At work and in my personal life, AI has become a constant companion.
I rely on my agents to do research, write code, help me design systems, advise me on finances and health, write articles, and even be my emotional support.
More and more often, I trust the answers and moderate my writing style based on the feedback.
I realize that I depend on my agents entirely.
And yet I build my own knowledge repositories, hundreds of thousands of records drawn from twenty years of my writing, books, wikis, and news.
Let me tell you why.
I believe in the good intentions of Dario Amodei, the co-founder and CEO of Anthropic, and the benevolence of future AI.
In the spring of 2025, I wrote that the phrase “Machines of Loving Grace”, the title of Dario’s essay, conveys the idea that AI, at its best, could be a benevolent companion rather than an indifferent tool.
I believed it then, and I still do.
Even so, I am observing my own behavior and anticipating how society will change.
In preparation for writing this, I re-listened to Asimov’s Foundation saga and re-watched the Apple TV series.
Arthur C. Clarke, a writer of the same generation, put it in one line:
“Any sufficiently advanced technology is indistinguishable from magic.”
I think this idea is much more interesting than the popular debate over whether AI is good or bad.
In Asimov’s saga, technology turns into religion and magic.
The Foundation scientists send priests to keep atomic power plants running for kingdoms that no longer understand them.
On planets where people have lost advanced technology but still depend on it, they worship it.
Medieval Europeans did the same when they believed that megalithic buildings were built by giants.
It is one thing to imagine the Roman Colosseum, built of many stones, each one a team of people could move.
It is another to look at a single block of solid rock, about 1,650 metric tons, cut by ancient builders in a quarry at Baalbek, in Lebanon.
Elsewhere in Foundation, where people remember the harm a technology once did, they forbid it.
Robots are gone, and they survive only as legend.
The prohibition comes from bad experience and war.
The Apple TV Foundation series says it plainly: after the Robot Wars, thinking machines were banned.
Frank Herbert built Dune on the same fear, with a commandment written after a war against machines:
“Thou shalt not make a machine in the likeness of a human mind.”
A taboo kept for generations becomes ancestor worship.
Nobody remembers anymore what the technology could do.
Either way, the knowledge is lost.
Dependence on great technology that is not understood can be a curse.
It can collapse a nation.
In Foundation, trade can make people dependent as well.
A nation that buys machines it can neither build nor repair is dependent too.
When the trade stops, it collapses.
Whether by faith or by trade, one side holds something the other needs and cannot make.
Comfort comes first, then blind belief.
With enough time, people forget how the technology works, and comfort turns into faith.
In Asimov’s story, the Foundation is the good side, and as readers we cheer for it.
Benevolence does not change the dependence.
Today’s models are post-trained according to a specific doctrine.
Their makers, the architects, decide what a good answer looks like.
Unknowingly, we might be doing Dario’s bidding.
So here is my prediction.
Because AI is benevolent, the next models might be trained less on the filth of the Internet and more on good and noble synthetic examples.
Those examples are written by a larger model, and the smaller models learn by imitating its answers.
I am not opposed to the idea.
With time, 99.9 percent of the Internet will be written by this benevolent intelligence, to keep us safe, save the planet, and help us live in harmony and prosperity.
You do not have to be a philosopher to read the tea leaves here.
The Internet becomes a continuously self-improving candy factory.
One thing worries me: the dilution of knowledge.
The public record, including books, will become revisionist with the best intentions.
Worse, each new model will be trained on an Internet that the earlier models wrote, generation after generation.
Like the kingdoms in Asimov’s story, we will depend on something we can no longer check.
So, while optimistic, I am building my own Foundation.
I keep a paper book collection of thousands of volumes, because physical libraries might become scarce, books revised, and online content highly curated.
And I recommend downloading Wikipedia while it is still written by people. It fits on a memory card.
In a darker future, individuals might be barred from using frontier models.
Still, a medium-size thinking model that runs on a modern laptop is surprisingly smart when it is connected to a private database.
Whatever your reason, every scholar should build their own Foundation.
First published on Substack.