Very few people have deep conviction in technology, or really even understand technology in vivo.

stagnation

In 1968, “2001: A Space Odyssey” was released to theaters. It was a microcosm of a cultural infatuation we had with the future. Technology had reinvented our lives many times over, and there was this cosmic hunger for what would come next.

We had split the atom, gone to space, and it felt like some MIT professor was on the cusp of traveling through time. The CIA was putting microphones into cats, making dragonfly-shaped drones, and building claws the size of a football field to pick Soviet subs off the sea floor.

In the coming years, after the Cambrian explosion of sci-fi, we began implementing these ideas. Critically, we forgot how to have them.

What would “2001: A Space Odyssey” look like if it were made in the present day? We have no imagined frontiers past space. We have no imagined endangering technologies past AI. It would simply be more space, more AI. When envisioning the future, our awestruck curiosity has soured into “Black Mirror”-style fetishization about the End of The World. In fact, the public doesn’t just disregard technology; it holds it in contempt, viewing it as a tool of subjugation in the hands of a capitalist technocracy.

Increasingly, it seems we are approaching the upper bound of what’s possible under human conception. As if we have mined the space of possible ideas; all that’s left is the drudgery of engineering which spawns wanton pessimism. We found all the cool, useful technology, and those who own the institutions which pervade this cycle will forever own us (i.e. permanent underclass). The engine of progress itself has stalled.

degradation

When not captivated by the future, we turn to the present. The present proves worse than it once was. The president won on this thesis: Make America Great Again. The mayor of New York, with polar-opposite politics, won on this thesis: Make New York Affordable.

In capitalist society, the latent sign of degradation is affordability. Literally signifying a lack of abundance. People have no hope for the future; technology won’t save us, but it’s what makes our lives worse. Society is a closed, zero-sum system to be rebalanced.

The dirty secret is that society is directionally correct. Technology has stagnated. Things are worse, “enshittification”. Capitalism’s competitive pressure towards innovation is no longer.

What you see is all there is. Humans reflexively generalize the visible unto universal truth, when perhaps there are more principled explanations. Is this enshittification only a regime local to the cheap-money era?

The PC revolution is at its tail, and we are slavishly bombarded by a flavor-of-the-week orthodoxy about what to build–one which Silicon Valley inculcated and capitalized and founders took for their own conviction.

Trivially, the normative behavior of Silicon Valley is acting as selection pressure for what exists. Taken to its end, power/knowledge becomes capital/conviction. Who publishes the blog posts? Who kingmakes the merchants of narrative? Everyone wants to be like their heroes… who chose them? It is in fact the institutions of the valley’s job to commoditize culture, repackage it as bridge loans of credibility.[0]

The milieu the founder is seeded from is only seemingly random–pervaded by narratives peddled and capitalized by people no more cognizant than him. A looped, self-assuring game of telephone constructively interfering into truth. If a group of KOLs deem something The Next Big Thing, for fear of missing it, or desire for acceptance, more people will attach themselves to it. Popularity begets itself. Conviction is incepted in the founder, which he mistakes for his own; he unconsciously assumes the folklore to be true simply because it appears everyone else does.

To have conviction in technology is to believe the stagnation is contingent, a contracted horizon of expectation, rather than endemic, a ceiling on what’s possible. Good technology will ride the rails of capitalism as intended, our stagnation is downstream society’s conventions and affectations–the mirage of a false horizon.

We have been focused on software and the personal computing revolution. But all of our attempts (e.g. the Metaverse) have failed to bring our lives online. The mundane, everyday reality of the vast majority of people has not gotten better at the same rate it once did. Or they simply have stopped believing it will.

Companies have begun treating the populace as something to be mined. To be exploited, like feudalists. There is finite value, and we must extract all of it. We have huge corpo-slop incumbent media. A flanderization of franchise. How much more impossible can the mission get?

bits, atoms, and hard problems

If we take a look at the “technologists” of the 60s, they had this intense polymathy. It was plausible that a single person could roughly understand the frontiers of many sub-fields of science.

Now, if you look at the mathematicians and scientists on the frontier, they are dizzyingly hyper-specialized. We are stuck in the wake of the past’s ambitions. The ambitions of modern ambitious people outgrew the small slices of frontier science one can hold in their head. They went where it was unfederated and simple: software.

Software, as a solution space, has been mined for many years. It’s a clean conceptual abstraction. Now it’s becoming disaggregated and the once-neglected stuff at the bottom of the stack is where value is accruing. This is the moneyballification of software startups, as Scott Wu would say. Founders outnumber ideas, and the whole game has been commoditized. The founder is not a convicted contrarian, but AIME-bred labor, grinding 996 on a thesis their VC published as a blog post and assigned out like homework. Gone are the days of the software founder artisan. We have software founder factories: fellowships, universities, and accelerators.

Perhaps, even without new ideas, we can tackle the pertinent issues head-on. We are back to the hard stuff. Still in the engineering slog, but at least, instead of competing in the commoditized, familiar game, one can go where the technology demands.

In one sense, this means hard tech, and in another it means dealing with operational complexity. There is fierce competition among the model providers, which represents perhaps the most complex software to date. And, for some reason, people believe the shovel sellers that will make the most money are solving software engineering problems for agents. Software engineering is simply very easy. There is just an abundance of very bad software engineers due to the sheer size of the infatuation and the raised floor due to AI. If you deeply believe the models will get better, unless you own the bare metal, you shouldn’t be working on traditional dev tools or most SaaS.

The true monopolies (in AGI’s critical path) are in the hardware supply chain. Mainly chips. If you want to start a networking hardware company, you have to tape out an ASIC, which costs a bunch of money. If you actually want to do anything computationally interesting, you’ll find it’s a race to the bottom: silicon.

Then there is the sheer operational complexity of enterprise. There has been no moneyballification here; the closest we have are MBAs. Very few people treat it like an engineering problem. Deploying AI across a large enterprise is extremely non-trivial. It’s a careful dance: figuring out what needs to be rethought from scratch versus automated as-is, locating the source of truth, identifying what’s critical and load-bearing, and finding a way to roll it all out continuously and monotonically. The most permanent solution is a temporary one, and most businesses are in a steady state with no incentive to change. Things work well enough. From the shareholders, to the CEO, to the employees, none want to take risk, none are rewarded for taking risk. Not by the market, not by their boss, not by anyone. And yet, the machine of capitalism demands them to be automated. The job of a founder is to figure out how.

But the current generation of founders weren’t raised to be good operators. And good operators weren’t raised to be disruptive technologists. Perhaps software is a tool to be used in enterprise, but it’s not the hard problem. The hard problem is conceptualizing a coherent mental model of an enterprise, a map you can use to optimize the territory. Then you ask the question of how you can effect change and simplify using software.

Until you suffer contact with reality, your mental model is wrong. And the disruptive software you have won’t work or go to market. In order to suffer the contact, you have to be a good operator. It precedes the software.

In the same way, it’s a mistake to view yourself as a 19yo founder with no experience. You can simply just do the thing and gain experience. Get a first-hand look, become an operator. It’s not as if you are structurally incapable of such things; they just take intentional, concerted effort.

Conviction is downstream of contact with reality.

AI is not a panacea

AI’s killer use case is software engineering. I and most others use it copiously. But it’s not clear advancements in AI alone constitute broad technological progress. Someone still has to use it to build great things.

The software I use every day doesn’t seem that much better. The value AI is bringing enterprise doesn’t seem to exceed token costs at scale outside SWE.

I recently spent >100m tokens debugging an issue. It was a long-horizon task spanning many days, so I mainly let it loop in the background. What I found is that it got 90% there fairly quickly and spent the rest of the time spinning in circles. I spent 1hr undoing all the dumb stuff it did, and then another 2hrs telling it what to do from first principles. And it was solved.

Claude tried many things, and it had, at some points, all the correct ideas, but the wrong premises. What I find is that thinking from first principles gives you a certain conviction to push through to falsification rather than giving up when faced with some opposing evidence.

This is especially salient in AI research and quantitative research. You have some principles and a thesis. You try it and it doesn’t work; do you abandon the thesis or look for engineering or implementation issues? The latter often sharpens your mental model, even if wrong.

A good researcher pushes through to some sort of falsification without becoming obstinate. A common pattern I see is this:

Claude: “Hmm A didn’t work because of x, y, z reasons, let me try some other random loosely related thing”

Me: “If we want C, and A => B => C, which is obviously true. How can you reconcile this with x, y, z and your conclusion?”

Claude: “Actually x, y, z is not evidence A is wrong, I just made a mistake while implementing it.”

(Ilya talking about taste, first-principles thinking, and how it leads to conviction.)

Businesses are a lot like both AI research and quantitative development in that you have some sort of thesis, but in business the thesis is often not capitalized by purely technical ability. How do you know if you need to tweak your GTM, or if the very idea you are pursuing is flawed? You often need to appear irrational past the point the average competent person would before finding success; you find this will through conviction in your thesis.[1]

In enterprise, it is not obvious how hiring more employees would make it more efficient even if those employees were ~free. AI is not yet as competent as an employee and certainly not free.

There’s a graveyard of people extolling the limitations of LLMs and getting proven wrong. But I think there is some sort of deeper issue here.

When I ask it for startup ideas, they lack a specific, cohesive flavor or viewpoint and are mean-seeking. They are not unusual or interesting. Even when asking about engineering problems, it gives mean-seeking advice, when there is a more unusual, higher-leverage bet to be made. It will almost never suggest a solution outside the consensus/hegemony.

This makes me hopeful, as although I may not be as smart as a quant or AI researcher, I probably could have unusual ideas.[2]

motivation

When I was in high school, one of my friend’s parents hung out with us when we did robotics. I used to talk to him about Java, Kotlin (he was a big Kotlin and Antlr guy) and architecting software. He was a true technologist.

It wasn’t until a few months ago that I realized he sold his company to Oracle during the dotcom boom. I literally had zero clue about that. I started asking him about it and sharing about myself, and the way he talked about the valley, the model around why people started startups, was so different from what I was used to.

He simply wrote software he thought was cool, and starting the startup was just how he could do that at the highest level. He viewed all the other startup-y parts of the endeavor as a chore. (He was the CTO, and had operators around him, to be fair.)

I have yet to meet a single founder under 25 in the valley who has motivations as pure as this. The ones who do are just engineers who work somewhere or work on open source. It’s, again, the “moneyballification” of startups; curiosity about technology is secondary to being really good at startups (whatever that means) and wanting to be a founder. But I refuse to believe this is the only mode of startups that can exist today.

If, truly, the min-maxxing founder is the only path to success, it would mean that the technology we envision is not itself marginally useful enough to naturally snowball into a startup. We have to hunt for some sort of arbitrage and exploit it.

We institutionalized startups. Raised a generation on The Social Network. It became cool. I won’t spill more ink on this premise, but you must remember that the proliferation of this culture in the valley is downstream of a lack of conviction in technology and a stagnation in technology.

new media // new technology

I have internally argued, for a long time, that this stagnation is impossible to escape. Accepted that these were the times we lived in. Humanity simply is not ingenious enough, or reality won’t rise up to the cause. If this were still the case, I would not be writing this essay.

My hope comes from quite an unusual place.

The last two movies I watched in theaters were “Marty Supreme” and “Obsession”, both of which are “new media”-adjacent enterprises.

In each case, there’s this incisive exposé of the unspoken parts of human experience. Most new media has this intriguing fascination. Beef, about the grudges we hold, how we go to war when threatened. Past Lives, about In-Yun and our platonic conceptions of romance muddled by the everyday. Marty Supreme, about 2004 Ye and the affliction of ambition.

Each premise, in principle, is mostly ~mundane (on purpose, in the case of ping pong in Marty Supreme). The anti hero’s journey. The arc of the character is not critical; they can stay unchanging. It’s about how the characters act under the circumstances which subjugate them. Seeing common but previously latent pathologies realized and verbalized on the screen.

And they are being rewarded by capitalism. These are not indie films; they spread through word-of-mouth. On the rails of web2 interconnect (TikTok). Box office success.

We are in technological involution. The incumbents have contracted; they have become apathetic to our humanity. Simultaneously, the allure of our technocapitalist republic is that we can change it. And it has already happened in media.

Obsession was produced by someone who got their start on YouTube. Produced on a relatively low budget. Capitalism is this freedom. A deep belief in our ability to just do things. You’re frustrated with a product? Build a better version and put it on TikTok Shop. Ask Claude how to make your vision a reality.

New Media, but where is the New Technology? If we think about how the valley uses web2 rails, it comes from the same culture of exploitation. Rage bait on Twitter. Undisclosed ads in the same format across hundreds of UGC accounts on TikTok. The same formulaic things over and over again.

If the idea you’re chasing is good enough, it will spread. That’s what conviction in technology is. Do you think Mamdani or the creators of Obsession had to pay some TikTok creator to make up a fake story where they name-drop your product halfway through? No, they simply spread their message and thesis. (Of course, with intention and craft. Which itself is the product.) [3]

When you use these new media rails, social platforms, in this way, what it tells me is that you lack conviction in technology. You are not a true technologist. You don’t come from the same Homebrew Computer Club hardware-hacker lineage that those before you did. You are not letting the merits of your product speak for themselves.

I don’t mean to moralize this, but it’s almost taken as a premise that everyone in the valley is a technologist with deep conviction. This is simply untrue. Most are chasing money and status, or infected by a culture that originates from those chasing money and status better than them. I have no problem with this, and almost every day I see someone espousing this as a “moral failure” which I don’t think is true. Money and status are pretty cool. But my point is to take the premise to its conclusion, which points towards stagnation and the end stages of technocapitalism.

It saddens me that the laptop Linux hardware hackers and the broccoli-haired 19yo Twitter founders rarely intersect. To me, this is what true excellence is. A pattern breaker, in service of the good.

It’s hard to say what this return to form in New Technology would look like. Perhaps the nature of humanity will have to be reimagined, per Bryan Johnson and the public’s favorite transhumanist, Peter Thiel. Perhaps it will mirror New Media.

All I know is that it will bring us closer to our humanity (or whatever it becomes) and spread through web2 rails and social platforms.

Be ambitious. Believe in technology. Know what you seek, for that is what you will become.

[0] I am not blaming VCs for this. Capital allocators will always act in aggregate. The VC market is most likely more efficient than ever. I have had mostly good experiences with them and they provide a valuable service. It is on the founder to pursue original ideas and convince the VC to fund them, not the other way around. However, I would tell VCs one thing:

[1] Perhaps I have this slightly backwards. Your thesis should be so actionable and obvious (perhaps only possible with some sort of secret or contrarian knowledge) that following it feels like you’re just bending down to pick up money from the floor. Practically inconceivable why no one else is doing it.

[2] The best part about unusual ideas and convex bets is that I only have to be right once. It doesn’t matter if other people or the AI are right more often if they all think the same way. Unusual ideas have far larger marginal utility even if the chance they are true seems small. This will become increasingly true as AI becomes better. That is, having different ideas than it.

[3] While editing this essay, I saw this beautiful ad (if you can call it that) by Midjourney:

It’s not a surprise it has done so well. The video is designed with intention and craft, and it stands simply as an exposition of technology. I asked Claude if it would work and it said no. I pushed and I couldn’t get it to make a coherent first principles argument from a physics perspective. This makes me incredibly bullish on the prospects of something like this.