on having a brain
a defense of thinking
Apparently, AI will eat the world. The trouble is, I don’t know what that means. We might all die forever, and we are supposed to accept that because uhh China. Or we might have fully automated luxury communism and swim in lakes all day. We might muddle through, and have something in the middle, where AI does the hard stuff and we are all yoga teachers or artisanal, data-center-free poetry producers. Maybe we have mass unemployment, or maybe institutional friction and regulation and real life being messy keeps us mere mortals in the workforce.
The point is: I have no idea what will happen and neither do you. Forecasting is difficult, especially when involving unprecedented, societal-scale deployments of sci-fi alien-summoning technology. But this is frustrating, because generally speaking, I’d like to know what to do with my life. If we’re going to be paperclipped in a few years, maybe I should be surfing in Hawaii instead of contributing to my 401k. If intelligence is too cheap to meter, maybe I should be taking peptides and looksmaxxing, until Claude 6.7 (embodied) tops the beauty benchmarks. On Twitter, the preferred life strategy for panicked tech workers is to vibecode their way out of the permanent underclass (and into a job at Anthropic). As Matt Yglesias writes, basically any question about how to plan for the future, whether at the individual or societal level, collapses into a question about how AI will unfold.
At the same time, there are actions that are robustly good, no matter what happens. You should still exercise and sleep well, because there is basically no scenario in which you would regret doing that. If you are an AI safety researcher with short timelines, you will be more productive if you sleep enough; if you believe “nothing ever happens,” you will still be glad you exercised.
No less important than your body is your mind. Your cognitive agency — your ability to think for yourself, to maintain connection with reality, to control your attention, to create — is what makes you an individual. Without those abilities, you become easy to manipulate, more subject to the whims of the world. In sum, your freedom is at stake.
Yet, I don’t think most people are taking care of their psychic environs, or realizing the mental pollution they are accruing. Social media and short-form video did their damage, but AI is both accelerating and serving as a justification for our collective cognitive atrophy.
I notice this in myself. I run a business, that (prepare for buzzword vomit) uses AI agent workflows to Transform Unstructured Data into Actionable Insights. Startups require speed, and AI coding is fast. The general cultural message in tech right now is “if you aren’t simultaneously running multiple Mac Minis to run a Napoleonic army of agents right now, you’re stupid and dumb and ngmi.”
The natural response to this, besides being anxious all the time, is to maximize your use of AI to speed up building. You have to, right? For a while, I tried this. I would spin up coding agents in an unholy amount of terminal windows, cooking up my half-baked thoughts into features. I built impressive, vibe-coded cathedrals — but, as I’m sure is a common experience by now — discovered they were merely a house of cards when prodded.
The maximal approach did not feel good. I did not know what was going on where, and the rapid, fragmented context-switching between projects and ideas was destroying my ability to focus on one feature and think through complexity. I was just spamming yes, add more slop like an assembly line worker. Which one of us is supposed to be the robot here?
But I’m running a startup. My feelings of overwhelm are ultimately secondary to the God of Speed. The issue is, this approach did not actually work. Bugs and unexpected behavior slipped in, the code was unreadable and overengineered, the AI made strange assumptions, and ended up slowing me down in the end.
Some might respond: skill issue. You just have to use one more framework bro trust me, just set up the proper harness and evals bro, you’re not even MCP-maxxing. And sure, I concede that you can go further in optimizing your setup. I also concede that we are moving up another layer in abstraction. You don’t need to know how assembly code works to write Python; now, you don’t need to know Python to make software. But the fundamental issue is: you still need to be in control over your abstractions, whatever layer you are at. This is important both to feel sane and human, and to actually build lasting, functional things.
I’m reading Matthew Crawford’s book The World Beyond Your Head. Crawford makes an analogy between a toddler playing with a Leap Frog learning tablet, and a gambler at a slot machine. The toddler is frustrated by her body and the world, which responds in unpredictable, uncontrollable ways to her actions, whether throwing a ball or trying to walk. A Leap Frog tablet, in contrast, gives her “pseudo-agency.” Clicking a button reliably produces a song; she can make something happen in the world. In the same way, when a gambler plays a slot machine, the machine becomes a source of certainty and control, even though it is passive and automatic.
Claude Code sometimes feels like a slot machine: when you collapse variability to “accept / reject”, when you retain only a tenuous connection to what you’re ostensibly making, and when you burn tokens more to feel productive than to make anything real. You are diminishing your own agency, despite feeling you are enhancing it. To use Will Manidis’ parlance, agentic coding tools can serve as “tool-shaped objects”: something that “produces the feeling of work—the friction, the labor, the sense of forward motion—but [doesn’t] produce work.”
I use the word “agency” throughout the piece. This might be annoying, given how “agency” is the buzzword du jour in San Francisco. The term has become whittled down to mean a particular, narrow form of Doing Things (cold emailing people, or starting an AI company). It has become synonymous with grasping for ambition and prestige, which, for many, is the opposite of freedom. Twitter dunked on Cate Hall, who is writing a book about agency, because she went to Yale — the assumption being, you can’t tell others how to raise their agency if you have status (even though she writes about how she felt completely unfree while there).
Crawford’s definition of agency, in contrast, is “the experience of seeing a direct effect of your actions in the world, and knowing that these actions are genuinely your own.” Agency is about freedom, and freedom is gained through skill. Expanding your agency means expanding your capacity to do things in the world.
My point is not that you shouldn’t use Claude Code (I love Claude Code!). As Manidis puts it: “The line between the tool and the tool-shaped object is not a line at all but a gradient, and the gradient shifts with every use case, every user, every prompt.” The way I work with coding agents now is more deliberate: building less at a time, thinking about and looping in teammates about features I want more carefully, going back-and-forth before jumping into implementation.
We have “extended brains”: our tools are extensions of our selves. Our brains are plastic and efficient, and will gladly mold themselves based on what we do, or don’t do, with them. Tools can relate to our cognitive capabilities in a few ways. They can form a new skill entirely, a capability that would not be possible without the tool. Tools can augment or provide scaffolding to an existing skill, and serve as “bicycles for the mind.” Tools can also replace a skill entirely. If you are completely offloading a skill to a tool, your brain has no reason to keep that ability itself and the skill will naturally degenerate. Often, this is fine: I am terrible at spatial navigation because I rely on Google Maps, and I don’t care, because spatial navigation is not a skill I value that much.
You should never offload your ability to think entirely. You should use AI in ways that are capability-augmenting, or replace skills you don’t care about, but never to entirely forego thinking for yourself. I use AI to research and steelman arguments and indulge random curiosities. To bring it back to vibe-coding, I find it heartening that so many non-technical people are taking ideas and expressing themselves through code. But if you lose your ability to think entirely, you will suffer, both materially and spiritually.
A corollary: you should write more, and you should never use AI to write. This is because writing is thinking. Using AI to write is one of the worst things you can do to your brain, just shy of huffing paint or bashing your head against a wall repeatedly. The words on the page are almost incidental to the process. Being able to sit with an idea, examine its edges and creases, to hold uncertainty, and to painstakingly capture your emotions and thoughts with the exactitude they deserve, is the root of great art, thought, and individuality. AI collapses weirdnesses, rough edges, and individuality, into a smooth, processed blend of slop.
And honestly? Your brain will not just rot — but you will sound like this.
The world is changing rapidly, and navigating the upcoming years will require adaptability, clarity and realism, and avoiding psychosis of various forms. Already, political and corporate actors are fracking your attention and manipulating you to buy, vote, hate, desire, fear, etc. China wants us to believe data centers are Satanic. The AI industry wants us to believe that China is manipulating us to think data centers are Satanic. Personalized AI that tries to one-shot you will make all of this worse. Don’t compromise your cognitive security even more by losing any remaining ability you have to discern what’s good and true and beautiful.
Careers will change. You might need to adapt yours, or invent a new one. You’ll be better positioned to do this if you retain the individuality that comes from exercising a mind that functions. Everyone else is using Claude too — what is your comparative advantage if you offload everything to AI? As everyone says, “taste” still matters. The way to build taste is by engaging with sustained and diverse inputs, creating things yourself, and maintaining your judgment.
Social media executives do not let their kids use phones. I predict that offloading thinking to AI will similarly become low-status and dangerous (and encouraged for the masses). Crawford writes about the degradation of the “attentional commons”: if you walk into, say, a typical airport, you are inundated with ads, with blaring screens, with sensory overload. In that airport, you can buy access to a business class lounge. The lounge might have better seating or food or whatever, but the primary feature that marks it as luxurious is the silence (what makes it “possible to think”). And as he writes, “it is those in the business lounge who make the decisions that determine the character of the peon lounge.” In the same way, the elite will retain their ability to exercise thinking (assuming they don’t AI psychosis themselves), while degrading it for everyone else.
Defending your ability to think will benefit you practically. But even if we wind up in a terminal state (dystopia or utopia), there is still an intrinsic reason that you should not abandon your cognitive agency. That is: your dignity. As Celine Nguyen puts it, “As a human being, intellectual discovery and gratification are your birthright. Nothing is more worthy, and more self-actualizing, than taking your interests seriously and pursuing them as far as you can go.” Even if we have dystopia/utopia, wouldn’t you want to live the final “normal” years having known you tried? That you refused to slop, wirehead, cheat, or brainrot, and instead, chose to strive for beauty, excellence, and truth? Ezra Klein points out that compared to the discussion around AI unemployment, we underdiscuss the consequences for human dignity. A way to retain your dignity is: creating things, exercising your mind, keeping yourself vital and alive and human.
I started reading the Old Testament recently. They are constantly questioning God, wrestling with God, arguing with God. You should not delegate your thinking even to God.
