For years, startups obsessed over recruiting the mythical "10x engineer," a rare, hyper-productive individual capable of writing elegant code and solving architectural puzzles at ten times the rate of a median peer.

But a new report from AI coding startup Cursor suggests that the 10x era is over. It has been replaced by something more radical: the software industry’s top 1% of AI-active developers now write 46 times more AI-generated lines of code per day than the median active user.

“If you zoom out of the code writing process and think about the traditional product development lifecycle, it used to look like a conveyor belt," said Tido Carriero, Cursor’s vice president of engineering, who leads Cursor’s 200-person engineering and product team. The workflow went like this: A product manager drafted a spec, a senior engineer broke it down into architecture, and mid-level engineers handcrafted the code. Historically, that handcrafting absorbed roughly 80% of an organization’s R&D budget.

"That whole factory line has totally combusted," he said.

Lines of code written is a wildly imperfect metric for productivity. More code can mean more progress; it can also mean a bigger mess. But Cursor’s report offers another metric that is harder to wave away: the top 1% of AI users also merge 15x more “pull requests” per week than the median user, meaning the code is not just being generated. It’s being accepted, finalized and pushed towards production.

So what makes a 46x engineer? According to Carriero, the answer’s not all that different from what made a 10x engineer valuable — but it’s being enhanced by new tooling. Instead of refactoring repositories of code by hand, the 46x developers instruct a network of agents to do the grunt work, then review, steer and stitch the results together.

"It’s about problem-solving skills and the ability to connect the dots," he said.

While Cursor started as an AI-assisted code editor, the team is now thinking far beyond that. Calling code writing “a bit of a commoditized situation,” Carriero said Cursor is now focused on the broader software development lifecycle. It is developing tools to help product managers decide what to build, along with coding infrastructure products for a world in which the massive influx of AI-created pull requests is causing Github, the de facto code repository, to experience severe reliability issues.

It’s hardly surprising that the industry as a whole is producing more, and faster, with AI. Coding speed has doubled over the past year, and growth continues to accelerate, the report found. The average code change is now 2.5 times larger than it was a year ago, with “mega” code changes of more than 1,000 lines of code changed, becoming increasingly common.

That boom has brought with it an associated hangover: “vibe slop”— lousy AI-generated code that ships quickly but breaks. Recently, two top AI developers responsible for the OpenClaw AI agent told the Wall Street Journal that they were deeply concerned about the amount of buggy, dangerous code flooding the world. Over the weekend, Uber’s operations chief said it was getting harder to justify AI budgets because there is still no link between higher AI token usage and useful consumer features.

Carriero said Cursor has made a push to educate enterprise customers on token efficiency. “We're trying to help admins navigate this from their worldview,” he said. “And also just show them all of the analytics on what's happening, so they can come to their own conclusions.”

The report comes as Cursor navigates a fearsome market for generative coding. Founded in 2022 by four friends who met at MIT, the startup was an early frontrunner and darling in the nascent field of AI code. But big advancements from Anthropic and its Claude Code tool have put Cursor and other coding startups on the defensive. After Anthropic released its Opus 4.5 model late last year, seen industrywide as a game-changer in AI coding, Cursor internally shifted into a “war time” posture as it sought to protect its territory in the market. Though Cursor started as an individual developer tool, the company is now focused on selling into the enterprise. Of its $3 billion run rate, 75% of that comes from other businesses, the company told Forbes. Its enterprise business grew three times in the first quarter, it said.

Since then, Cursor has entered into a $60 billion deal with SpaceX to train its models on its Colossus supercomputer, with the expectation that Elon Musk’s space and AI company acquires it shortly after SpaceX’s IPO later this year. No products will be shut down as a result of the acquisition, Carriero said, noting that Cursor’s models will get a boost because of the infrastructure upgrade.

“We've just been compute constrained for a long time, so getting out of that world is very exciting,” he said.