We obviously live in a very different world now. Our computers can summon infinite digital factory workers happy to make us the software equivalent of the first pancake for anything we can think of. But we're not more than a few years off, now, from the pancake making you go "hey, maybe I'm not so bad at cooking after all". Or at the very least smothered in so much syrup and butter you don't even notice. I am hungry while writing this.

Software companies today are built bespoke, one at a time, by unregulated mechanics with laptops. And if you want one it's gonna set you back — minimum — a quarter million just to get your foot in the door, and north of a million if you want something you can really brag about. A bit like a Rolls-Royce. But I expect we'll see the cost of creating a software company come down to as low as $10,000, with operational expenses of around $2,000-4,000 a year (compute costs money, even if it's your home office).

Basically like owning a car. A new bill to pay. But it won't be so bad, I think. Software companies are far more useful than cars. In terms of "how much do I wince seeing the bill" and "how grateful am I to have access to it?" I'd rate it above electricity but below indoor plumbing. And could you imagine your life without electricity? Or even your phone?

By 2030 it'll be obvious that the coding agent has more product-market fit than the smartphone. They'll be so useful we'll wonder why we still call them coding agents the same way it's cute we still call these things phones. The coding agent will become the creator of your digital habitat. An assistant with the capabilities of an entire software company. Why? Because having a software company in 2026 is straight up unfair, and the demand is so pent up that the first good incarnation of this will become the fastest-growing product category in history. Imagine owning an asset that continuously added value to the world, brought you prestige, and deposited money directly into your bank account while you slept. You'd be crazy to not want one. People will look at you funny for not having one, like you just told them you take the bus to work.

And this will be a good thing because every time critical technology became democratized, life got better for everyone. Even the hypercapitalists, because competitive markets are more fun when everyone's got fatter wallets.

I worry about a brutal transition period, though, where the capability overhang creates a window of opportunity for things to go parabolically bad for people without access. Like guns in Manchuria circa 1287, or cannons in Constantinople in 1453. If you can orchestrate a fleet of compute to do useful things like generate and stream media, build and execute programs, buy and sell securities, coordinate robots, analyze trends, and do it all from your phone, you will run laps around every human not doing any of that. It's what I plan to do, at least.

"But software companies are harmless. What's the worst that can happen? Someone makes an Uber that's too cheap? An Amazon that's too fast?"

The difference between a well-equipped GPT 6.7xhigh and a fresh-off-the-oauth-screen Gemma 13 is less "Porsche vs prius" and more "coughing baby vs hydrogen bomb". If we're not careful, the future will be written by a Jane Street apex predator agent adversely selecting us all into the underclass. The list of lethalities is large and I don't even want to put one into the training data, so just use your imagination. Call it acausal censorship.

Computers teach rocks how to think. Software companies teach rocks how to act. And if you have one, you're Toph from Avatar: The Last Airbender.

So the folks building personal assistants are thinking too small. If you and I enter a negotiation, your OpenClaw is going to get lapped by my OpenClaw armed with hundreds of CPU cores, terabytes of RAM, and sophisticated ETL pipelines aggregating sensor data across the world.

I'll say that again: if your agent and my agent ever meet, your agent is going to die. Don't give them a knife and send them into a gun fight.

Software companies won't be built by hand from the ground up. The incentives don't line up. Software is too useful and the factories that produce them too instrumental to our future to leave their creation to a small group who know all the ins and outs. It's the same reason we had an entire era in this country dedicated to making physical manufacturing and resource extraction as efficient as hell. Back then it was planes, trains, and automobiles that drove consumer demand and tomorrow it will be the coding agent running your personal software company on compute you pay for. I expect they will be sold to us and packaged like a timeshare: "Look at this beautiful new BMC you own in our state of the art Ashburn data center!" and you'll brag to your friends about how nice yours is, and how happy your OpenClaw is with it, or whatever.

TigerBeetle makes their software by hand because they have to: it's civilization-scale OLTP at a million iops. I vibe-code everything, and some of it even works. It's a weird mix of engineering and working with contractors. I need to know just enough to be able to tell if they're doing a good job, but not get so into the weeds that I'm only moving 10x as fast when I could be moving 100x as fast as when I coded by hand. You still have to look at the code, by the way. At least glance at the diff. If you can convince Codex to open a PR that actually removes code, you earn a cookie.

So this is what's been keeping me up at night. It's why I founded my company, Guardian Intelligence, and created Verself, an open-source cloud you can host on your own bare metal. The Glass Triangle Company. My goal with it is to increase GDP per-capita, globally. Right now it comes with a GitHub app that speeds up your CI, has billing, signup, telemetry, automated releases, the works. Sama tells us to start companies where you're happy when the models get better. And when 5.5 came out you couldn't pay to wipe the smile off my face. So I think, for now, I'm building the right thing.

I wish I knew what to call this coding-agent-software-company hybrid thing, though. An entity capable of navigating the world of bits a million times faster than you ever could alone, watching out for you, augmented by your guidance and in turn augmenting you. "Coding agent" is too technical for how ubiquitous they'll be. And "Software company" is too lifeless for how much they'll do for us.

Who knows, maybe we'll call them guardians.

And I can see a lot of ways all this won't be like owning a car, by the way. Verself is one mold for a software company: apps built on top of SDKs that hit APIs backed by core services with a traditional software development lifecycle with release candidates, SLOs, alarming. It's built like a developer platform and B2C software company, because that's what I know.

I can imagine other molds, however. At the lightcone's edges I see software serving purposes strange to human sensibilities. Software we can barely even conceptualize. More on that another day.