Tech & AI Twitter Unfiltered: Cursor’s $60B Exit and the Cracks Beneath the Hype

Cursor, the fastest growing software company in history, just sold to XAI for sixty billion dollars. Last week we talked about the raw power of the big foundation models. This week, we saw what happens when you build a ninety-story skyscraper on one of those foundations… and the ground underneath starts to move. The story of Cursor’s exit isn’t about a failure. It’s about a success so meteoric, so impossibly fast, that it revealed a crack in the very fabric of the new AI economy. They hit two billion dollars in annual recurring revenue in thirteen months. They signed multi-year contracts with seventy percent of the Fortune 1000.

And then, they sold. Because the founders saw the path from two billion to one hundred billion dollars, and they decided they were not willing to underwrite that risk. We're going to come back to exactly what that risk was—and why it affects every single person building in tech today. But first, let's sweep the rest of the landscape. The tectonic plates are shifting everywhere. OpenAI just partnered with Yubico to launch Advanced Account Security for ChatGPT. This isn't a software patch. This is a hardware solution. They’re selling co-branded YubiKeys for sixty-eight dollars a pair. When you enable this, your password login is disabled.

Your email and SMS recovery are disabled. You have two cryptographic credentials, and if you lose them, there is no customer support line to call. OpenAI is now treating access to its AI the way a bank treats a vault or the way crypto treats a wallet. This is a direct response to the over one hundred thousand stolen ChatGPT credentials that were found floating on the dark web back in 2024. The era of using your cat’s name as a password for the most powerful tool on the planet is officially over. Meanwhile, the Trump administration just put a hard stop on Anthropic’s plan to expand access to its cybersecurity model, Mythos.

This is the AI that can autonomously discover zero-day vulnerabilities. In testing, it had a seventy-three percent success rate on expert-level cybersecurity tasks. The government’s official line cites security and compute resource concerns. The unofficial line is pretty clear: we are not ready to let an autonomous cyber-weapon off its leash, no matter how good its creators claim its intentions are. This is the friction point where innovation runs headlong into national security, and for now, security just won. And speaking of friction… Elon Musk’s grand plan to put data centers in orbit to power AI was met with a polite, but firm, dose of reality from OpenAI’s own Sam Altman.

When asked about it, Altman’s response was, and I quote: "Good luck with that. I don’t even think he’s that serious about it." It’s the tech equivalent of "bless your heart." But the real story, the one that connects all of these threads, is happening at a much lower level. It’s happening in the code itself. Evan Armstrong reported this week that sixty-five percent of all new code being written at Snap is now AI-generated. At Google, that number is seventy-five percent. Let that sink in. Not "assisted by AI." Not "checked by AI." Generated. At that rate, Google will hit one hundred percent AI-generated code before the end of 2027.

This isn't some far-off future. This is a tidal wave that has already hit the beach. And it leads directly to the final, and perhaps most significant, data point of the week. Stripe CEO Patrick Collison reported a “parabolic rise” in new firm creation. Not just new websites. New firms. Entirely new businesses being spun up, powered by AI tools like OpenAI’s Codex and Anthropic’s Claude Code. Collison’s quote was stark: “Because of AI, the entire economy is replatforming.” And that word—replatforming—is the key. It’s what happened to Cursor, it’s what’s happening to developers, and it’s what’s about to happen to the internet itself.

So let’s go back to Cursor. Let’s do the deep dive. How do you become the fastest-growing software business in history and decide to cash out? Two billion dollars in revenue in thirteen months. Seventy percent of the Fortune 1000. They were winning, by every metric that has ever mattered in Silicon Valley. They built a better, faster, AI-native code editor that developers loved. They were out-selling Microsoft Copilot in head-to-head enterprise bake-offs. This was not a struggling company. So what was the risk the founders weren't willing to underwrite? It was platform risk. But a new kind of platform risk. Cursor was a reseller.

They built an incredible user experience on top of foundation models from companies like Anthropic and OpenAI. They were the beautiful, high-performance showroom built on top of a rented factory. The problem is, the factory owner can always decide to open their own showroom, right next door, and sell the product at cost. And that’s exactly what happened. Ethan Ding’s analysis on this was brutal and correct. Anthropic didn’t just compete with Cursor on their Claude Code offering. They ran their first-party agent margins close to zero. They effectively sold their models to developers at a price that was five times lower than what a reseller like Cursor could possibly offer.

They weren't just competing. They were deliberately keeping the resale market underwater. This is the new reality. If you build your business on someone else's AI, you are not their partner. You are their customer. And you are, potentially, their future competitor. The platform giveth, and the platform can take it all away by changing a single number in their pricing API. The founders of Cursor looked at the path to one hundred billion dollars and realized it was a tightrope walk over a volcano, and the company that owned the rope was actively trying to set it on fire. So they sold. To XAI. To a company that owns its own models.

They didn't just sell their company; they sold their platform risk. Now, connect that to the second number. Seventy-five percent of new code at Google is AI-generated. The very same models that made Cursor’s business model untenable are now fundamentally changing the job of the people Cursor was selling to. For years, the debate was "will AI take developers' jobs?" That was the wrong question. The right question was "what will a developer's job become?" And now we have the answer. The data from Evan Armstrong is critical here. While this AI code-generation wave has been happening, entry-level developer job postings have dropped sixty-seven percent between 2023 and 2024.

A collapse. But job postings for senior engineering and product management roles? They’re going up. AI is not replacing developers. It’s hollowing out the middle. It’s automating the grunt work, the boilerplate, the stuff you used to learn in your first two years on the job. The new entry-level requirement is the ability to operate the AI. To direct it. To verify its output. The job is shifting from being a writer of code to being an editor and architect of AI-generated code. A senior engineer with a powerful AI assistant is not a ten-percent-better engineer. They are a ten-X engineer. They can do the work of an entire team from five years ago.

This is a massive force multiplier for talent. It also means the ladder that people used to climb—from junior to mid-level to senior—has had several of its rungs sawed off. So you have this perfect storm. The tools developers use, like Cursor, are being absorbed or made obsolete by the platforms. And the work the developers do is being fundamentally transformed by those same platforms. It's a total replatforming of the software industry, from top to bottom. And this brings us to the final, biggest piece of the puzzle. What happens next? This replatforming of software isn't the end of the story. It's the beginning. It’s the necessary groundwork for the next phase: the agent economy.

Patrick Collison’s observation about the "parabolic rise" in new firm creation at Stripe is the canary in the coal mine. These aren't just people spinning up Shopify stores. These are AI agents, built with this new, AI-generated code, being deployed to perform economic tasks. Think about what Craig Hepburn said: "The entire commercial architecture of the internet… was designed around the human as customer. Replace the customer with software and almost none of it survives." For thirty years, the internet has had a glaring hole in its architecture. There was no native payment layer. Every transaction required a human to pull out a credit card, to click a button, to approve a purchase.

It was built for human-scale commerce. What Stripe, and others, are now building is the payment infrastructure for software agents. A system that can handle billions of autonomous agents making trillions of micro-transactions per second. This isn’t about making it easier for you to buy a book on Amazon. This is about one AI agent paying another AI agent a fraction of a cent to perform a specific task, like analyzing a data set or rewriting a paragraph of text. This is the trillion-dollar transaction layer being installed beneath the internet we know. It's happening because the cost of creating and deploying software—the very code that makes up these agents—has just dropped to near zero, thanks to the AI models that are generating seventy-five percent of it.

The story of Cursor’s sixty-billion-dollar exit isn't just about a software company. It's a punctuation mark at the end of an era. The era of building tools for human developers. The new era, the one that’s just beginning, is about something else entirely. We aren't just building tools for the economy anymore. We're building the economy out of tools.