import { NoteReference } from "~/components/NoteReference";

sorta rambly thoughts but "framework" for thinking about building startups ready for a [powerful ai](https://arc.net/l/quote/xebyehid); most companies ik and founders i talk to seem to just skirt this problem, and i feel like that was fine to do last year more so but increasingly less fine

i've been finding it hard to see through the fog of war what the future looks like in terms of what are real durable business and not - this is an evolving framework to think about how to build for that

ideas/frameworks to borrow from

- bezos has this quote to the effect of everyone asks what do you think will be different in 5/10 years but the question to ask is what stays the same - this is what you can build a business around
- i love both frameworks of "[marginal returns on intelligence](https://arc.net/l/quote/usalpydw)" and "[compressed 21st century](https://arc.net/l/quote/bkqmrwdy)"

in software, outside the labs or building models to compete where the labs aren't focused, there is a knife edge to walk on where you don't get eaten by the model provider and don't get so easy to build with the existing model that the enterprise does it themselves (this is before use the model for codegen) - i find most ai agent type companies to fall of this knife's edge.

two notes

1. i'm thinking of baseline software commoditizing as a reaper that increases competition, lower prices but not necessarily death of business
2. the knife edge is thinking about how to not be collateral damage from models - i think here it is better to compete head-on and navigate an attack vector that not thinking clearly about it and being collateral damage

even my fav ai company (non lab) - cursor with insane pmf is not obvious to me that have a durable 5-10 year business; _at some point codegen is so good, the comparative advantage in writing code is gone (can be system architecture or product or design but not writing code), before this the codegen is good enough it is like reviewing a pr - ideally on the web._ <NoteReference slug="building-for-agi-future" id={2} summary="counter from friend" direction="right" distance={10}><a className="text-[#8f2a14] underline" href="https://x.com/KaranVaidya6" target="_blank" rel="noopener noreferrer">karan</a> raises a counterpoint that models tend to get to 80/90% on harder and harder tasks before saturating 95-100% on even simple task - this would give the comparative advantage back and need for an ideanother point is cursor could evolve into that next ux on the web or something better.</NoteReference>

i think there are some companies walking this line and sure some will make it out but feels like a very tight line to walk. maybe [composio](https://composio.dev/), [browserbase](https://www.browserbase.com/), … ?

_tangent ramble:_ it doesn't as obviously feel like some traditional software companies are on such a collision course with AI.

some obviously are, just through more competition less prices - this is mostly through commoditising the ability to write that software. that being said this doesn't feel as **acute** as being on a collision course with the lab directly or the enterprise integrating the labs work.

examples here to explain what i mean - it doesn't look like [bun](https://bun.sh/) or [astral](https://astral.sh/) business is meaningfully affected (to whatever extent they had a business) but what about vercel? i could see them being fine? but also maybe dx doesn't matter in this world so they aren't? feels like stripe is fine.

expanding on what stays the same, some ideas, some concepts, some businesses

- airbnb most probably - we still probably want vacations vs uber that seems pretty dead with waymo and tesla eating their lunch, probably still been fine for a while but will whimper and taper out
 - network effect business here are interesting as probably have low marginal returns to intelligence in breaking them
- stripe most probably - unless no income weird ubi, probably paying for things, maybe bridge or whatever is the agents way to pay for things
- sports - nike wouldn't really be in trouble, have had stockfish for years don't really like watching stockfish play chess
- music - think it is super important to be a shared meme, i think ai music tooling to making shared memes is interesting but not sold on the personalised music(vs very sold on personalised software)
- movies/tv - blockbusters probably the same, shared meme; netflix slop can probably be ai generated and personalised
- games - closer to movies some shared meme, some personalised.
- human connection stuff - wanting friends, love
 - probably decent durable business here
- … ? _should expand this list much longer_
- maybe example from conversation is underwriting insurance for AI Agents

### _tangent:_ opportunity cost in a revolution

there will be many things take make $1M, $1M/year even $10M/year but in the last wave of this scale - there were the real trillion dollar market opportunities or at the very least many 100+ billion market opportunities. i think the last 10 ish years didn't have this scale of opportunities outside deep-tech go build the ai revolution

this is all to say, choosing market in a revolution is super important cause the size of the pie up for grabs in enormous - this is much where you run into the steamroller

note: i hate the stay in your lane sorta advice, find some tiny niche or crevice where the steamroller doesn't touch and beg of scraps. imo startups are about building generational iconoclast companies that disrupt industries not begging for scraps in front of a steamroller.

if you are thinking in these lines and want to chat or collaborate, find some time to [chat](https://cal.com/cryogenicplanet)

framework for thinking about agi "proof" startups

really building for an AGI future?

Scalar retro and post-mortem

import { NoteReference } from "~/components/NoteReference";

_**i**'ve made this argument before but want to write it down in one place, none of this is that ground breaking or deeply original - just when i talk to more "normie" friends they don't fully get it and this is trying to synthesize the idea clearly_

the goal of humanity should be maximizing energy use per capita, this means every human should use 1000 times or 1M or 1B more energy than they do now.

this might sounds like a strange goal at first - for many reasons. first why would you want to increase consumption of something and not just get xyz outcome?

before making the point about why increasing consumption is good, let's create some common ground first.

i would argue money is fake - it is just a ledger for "the real currency"/raw input - energy. energy is definitionally wealth. <NoteReference slug="goal-increase-energy-consumption" id={1} direction="right" distance={20} summary="no low energy rich countries"><Img src="/writing/media/energyGDP.png" className="w-[500px] h-auto" alt="energy spend" /></NoteReference> this graph comes up a lot, as to say there are no low energy rich countries.

i think it is fairly intuitive to most how electricity spend is energy use but human labor is a type of energy spend, low efficiency high versatile energy spend. <NoteReference slug="goal-increase-energy-consumption" id={2} direction="right" distance={0} summary="note on efficiency">efficiency does matter but more energy spend is better than less even inefficiently - but yes the more leveraged energy spend you have the better</NoteReference>

so the idea of energy being real wealth is fairly intuitive and well understood. what does this look like for an individual tho? rather than a country?

i would argue that your personal wealth is essentially how much energy you can leverage to make your life better - this can be an AC or using a plane or hiring a maid or at minimum just the food you eat.

so it makes sense that the more energy you consume, the richer you are and that poorer people should go up the energy consumption ladder to reach the level of wealth of the west. but the heart of this piece is that it is **actually really hard to use energy**.

if i told you to go spend 1GW tomorrow <NoteReference slug="goal-increase-energy-consumption" id={3} direction="right" distance={20} yOffset={-500}>1 GW is equivalent to about 1.3 million horsepower, but that's still a bit abstract. Here's a more tangible way to think about it: 1 GW is roughly the power produced by 2,000 Olympic-sized swimming pools filled with jet engines stacked on top of each other 😄 Or more practically: Imagine all the household power used by a city of about 750,000 homes running simultaneously - all their air conditioners, ovens, TVs, washing machines, and everything else running at once. That's approximately 1 GW of power consumption. Another way to visualize it: 1 GW is about the same amount of power as: Doc Brown's DeLorean time machine from "Back to the Future" needed (1.21 GW), A large nuclear reactor operating at full capacity, 3.125 million solar panels operating under ideal conditions, 431 utility-scale wind turbines running at full speed</NoteReference> it would be impossible for you to spend that, <NoteReference slug="goal-increase-energy-consumption" id={4} direction="right" distance={20} summary="average american does 200kWh/day">ChatGPT math but sounds reasonable, <a className="underline" href="https://chatgpt.com/share/67640130-85f8-8011-96b0-fe6b2078586a">source</a></NoteReference>

you would have to spend years inventing new technology to do so. one such technology might look like running your own supercomputer of power AIs to do extremely leveraged work? or using starship to commute across the globe <NoteReference slug="goal-increase-energy-consumption" id={5} direction="right" distance={20} yOffset={-100}>Apparently starship consumes something like 120 GW to lift off? ChatGPT math again tho our boy sonnet contested that so not sure. <a className="underline" href="https://chatgpt.com/share/6763ffad-01a0-8011-bd98-088c5b4477f4">source</a> </NoteReference>. this is all to say that to increase energy consumption to 1GWh/day for an average american would mean them using three orders of magnitude more energy than they do now. this would mean their lives would be unimaginably better than ours.

there is a sense in economics of chasing forever growth, and if that is possible - you are chasing growth for growth's sake as it makes your life better. i think a better way to frame this is chasing how much energy can we consume, and we know the limits to this - this is in some sense moving up the [Kardashev scale.](https://en.wikipedia.org/wiki/Kardashev_scale)

i think the framing that is lost in the Kardashev scale convo is how it is actually about increasing energy consumption per capita - enabling each person more leveraged energy use to make their lives better.

A "good" goal for humanity to increase energy consumption

Goal: Increase Energy Consumption

> note: this is a reskin of an essay i wrote a couple of years back where i polished it up and reposted it after bring this new site up, the original was written by hand this was [vibes written](/writing/vibes-writing)



Going to the gym is **a relic of an outdated human operating system—one designed for a world that no longer exists.** We have all this technology, all this innovation, and yet millions of people still spend hours every week running in place like hamsters, all to counteract the basic fact that we have too much food. This is stupid. The future should be calorie-agnostic. Workouts should be for masochists and hobbyists—the rest of us should demand a better operating system.

### **We Are Malfunctioning Meat Computers**

![Source: wait but why](https://prod-files-secure.s3.us-west-2.amazonaws.com/aa7b5714-a7d8-4439-9feb-4ad60c4a5ac3/c6d846b3-53d6-4c8b-9dac-e93b4399489a/image.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=ASIAZI2LB466RPJQXCBS%2F20260828%2Fus-west-2%2Fs3%2Faws4_request&X-Amz-Date=20260828T042553Z&X-Amz-Expires=3600&X-Amz-Security-Token=IQoJb3JpZ2luX2VjEHwaCXVzLXdlc3QtMiJIMEYCIQCNes%2BvT9fOtbuerUNYrs33WQVn7g22J%2BevgtJfCBzkiwIhAJfCmv0dwkPBLkro1LklrXe9GhoLhWq4v2KbOLTAgRy6Kv8DCEQQABoMNjM3NDIzMTgzODA1Igz7jTQBzZKbxwMUIC8q3AMgLdyUkLKutMcKinaKnbIrdUmvKWMo8tMzOamE0kenqJg6fxqb8PoBw2ftrZnVKoWfTOOLFY618Wi1p0O3YOyck9NwL%2B1n1%2B1Sx8HURPBGWNHfBi%2FbKzyAAgk7gtLbJ6fDTEBnDIiHMYROrWorD6uBc9G8g8E%2BaFZlFsh%2FOJaw9fhyXvwl3Hia8hvvQmD8AvGZj5RcaYYhxVwNI1D4DuoQeJX3MZgSDCnrPrb1AiQo5KsoN9WRNr2vycZycbudn1ZiaT8EQFrFpeOTGFN1GLE6HAcUqonPBnmKIA5Q02CZjMjUfNS96cjr3c0aj0RWBJDFtfM9idXkmGRe84NGJTXRfJImW1%2BmJQkupI0HDRZIITR3FJWYqQkMY1d3esy3fSvx2FkH7Sz8itXfFJyCDFmcEI4AhVsU7UIH9Yl%2B%2BvFhxBWtpDIRwM35B4AK0ygF3fZioqcGKxjznQUVrDEH9qZdH9CL5rQGzjzhkcJkvjmSTWyKwKNn6yF%2B8TZrpaIwrL9NgJaQGHnZ0cYf3VJeqMUHIpR1E%2FrK6oXGR4%2BHRCPoEtCIUmSMVsJ6QDn3wdU5ijnxaClKdk1mbmngiLca4Go8ffMIIe3b2QZHeSf4qJjEH%2BH7e6HXOH%2FArumatTDU%2FMPUBjqkAXPVAYphv39r4RWX5HLbj%2B%2F1bprARTaOk4nmoAAQUPJW6AfokvsJIj%2BXPb76J%2FnZbFvCcspbmC59H2CpRe45%2BCybQZYgbhM5skTTjkic84JUvv6xFVNSedYQHrtlNHZxBJHPKeYyhvjc93iaATUapxKH1zc3EHpcUUa2NyK6oy%2FiTiU8d0LM9Uqp1m2G%2BFSSWtMdoC59punHxrt7tfy4AKJQgTc6&X-Amz-Signature=3ec266c4cb16ede78f19e694496fa7adbca4408fac164e4e226a69d8f5c0b74d&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject)

Nothing about modern life is natural. Our bodies? 200,000 years of software built for survival, not leisure. We were designed to **panic about starvation,** not count macros. Our metabolism is still operating in caveman mode, screaming: **“Store the fat! Famine is coming!”** Meanwhile, famine is not coming. If anything, the opposite—an infinite buffet of dopamine-laced sludge, precision-engineered to hijack our taste buds and make us balloon up like foie gras geese.

A quick aside: ever wonder why cancer is such a big deal now? Because for most of history, **something else killed us first.** Same with obesity, diabetes—diseases of abundance that our bodies were never built to handle.

![Graphic from: [https://youtu.be/USmu0scNcSs](https://youtu.be/USmu0scNcSs) ](https://prod-files-secure.s3.us-west-2.amazonaws.com/aa7b5714-a7d8-4439-9feb-4ad60c4a5ac3/6f1fa63a-80af-4969-bafc-a4faa0c225e0/image.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=ASIAZI2LB466RPJQXCBS%2F20260828%2Fus-west-2%2Fs3%2Faws4_request&X-Amz-Date=20260828T042553Z&X-Amz-Expires=3600&X-Amz-Security-Token=IQoJb3JpZ2luX2VjEHwaCXVzLXdlc3QtMiJIMEYCIQCNes%2BvT9fOtbuerUNYrs33WQVn7g22J%2BevgtJfCBzkiwIhAJfCmv0dwkPBLkro1LklrXe9GhoLhWq4v2KbOLTAgRy6Kv8DCEQQABoMNjM3NDIzMTgzODA1Igz7jTQBzZKbxwMUIC8q3AMgLdyUkLKutMcKinaKnbIrdUmvKWMo8tMzOamE0kenqJg6fxqb8PoBw2ftrZnVKoWfTOOLFY618Wi1p0O3YOyck9NwL%2B1n1%2B1Sx8HURPBGWNHfBi%2FbKzyAAgk7gtLbJ6fDTEBnDIiHMYROrWorD6uBc9G8g8E%2BaFZlFsh%2FOJaw9fhyXvwl3Hia8hvvQmD8AvGZj5RcaYYhxVwNI1D4DuoQeJX3MZgSDCnrPrb1AiQo5KsoN9WRNr2vycZycbudn1ZiaT8EQFrFpeOTGFN1GLE6HAcUqonPBnmKIA5Q02CZjMjUfNS96cjr3c0aj0RWBJDFtfM9idXkmGRe84NGJTXRfJImW1%2BmJQkupI0HDRZIITR3FJWYqQkMY1d3esy3fSvx2FkH7Sz8itXfFJyCDFmcEI4AhVsU7UIH9Yl%2B%2BvFhxBWtpDIRwM35B4AK0ygF3fZioqcGKxjznQUVrDEH9qZdH9CL5rQGzjzhkcJkvjmSTWyKwKNn6yF%2B8TZrpaIwrL9NgJaQGHnZ0cYf3VJeqMUHIpR1E%2FrK6oXGR4%2BHRCPoEtCIUmSMVsJ6QDn3wdU5ijnxaClKdk1mbmngiLca4Go8ffMIIe3b2QZHeSf4qJjEH%2BH7e6HXOH%2FArumatTDU%2FMPUBjqkAXPVAYphv39r4RWX5HLbj%2B%2F1bprARTaOk4nmoAAQUPJW6AfokvsJIj%2BXPb76J%2FnZbFvCcspbmC59H2CpRe45%2BCybQZYgbhM5skTTjkic84JUvv6xFVNSedYQHrtlNHZxBJHPKeYyhvjc93iaATUapxKH1zc3EHpcUUa2NyK6oy%2FiTiU8d0LM9Uqp1m2G%2BFSSWtMdoC59punHxrt7tfy4AKJQgTc6&X-Amz-Signature=2ebc7da8f0cbde5f426919fa3e46b7ba1e33380284d778688e98a446d3ef57e0&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject)

### **Stop Blaming People, Blame the OS**

The dumbest possible take on obesity is **“People are lazy.”** That’s like blaming a computer for running buggy software. We don’t need better *willpower*, we need better engineering.

**We should be solving this with technology, not discipline.** Imagine a world where you can eat anything, whenever you want, and remain **perfectly healthy**—not because you’ve “optimized your macros” like a psycho, but because **science has already solved it for you.**

Palmer Luckey has framed this problem correctly: we should not be trying to guilt or force people into dieting and exercising. That’s a brute-force approach to a problem that requires real technological intervention. If we can solve this at a fundamental level, why shouldn’t we?

### **The Gym Is Just a Commute for Your Body**

Nobody actually *likes* commuting. It’s a tax on your time. Sure, some people love driving, but the majority? **A colossal waste of man-hours.** This is how we should think about the gym. **Most people don’t actually enjoy lifting heavy things and putting them down—they do it because they have to.**

Self-driving cars will eliminate commutes, and **we should be using the same mindset for working out.** The goal is not to “make fitness fun,” the goal is to make it **irrelevant.** You should **be fit by default.** The only people who should be going to the gym are the ones who genuinely love pain and suffering (which, to be fair, is a respectable hobby). The rest of us should have technology do the work.

### **The Future: Food Without Consequence**

We’re thinking too small. “Healthy eating” is a brute-force solution. **We need engineered food—food that tastes exactly like a double bacon cheeseburger but has zero impact on your body.**

This isn’t impossible. We already bioengineer everything from corn to lab-grown meat. Why not take it further? Why are we still accepting **caloric reality** like peasants? We should be eating **synthetic, calorie-perfect, zero-consequence food.** Not *fake* food—real food, perfected.

### **We Have to Take the Blame—And Fix It**

The fact that this problem is still unsolved is a failure of **us, the builders.** It’s easy to point fingers at the food industry, the fitness industry, or bad personal habits, but the truth is, **if we were serious about fixing this, we would have by now.**

Technologists, engineers, biohackers—we should own this failure. We created a world where calorie-rich food is abundant but calorie-burning is optional. We can’t just shrug and tell people to “try harder.” We need to be thinking bigger.

This is something that has been circulating in my brain for a while. It feels like a lingering piece of an idea maze—one I haven't fully mapped out yet, but I know is important. If you're working in this space, I would love to chat, to jam, to be helpful, to collaborate. Maybe this is something I’ll work on myself in the future. But someone needs to build this. **This should exist.**

The future isn’t about fitness—it’s about **making fitness obsolete.** Imagine telling someone from 2100 that we used to spend hours every week *manually burning calories* because we hadn’t figured out a better way yet. They’d think we were **insane.**

### **Final Note: If You Crack This, Don’t Go to Mars**

If you get the biotech upgrade that lets you eat everything and stay shredded, **do not move to Mars.** That’s not going to be a post-scarcity society. If things go south, **you’re going to want your original metabolism back.**

But Earth? Here, in the land of abundance? We have **no excuse** for still doing things the old way. Fix this, or future generations will laugh at us.

The gym is a tax on human failure. It’s time to zero it out.







obesity is an engineering problem not a willpower problem

Working Out Should Be Thought of More Like a Daily Commute—A Problem to Solve, Not Accept

> i don't have the answers and i am thinking out loud, this should sorta be read with my [really building for agi post](https://cryo.wtf/writing/building-for-agi-future)



for years, the battle between startups and incumbents was a fight between **bloat and distribution**. could startups stay lean enough to outmaneuver the giants? could incumbents leverage distribution to crush their smaller, faster-moving rivals? that was the equation. but now, there’s a third axis—**test-time compute**—and it changes everything.

![previously](https://prod-files-secure.s3.us-west-2.amazonaws.com/aa7b5714-a7d8-4439-9feb-4ad60c4a5ac3/dd92691f-4fe9-4282-9149-216801f2936b/image.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=ASIAZI2LB4666BHW2BCX%2F20260828%2Fus-west-2%2Fs3%2Faws4_request&X-Amz-Date=20260828T042554Z&X-Amz-Expires=3600&X-Amz-Security-Token=IQoJb3JpZ2luX2VjEHwaCXVzLXdlc3QtMiJIMEYCIQDdlq7Mh0Vzuufib4OVxZJZW0QP3y3fFjoYOBdudssxvQIhAJdD1ND140hB3H8xLOTZykDuOBddr09diyRhJivoKaEDKv8DCEQQABoMNjM3NDIzMTgzODA1IgxC%2BRPmxVv9T7%2Bjmngq3AO3B1kmVpf%2FmhQ23HkPVyp3Ayw7U9LuwMmMGrpxoO3%2BQJMe5oTd0uuqvWdcUlxoDK09AgMX0aZX4RUdMQSdIBRGLQ%2FE2DtehfYuz52NvNi%2F%2BFX5WEWKab0EBXFGXiSjO%2Bqgdbtw6gN097GFKUTufRusY%2BWIFbhkf36JjVlv7fQIr67R9UBZ7h%2F6LdwzP%2FQgVMxtnKzuFbt%2Fz6jpy6utYHK5WSmNf6hRh48N5qru%2ByOjQT%2Fi3sR%2FAw4i79sycQE%2FcvkloIeUycaBwJFpXC5E22DGDb2Y2hqdJlHE9HE%2BcAnyOa2dWhBsSfgkh4oYaxz5n8uiiogzMGuYQ5Y%2BaBMWoU0lsdoadBnFh56TPP3yeDxUqMnMUMB6qVx7hrUMsBdWSzQGCCFtJEdy15fg2sJMCqIDlUbXAH4ra11wDHTwRp1Tw8bUrS7TSTUEQnLIUQntBfAz87%2F3FFLkAMT%2FTxHKLO7aKkJEFKy8GtvBxfpFVTIgOQvxT7H2UQkSp9TdWRtOB1QhbUpu5G7SX%2FhyIfRNnBu6XB3nZvD1y9SDpMIFnB1NpEoLTyiITLzCRViujAQTdY9rt2%2BuAbCNhenOnnXbqWvq7QxczVwwCTHdHFLDuHHLSR%2BX9Wr9BzUXTOTr7TCK%2FMPUBjqkAZmuytpcFrdTYAnWV8EAl6Seow0xN1MBmGJgXf07pFFdJTQtwVtiVvlbOHF1eppf0y2BqRm9nkVpu8Os9YjdpGqZbVRINRZunara%2BZnpJx7eSPs4J0mZuwvRPkU6ky1eRCSmsWpijdp8HLnOxdFv4%2BOnhk69MJrjx6LQJxGEEsnaDRFyex9sPQdfiXY27YOfJ0JwON02EDbhUYxnVbxAA9lwox7z&X-Amz-Signature=37be0372e93359abd730cae8f1ece904d635f47fe79c3f04633e617d54916bb3&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject)

### what is test-time compute and why it matters

test-time compute is running a model for longer, allowing it to think more. this is what powers advanced reasoning models—your O-series models, your chain-of-thought prompting, tree-of-thought exploration, monte carlo tree search (MCTS). it’s about generating better answers not just by being smarter initially, but by iterating, refining, and reasoning through a problem dynamically. every additional second of test-time compute gives an ai system the ability to re-evaluate, simulate different possibilities, and optimize outcomes, making intelligence itself a function of budget.

in business, many decisions come down to simple capex ROI math—if spending \$50 billion on compute increases conversion rates by 5% and that 5% drives \$60 billion in revenue, the decision is easy: just spend more. incumbents can turn on this loop and pour infinite money into refining intelligence-driven optimizations, whether it’s maximizing engagement on instagram, fine-tuning ad targeting, or optimizing search placement.

test-time compute becomes the ultimate strategic weapon—an unbounded ability to generate better answers by thinking harder and longer. and in an environment where intelligence can be brute-forced through spending, the company with the deepest pockets will always win.

this is why startups **must shift their definition of intelligence**. the old game of 'we'll just be better engineers' doesn't work when the real advantage comes from throwing infinite compute at a problem. startups need to be smart not in raw intelligence, but in picking **problems where intelligence itself isn't the currency of competition.**

### problem selection matters more than ever

picking the right problem has always been crucial in building a successful startup, but in the AI age, it's more important than ever. before, startups could win by building better software, designing more efficient systems, or optimizing workflows in ways incumbents couldn't match. but now, intelligence itself has become a commodity—one that can be scaled infinitely with compute.

this means that startups can no longer rely solely on being 'smarter' in the traditional sense. instead, they need to be smarter about where they apply intelligence. the key is to find **problems where the marginal returns to intelligence are low**—where throwing more compute at the problem doesn't yield significantly better results. this is where startups can still have an edge, not by outthinking in the raw sense, but by out-positioning incumbents through problem selection.

quick tangent on what is marginal returns to intelligence, this is straight from [dario amodei's excellent machines of loving grace](https://arc.net/l/quote/usalpydw)

> Economists often talk about “factors of production”: things like labor, land, and capital. The phrase “marginal returns to labor/land/capital” captures the idea that in a given situation, a given factor may or may not be the limiting one – for example, an air force needs both planes and pilots, and hiring more pilots doesn’t help much if you’re out of planes. I believe that in the AI age, we should be talking about the marginal returns to intelligence  and trying to figure out what the other factors are that are complementary to intelligence and that become limiting factors when intelligence is very high. We are not used to thinking in this way—to asking “how much does being smarter help with this task, and on what timescale?”—but it seems like the right way to conceptualize a world with very powerful AI.

this is the fundamental insight: if intelligence alone determines success, the company that can afford to spend the most on compute wins. but if a problem has other constraints or bottlenecks that startups can solve they could win

### where startups could win

the key for startups is to identify **problems where intelligence is not the primary driver of success**. these include:

- problems where **intelligence isn’t the bottleneck** - maybe compliance, getting things done in the real world or **network effects (this feels like a big one).**
- markets where incumbents **don’t see value in deploying high test-time compute** because they don’t think it’s worth the spend.
- ??? - email me ideas [hey@cryo.wtf](mailto:hey@cryo.wtf)
if intelligence is just one input among many, then startups still have ways to compete but they should make sure to compete outside the axis of intelligence

*i am caveating small outcomes, this is always a thing a 200M business is rounding error for microsoft but might be great outcome for a founder, i am talking at scale*



this essay was [vibes written](/writing/vibes-writing)





how do startups compete with test time compute

Rahul Tarak