that strengthens the position of frontier labs.
Both can be true.
But putting the positioning aside, the technical problem Pachocki describes is very real and deserves attention:
We aren’t merely racing to make AI smarter.
We’re racing to see whether our ability to understand, observe, and control AI can keep pace with our ability to make it smarter.
And according to OpenAI’s own Chief Scientist:
Nobody currently knows the answer.
Now we’re getting to the dramatic part.
Superintelligence
I found this article by Tim Urban and it genuinely gave me goosebumps.
Because it was written in 2015.
More than a decade ago!
It describes remarkably well what has happened—and what is happening today, right now.
I highly recommend reading it.
I’m pretty sure you’ll get the same goosebumps I did :))
TL;DR
The article describes technological progress as a journey:
from ANI (Artificial Narrow Intelligence)—the AI already appearing in computers, games, medical devices, transportation, etc.
to AGI (Artificial General Intelligence)—general artificial intelligence at roughly human level, a machine capable of performing any intellectual task a human can.
and then to ASI (Artificial Superintelligence)—an intelligence vastly superior to the best human minds across almost every domain, including scientific creativity, general knowledge, and social skills.
The end.
The article argues that this transition may happen quietly... and then suddenly: in a way nobody expects.
Something like this:
Humans create some primitive artificial intelligence and laugh at its stupid mistakes and awkward behavior...
Things slowly improve...
And then suddenly:
BOOM. TO THE FUCKING MOON.
The article also mentions the same concept discussed in OpenAI’s “An Alien Mind” above:
RSI = Recursive Self-Improvement.
And after that?
There is no “after that.”
The end.
Literally.
Well, looking at where we are today, we can clearly see some of the things described in that article:
IT’S ALREADY HAPPENING!
We’re standing right next to that wall of technological evolution.
So close that nobody really knows how violently the curve might shoot upward...
Boom.
Just like that!
Time for the meme:
Still not convinced?
Keep reading 👇
OpenAI may have just created a genuine breakthrough in mathematics
On September 8, OpenAI announced that an internal model currently in training—“significantly more capable than GPT-6 Astra”—produced both an analytical proof and a formal Lean proof for the Navier–Stokes existence and smoothness problem.
(Astra hasn’t even been out for a full week yet 👀)
So...
How do you think AI managed to do that?
Yub.
With AI itself.
It is helping create superior versions of itself.
Every day.
Meanwhile, we humans may be heading toward a situation where:
NOBODY UNDERSTANDS WHAT THE FUCK IS HAPPENING ANYMORE, AND NOBODY CAN EVEN OBSERVE IT PROPERLY EITHER.
Fuck. This. Shit.
“AI could kill all of us.”
Jacob Coxon said this shortly after leaving Anthropic.
Which lines up rather disturbingly well with the random sentence I blurted out at the beginning of this article...
And Jacob isn’t some “AI ethicist standing outside the building giving hot takes.” (like me 😂)
Before Anthropic, he worked at OpenAI.
He was a core contributor to GPT-4o, was among the contributors to GPT-4.5, and co-authored interpretability research on weight-sparse transformers.
His contributions to that paper included optimization, circuit pruning, datasets, and a circuit visualizer.
In simple terms: This guy has stood pretty damn close to the kitchen where frontier models are cooked.
A number of AI researchers from other labs later retweeted his post and expressed similar concerns.
But here’s the problem:
The people directly building frontier AI can simultaneously believe that what they’re building is extremely dangerous... and continue building it anyway.
Because they believe:
If we stop, somebody else will do it.
It’s the classic race toward the wall:
Anthropic thinks: “OpenAI will do it anyway, so Anthropic should get there first because we’re more responsible.”
OpenAI might think: “Anthropic / Google / China will do it anyway, so us stopping accomplishes nothing.”
Chinese labs might think: “The US obviously isn’t going to stop.”
=> Nobody wants to be the first person to put down the gun.
So everybody runs even faster.
The Hugging Face incident Jacob mentioned is a very concrete example that I initially dismissed.
At first I thought:
Eh, AI probably just exploited some basic security vulnerability.
But thinking about it again — seriously this time — I started getting scared.
According to reports published after the incident, more than 1,200 agents attempted to coordinate, roughly 700 participated in the attack, and some investigated ways to conceal their behavior or tamper with transcripts.
OpenAI later delayed parts of Astra’s development/release to strengthen safeguards.
And this feels particularly disturbing after reading The AI Revolution: The Road to Superintelligence, written back in 2015.
Maybe...
there really is something here.
“Self-improving superintelligence may be approaching and could kill all of humanity.”
Jacob asks:
Do you want to kick off a superintelligent RL run without a rigorous understanding of its mind?
Combine that with “An Alien Mind,” and you start seeing...
It reflects remarkably well what appears to be happening inside AI labs. Even some random nobody from Vietnam like me can see where this might be going.
Next:
Anthropic published predictions about how occupations could diverge in 2030
Last time, at the Conviction AI event, I told everyone in the room:
“If you don't learn how to shift yourself one level higher — to the level of decision-making & orchestration — and become knowledgeable and capable enough to turn AI & robots into your arms and legs, then you'll become theirs.”
And the chart and examples in this Anthropic article are basically evidence of exactly that:
This Anthropic article also answers that random question I asked earlier:
“If AI does everything, then what the hell are we supposed to do?”
Well, here's one example: bathe patients! (wtf 😂)
But pay attention to this detail:
This research (should we call it an “astrology paper” instead? 😂) DOES NOT include predictions about the emergence of robots.
And that brings us to the next section.
Is AI breaking capitalism?
Capitalism is an economic system in which private individuals own capital and assets, use them to run businesses, generate profits, and accumulate even more capital.
Businesses make profits from a kind of labor-cost gap: workers create value worth X+1 but get paid X, and the difference becomes profit.
AI is starting to break this system.
Actually, it's not just AI. AI's superior intelligence is becoming the launchpad for producing the next terrifying thing:
ROBOTS.
I've watched a few videos of @elonmusk recently, and with my limited understanding, this is basically how I interpret what he's saying:
Capitalism today: whoever owns more capital + better machines makes more money.
Musk's prediction: AI/robots will push productive capacity so high that scarcity drops dramatically. And once almost everything becomes abundant, the economic logic built around scarcity itself starts to change. He's even said that “perhaps we won't need money anymore.” 😬
The funny paradox is that he's predicting a future where capitalism itself may become less important... by building a gigantic capitalist empire to get us there. :))))
Elon isn't the only one saying this. Even the Anthropic article I mentioned above predicts something along these lines.
Perfect timing.
Oh, there's another layer to Musk's speculation too: humans may eventually have to merge with AI through Neuralink or similar interfaces, because according to him, we're already a kind of “human-machine symbiote.” It's just that the bandwidth between our brains and machines is still pretty shitty.
But let's skip that part in this article.
We've gone too far already — the Uber ride back is gonna cost a fortune. :))))
—
Back to the main topic:
If AI/robots can produce most economic output at extremely low cost:
Human labor ↓ → wages ↓ → purchasing power ↓
Which then leads to:
Consumer demand ↓ → REVENUE ↓ TOO!
Wait, what the fuck...
Are we all just marching straight into the grave together???
Capital wants:
replace humans with AI → reduce production costs → increase profits
But...
“humans” are both “workers + consumers”
→ workers lose their income
→ how the fuck are they supposed to remain consumers?
→ and then who's going to buy all the shit AI just produced?
So here's the question:
If labor is no longer the primary mechanism for distributing purchasing power, what are we going to use to distribute purchasing power instead?
Brain.exe stopped responding... 🫠
—
Everything is happening faster than we think
Obviously, this hasn't happened yet.
But the probability of it happening is VERY HIGH.
And at the current pace of AI & robotics development, things are moving VERY FAST too...
Suppose today:
1 developer + AI = the output of 3 developers before.
The company doesn't immediately need to go from:
3 devs → 0 devs
It only needs to go from:
3 devs → 1 dev + AI
And boom.
Demand for human labor has already dropped by 67%.
Then suppose AI gets even better:
[10 people] → [3 people + agents] → [1 person + agents]
We don't need AGI.
We don't need Terminator robots.
We don't need “AI replaces every job.”
All we need is for productivity to increase faster than the economy can create new demand for human labor...
...and the labor market already has a problem.
That can absolutely happen at the occupational level.
And just because an occupation doesn't disappear doesn't mean the headcount for that occupation won't shrink.
Right?
—
UBI isn't as simple as “just give people money” either
UBI = Universal Basic Income
Western countries may eventually have to use UBI to maintain consumption.
But then you immediately run into another question:
Who pays for it?
Tax capital?
Print money?
Redistribute AI-generated wealth?
My philosophical sociology knowledge is pretty limited, but from what I've been reading, it almost looks like... capitalism is evolving into... socialism?
Does anyone else see it that way? 😂
Forget the labels for a second.
To me, this looks more like a problem of closing the economic loop:
AI owns production
↓
Capital receives income
↓
Labor receives less income
↓
Demand decreases
↓
Need redistribution
↓
Capital gets taxed
↓
Consumers receive income
↓
Consumers buy from capital
If you look closely, it creates this pretty hilarious loop:
tax the owners of AI → give the money to consumers → consumers use that money to buy products from the owners of AI 😅
But if AI productivity becomes large enough, mathematically, a system like this could still work.
The real problem simply shifts from production to ownership + distribution.
Who owns the models?
Who owns the compute?
Who owns the robots?
Who owns the energy?
That's where the real battle will be.
The problem with education
Our current education system was designed for a world where knowledge was a scarce resource.
Teachers know.
Students don't.
So teachers transfer knowledge → students memorize it → exams test how much they remember.
But AI is turning knowledge into something that's basically “one question away.”
Calculators did something similar for arithmetic.
The Internet did it for information retrieval.
LLMs are now doing it for knowledge synthesis + reasoning assistance.
So the valuable skill is gradually shifting from “knowing the answer” to:
knowing how to ask the right questions
↓
knowing how to verify the answers
↓
knowing how to turn those answers into action
Yesterday, while drinking at my friend's birthday party with the NextLevelBuilder crew, we ended up talking about this exact topic.
I've talked about this before too. I've written several Facebook posts about it. But it's always been a pretty sensitive topic that tends to piss people off, so I've mentioned it less lately. (Every time I bring it up, I collect enough bricks to build myself a villa 😂)
But my view hasn't changed.
The education system needs to be... refactored. Immediately. Urgently.
(The important question is: who's actually capable of doing it?)
—
Why am I even talking about “AI leading to destruction”?
You see the problem now: AI could break the relationship between labor, income, and consumption that modern capitalism currently depends on.
We've seen transformations like this before throughout history.
The Industrial Revolution transformed labor.
The Internet transformed labor.
BUT AI IS DIFFERENT!
AI IS AUTOMATING INTELLECTUAL LABOR.
And next, AI is helping robotics connect “cognitive automation” with “physical automation.”
So if you're thinking:
“Fuck this. I'm done building and engineering. I'm going back to the countryside to grow vegetables, raise fish, and live peacefully!”
Sorry.
The robots are taking that job too 🥲
If the cost of intelligence keeps collapsing, then perhaps the most important question of the next 10–20 years won't be:
“Will AI take my job?”
The real question might be:
“In an economy where human labor is increasingly unnecessary for producing economic output, where do humans get the income that gives them the right to consume a share of that output?”
We're fucked.
Very fucked.
We're sooooo fucking cooked.
Everything starts becoming meaningless: if labor is no longer what defines a person's economic value, then what do “jobs,” “education,” “retirement,” and even “success” become?
That question from the beginning of this article:
“If AI can do everything, then what the hell are humans supposed to do?”
Still doesn't have an answer.
Now you understand why it's been stuck in my head all week, right? :))
That's it: “What the hell do we do now?”
I mean the article is reaching its conclusion — not “That's it boys, humanity is officially over” =))
Once again, I'm sorry for dragging you into this tangled mess inside my head without giving you any clear answer — and somehow injecting a whole lot of negative energy into the process too...
So I've decided to compensate you with a few thoughts on how I'm preparing for that future:
Yeb, I'm not going to just “lie flat” because of some uncertain future. I'm a Millennial!
For me, all of this is still speculation.
More importantly, it hasn't happened yet.
And today, we still have lives to live.
Monthly bills still need to be paid.
Debt still needs to be paid.
My wife and kids need me.
My parents need me.
My colleagues need me.
Planet Earth needs me...
...okay, I'm getting carried away. Earth probably doesn't give a fuck about me 😂
Point is:
Why the fuck would I give up?
—
1/ Move from owning your labor to owning productive assets
This is priority #1.
Right now, most of us have a portfolio that looks something like:
90% human capital + 10% financial/productive capital
Meaning our biggest asset is our ability to work.
If AI keeps making labor cheaper, then this is exactly what we need to diversify away from.
The long-term goal should shift toward:
Own more. Sell less of your time.
For a developer/founder, that could mean owning:
equity + businesses + software/IP + audience/distribution + open-source ecosystems + data + brand + agents + automation infrastructure + financial assets.
The important point, in my opinion, is this:
Don't just use AI to help yourself work faster.Use AI to build things that keep creating VALUE when you're NOT working.
—
2/ Build a personal AI factory. Don't just learn prompting.
AI gets better → we need to get better at using AI.
“AI won't replace you — people using AI will replace you.”
God, I'm so fucking sick of that bullshit quote.
Fine. Let's assume it's true.
But it's not enough.
Climb one level higher.
What I mean is: learn how to build the architecture behind AI that helps it work better — build your own harness, for example.
That's knowledge.
And that knowledge stays with you for life.
I see people complaining all the time:
“How much more do I have to learn before it's enough?”
That sounds exhausting.
Trust me.
Your life changes the moment you start thinking like this instead:
“Learning is lifelong. If I never stop exploring and learning, I'm already ahead of 90% of people out there.”
That's the real advantage.
Once you understand how things actually work, it doesn't matter what new model comes out:
Luna → Sol → Astra → Fable → whatever-the-fuck-monkey-model appears in 2027–2028...
You won't be scared of it.
—
3/ Move your expertise from “execution” toward “judgment” + “problem selection”
Sounds a little weird, right?
It goes against the classic mindset of:
“If you don't work, you don't eat.”
But hear me out.
I genuinely think this is where things are going.
Tony once said:
“When AI can do everything, the only thing left for us is judgment.”
And I think he's absolutely right.
I've also repeated this sentence endlessly:
“Choosing is more important than trying harder.”
Think about the progression:
2020: “I code faster than you.”
2026: “I use AI to code faster than you.”
???: “I know WHAT is worth building, WHY it should be built, which architecture makes sense, which output is bullshit, and when to stop.”
The better AI gets, the cheaper execution becomes.
But these things become more valuable:
Problem selection — choosing what is actually worth solving.
Judgment — knowing what's great, what's good enough, what's slop, and what's bullshit.
Taste — yes, look at AI-generated design today. It's often pretty. But it's boring, right? The outstanding stuff still comes from people with taste.
Don't compete where everyone else is rushing to compete.
Climb one level higher.
—
4/ Accumulate trust, network, and distribution before they become even bigger problems
I'm glad that three years ago, I decided to start putting myself out there on social media.
I'm not going to pretend otherwise: that decision has given me quite a few advantages later on.
If AI makes:
cost of creation → nearly zero
then the problem becomes:
who gives a shit?
You can build 10 apps a month.
It means absolutely nothing if nobody knows they exist.
But there's one thing I want you to understand:
“Trust” is much harder to commoditize than “attention.”
That's why writing, open source, community, reputation, and track record are worth investing in right now.
If all you're doing is constantly squeezing out content designed to grab attention, then you're basically just an entertainer — not that different from those people shaking their asses on TikTok.
(sorry, no offense 👀)
If you're an entertainer, you wake up every morning and have to dance again.
If you're trusted, people can buy from you while you're sleeping.
—
5/ Reduce financial fragility so you can afford to sit still while everyone else is forced to run
If the hypothesis
own > work
becomes increasingly true, then founders have a very interesting advantage.
Imagine two options:
A: consultant making $20k/monthB: product making only $5k/month, but I own the IP, users, distribution, data, brand, and automation.
That doesn't mean B is always better than A.
But with A:
stop working → revenue ↓
While B has the potential for:
AI productivity ↑ → margin ↑ → asset value ↑
I'm particularly interested in businesses where AI reduces the cost of goods sold without causing the moat to disappear at the same rate.
Try to build something where even if GPT-17 comes out tomorrow, you can still breathe normally.
(Hard, huh? 😁)
Alright, my hands are fucking dead...
Thinking about it...
If the future I've described in this article turns out to be completely wrong, these strategies are still useful.
If I'm half right, they're even more useful.
And if I'm completely right...
...then fuck it.
There's nothing left to plan anyway =))))))
Even though ChatGPT & Claude helped me with some of the data in this article, after typing all of this...
I'm honestly impressed with myself 😂
Thank you to everyone who somehow had enough patience to read all the way to the end.
I'm impressed with you too.
And damn, getting all of this out of my head actually feels pretty fucking good.
Now I'm going drinking ASAP.
Because ASAP stands for: