Cursor emblem is displayed in opposition to a backdrop of code.
NurPhoto through Getty Pictures
For years, startups obsessed over recruiting the legendary “10x engineer,” a uncommon, hyper-productive particular person able to writing elegant code and fixing architectural puzzles at ten occasions the speed of a median peer.
However a new report from AI coding startup Cursor means that the 10x period is over. It has been changed by one thing extra radical: the software program trade’s high 1% of AI-active builders now write 46 occasions extra AI-generated traces of code per day than the median lively consumer.
“When you zoom out of the code writing course of and take into consideration the normal product improvement lifecycle, it used to appear like a conveyor belt,” stated Tido Carriero, Cursor’s vp of engineering, who leads Cursor’s 200-person engineering and product group. The workflow went like this: A product supervisor drafted a spec, a senior engineer broke it down into structure, and mid-level engineers handcrafted the code. Traditionally, that handcrafting absorbed roughly 80% of a company’s R&D finances.
“That complete manufacturing facility line has completely combusted,” he stated.
Traces of code written is a wildly imperfect metric for productiveness. Extra code can imply extra progress; it could additionally imply a much bigger mess. However Cursor’s report provides one other metric that’s more durable to wave away: the highest 1% of AI customers additionally merge 15x extra “pull requests” per week than the median consumer, which means the code isn’t just being generated. It’s being accepted, finalized and pushed in direction of manufacturing.
So what makes a 46x engineer? In keeping with Carriero, the reply’s not all that totally different from what made a 10x engineer beneficial — but it surely’s being enhanced by new tooling. As a substitute of refactoring repositories of code by hand, the 46x builders instruct a community of brokers to do the grunt work, then assessment, steer and sew the outcomes collectively.
“It’s about problem-solving expertise and the power to attach the dots,” he stated.
Whereas Cursor began as an AI-assisted code editor, the group is now considering far past that. Calling code writing “a little bit of a commoditized scenario,” Carriero stated Cursor is now centered on the broader software program improvement lifecycle. It’s growing instruments to assist product managers resolve what to construct, together with coding infrastructure merchandise for a world through which the large inflow of AI-created pull requests is inflicting Github, the de facto code repository, to expertise extreme reliability points.
It’s hardly stunning that the trade as an entire is producing extra, and quicker, with AI. Coding velocity has doubled over the previous yr, and progress continues to speed up, the report discovered. The common code change is now 2.5 occasions bigger than it was a yr in the past, with “mega” code modifications of greater than 1,000 traces of code modified, changing into more and more frequent.
That increase has introduced with it an related hangover: “vibe slop”— awful AI-generated code that ships rapidly however breaks. Not too long ago, two high AI builders chargeable for the OpenClaw AI agent told the Wall Road Journal that they have been deeply involved concerning the quantity of buggy, harmful code flooding the world. Over the weekend, Uber’s operations chief said it was getting more durable to justify AI budgets as a result of there’s nonetheless no hyperlink between increased AI token utilization and helpful shopper options.
Carriero stated Cursor has made a push to coach enterprise prospects on token effectivity. “We’re making an attempt to assist admins navigate this from their worldview,” he stated. “And likewise simply present them all the analytics on what’s occurring, to allow them to come to their very own conclusions.”
The report comes as Cursor navigates a fearsome marketplace for generative coding. Based in 2022 by 4 associates who met at MIT, the startup was an early frontrunner and darling within the nascent discipline of AI code. However huge developments from Anthropic and its Claude Code instrument have put Cursor and different coding startups on the defensive. After Anthropic launched its Opus 4.5 mannequin late final yr, seen industrywide as a game-changer in AI coding, Cursor internally shifted right into a “struggle time” posture because it sought to guard its territory available in the market. Although Cursor began as a person developer instrument, the corporate is now centered on promoting into the enterprise. Of its $3 billion run charge, 75% of that comes from different companies, the corporate advised Forbes. Its enterprise enterprise grew 3 times within the first quarter, it stated.
Since then, Cursor has entered right into a $60 billion cope with SpaceX to coach its fashions on its Colossus supercomputer, with the expectation that Elon Musk’s house and AI firm acquires it shortly after SpaceX’s IPO later this yr. No merchandise will likely be shut down on account of the acquisition, Carriero stated, noting that Cursor’s fashions will get a lift due to the infrastructure improve.
“We have simply been compute constrained for a very long time, so getting out of that world may be very thrilling,” he stated.