According to Business Insider, Cursor’s engineering head Jason Ginsberg revealed on the “LangChain” podcast that some of the AI coding company’s most important features started as informal side projects. He built the now-launched “Debug Mode” over the Thanksgiving holiday simply because he wanted it, while the company’s defining agent feature was prototyped by a single skeptical engineer. Ginsberg said their key metric for shipping is internal adoption by their own tiny team, which numbered about 20 people at the start of 2025. He attributed the small size to an extremely high hiring bar and a slow process, which creates a talent-dense structure that allows Cursor to operate with minimal formal process and move quickly.
The Power of Bottom-Up Tinkering
Here’s the thing that’s so fascinating about this. In an industry obsessed with massive roadmaps and top-down strategy from product managers, Cursor is basically running on hacker energy. An engineer wants a debugging tool over a holiday break? He builds it. A colleague thinks an agent could work despite skepticism? He prototypes it “super quickly.” And if the team starts using it internally, that’s the green light to ship. It’s a remarkably pure, almost old-school Silicon Valley approach. Instead of endless alignment docs and meetings, they resolve disagreements through code. That’s a powerful filter. It means every feature that makes it out has already proven its utility in the most demanding environment: the daily workflow of the brilliant people building it.
The Small-Team Doctrine Goes Mainstream
But Cursor isn’t an outlier anymore. It’s part of a major trend. Ginsberg’s comments echo a philosophy that’s become gospel across the AI frontier. Look at Meta: Mark Zuckerberg recently said he’s “convinced” small, talent-dense teams are optimal for frontier research, a belief embodied in his superintelligence AI unit that’s a tiny sliver of Meta’s 70,000+ employees. Sam Altman predicted billion-dollar valuations for 10-person companies. There’s a clear consensus forming. The complexity of modern AI might suggest you need armies of people, but the opposite seems true. Speed, clarity of vision, and minimal communication overhead are the new superpowers. A small, elite team can pivot on a dime, and as Cursor shows, its best ideas can come from anywhere.
When Process Is the Enemy
So what’s the real takeaway? It’s a stark warning against premature bureaucratization. For a company like Cursor, adding formal process too early would likely kill the very magic that lets them innovate. The debate-by-code model only works when everyone is exceptional and trust is high. Scale that to 200 engineers, and it collapses into chaos. The challenge for every fast-growing startup, especially in AI, is knowing when to start implementing guardrails without stifling the bottom-up creativity that got them there. Cursor’s story suggests you should delay that moment for as long as humanly possible. Let the side projects flourish. Because your next flagship feature is probably being built by an engineer who’s just scratching their own itch.
