Software Is Eating the World, AUTONOMOUSLY! - Marketing 2030 🦞 Store

Skip to content

Software Is Eating the World, Autonomously

Remove one word from a company name and you get a movie.

In The Social Network, the moment that matters is not the coding montage or the lawsuits. Those made the movie what it is but there is this scene where Sean Parker leans across the table and tells Mark Zuckerberg to drop "the" from "The Facebook." Two syllables. No new feature. No new code. And yet everyone who has seen the film understands, instinctively, that Facebook became a different kind of thing the moment the word ‘the’ disappeared from its name. It stopped sounding like a service you visited and started sounding like a cool place on the internet you lived with friends and family.

I want to do the opposite. I want to add a word.

In 2011, Marc Andreessen of venture capital firm, Andreessen Horowitz, the firm that invested early in Airbnb, Lyft, Instagram, Pinterest, Roblox and many more; wrote an essay for the Wall Street Journal that became one of the most quoted pieces of business writing this century: "Why Software Is Eating the World." The thesis was simple and, in hindsight, wildly correct. Every industry that could be run by software eventually would be. Bookshops became Amazon. Taxis became Uber. Hotels became Airbnb. Andreessen was writing at a moment when most of the establishment still thought of software companies as a sub-category of the economy. He argued they were about to become THE economy.

He was right. And for fifteen years, people have tried to write the sequel. There was "Models Will Run the World," a 2018 Wall Street Journal argument that self-improving processes would take software’s appetite even further. There was "Software Ate the World, and Soon It Will Write Itself," a 2020 piece anticipating that AI would eventually generate the code that used to require a developer. Each update assumed the same basic shape: software gets smarter, software eats more of the world, rinse and repeat. Incremental appetite. A bigger stomach, not a different creature entirely.

Here is the sentence I think actually deserves to sit next to Andreessen’s original:

Software is eating the world, autonomously.

That is not an upgrade to the original claim. It is a different claim entirely. The old software ate industries by giving humans a faster way to do what they already did: book a room, hail a ride, buy a book and shop online. A human was still the operator. A human still clicked, chose, compared, paid, and complained if it went wrong. A human was still in the loop. Autonomous software removes that human from the loop, from the middle of the transaction entirely. It does not make the human faster. It makes the human optional.

That is the difference between "the" disappearing from Facebook and "autonomously" appearing after "software is eating the world." One is a rebrand. The other is a paradigm shift.

And paradigm shifts are always accompanied, with almost comic reliability, by a period in which everybody loses their mind slightly and calls it a bubble or hype.

The AI Bubble Is Real. So Is the Thing Hiding Inside It.

Let’s not pretend the skepticism is unearned.

I wrote, in a previous book, The Gilded Cage, that if the entire AI industry’s ambition was to turn a stochastic parrot into an artificial general intelligence (AGI), while venture capital quietly subsidized the difference between what a token actually costs and what a subscriber actually pays, and every third person on the planet was busy prompting a picture of a cat in a sombrero, then yes — it does look, from certain angles, exactly like a bubble. A very well-funded, extremely online, occasionally adorable bubble. But an AI bubble.

The financing by AI companies has been circular in a way that would make a structural engineer extremely nervous. AI LLM companies raise money from investors who are sometimes also their customers, who are sometimes also their infrastructure providers, in arrangements that make it genuinely difficult to tell whether revenue is being earned or recycled. The subscription model that made AI tools feel affordable — twenty dollars a month for something that felt like hiring a genius — was never really a price. It was a promotion, and promotional prices, eventually have to come to an end. Anthropic and OpenAI, as they move toward public markets and away from investor subsidy, are shifting from flat subscriptions toward pay-per-token pricing that exposes the real cost of the compute underneath. Suddenly the AI genius has an hourly rate, and the hourly rate is expensive.

Then there is the fight over who actually owns the value once the compute has been rented from AI companies.

Alex Karp, the chief executive of Palantir, said in early July that something had gone "completely wrong" with the token-based economics of the leading AI labs, arguing that closed-source AI companies were essentially renting out compute...