Another day another deal which makes you question the meaning of money itself. SpaceX said this morning it has an option to buy Cursor for $60 billion later this year, or pay the coding startup $10 billion for the work they are already doing together, Elon Musk said on his bully pulpit. Chasing the “code” opportunity has been top priority inside xAI, so this adds up.

SpaceX isn’t alone.

Over at Google, Sergey Brin has come out of semi-retirement to personally drive a DeepMind “strike team” whose job is to catch Anthropic in coding. (He wants Gemini to start writing Gemini. Recursive, that.) OpenAI just rolled out a Codex revamp last week with desktop control, memory, and multi-agent workflows aimed at the same target.

And they are all coming to the same conclusion because they are looking at Anthropic with lustful jealousy.

Anthropic says Claude Code is now growing revenues at a $2.5 billion run-rate, a number that has doubled since January 1. Claude Code was launched in May 2025. Six months later it was at $1 billion. Its sales are growing faster than a 1980s F1 monster, pulling the whole company along with it. Anthropic hit $14 billion in ARR in February, $19 billion in March, and around $30 billion this month.

Despite all the hoopla about AGI, and changing the course of human history (and it will), the fact remains that most of these big boys know one thing. Money is in automating software. Money gets the valuations. And valuations get the money.

In plain English, every tier-one AI lab, plus the guy building rockets to Mars, is pouring money, talent, and compute into one thing. Software that writes software. The holy grail (at present) is agents that can code and automate everything. Brin said as much in an internal memo. Anthropic’s own engineers say Claude Code now writes between 70 and 90 percent of their code. A recent analysis found that 4 percent of all public commits on GitHub are authored by Claude. That number was zero eighteen months ago.

And it isn’t just coders. Consider Meta. The company told its US employees this week that their computers will now record mouse clicks, keystrokes, and screenshots while they work. The data will train, in Meta’s own words, “AI agents that can perform work tasks autonomously.” Meta will spend up to $135 billion on AI infrastructure this year, more than the whole company generated in cash in 2025. On May 20, about 8,000 of those same employees, roughly 10 percent of the workforce, get laid off. They are training their own replacements.

Marc Andreessen famously said software was eating the world. Software may be eating the world, but looking around the AI landscape the buffet is starting by very much trying to eat its own. The people who build software are busy wiring up the machines that will build software in their place. Call it automation. Call it augmentation. Call it whatever makes you sleep. What it actually is, is software industrialization, happening right in front of our eyes.

April 22, 2026, San Francisco