In a recent episode of Lightcone, Varun Mohan, CEO and co-founder of Windsurf, sat down with Y Combinator to share a story that’s become emblematic of the AI era: a total company pivot, executed over a weekend, leading to one of the most widely adopted AI development platforms today. What began as a GPU virtualization startup pivoted—almost overnight—into a fast-moving player in the AI coding revolution. The result? Windsurf: a platform empowering developers, domain experts, and even non-coders to “vibe code” their way into building functional software with unprecedented speed and precision.
This wasn’t just a product shift—it was a philosophical one. Mohan articulates it clearly: “Every insight is a depreciating asset.” That ethos has driven Windsurf’s rapid-fire evolution from infra to interface, from passive autocomplete to active agent, and from a niche dev tool to a future-proof software co-pilot.
In mid-2022, Windsurf (then called Exofunction) was managing over 10,000 GPUs and pulling in millions in revenue. But Varun and his team saw the writing on the wall: Transformers were consolidating the AI stack, and infrastructure was becoming a commodity.
Rather than wait it out, they took decisive action. Over a single weekend, the team pivoted to building Kodium—an AI coding assistant that quickly evolved into Windsurf. Their first models weren’t better than GitHub Copilot, but they were fast, free, and improved relentlessly. Within months, they were outpacing Copilot in core tasks like in-line code editing and complex codebase manipulation.
Fast forward to 2025, and Windsurf isn’t just a dev tool—it’s an AI-native IDE. With millions of users, deep enterprise integrations (think JPMorgan and Dell), and a newly launched full-stack editor forked from VS Code, the company has moved far beyond autocomplete.
Key to their differentiation is Cascade, their in-house AI agent. Cascade doesn’t just assist—it reads, understands, and edits massive codebases in real time. It enables developers to offload version updates, boilerplate code, and even complex deployment workflows. More impressively, non-technical users—sales reps, ops managers, partnership leads—are now building and deploying tools without touching a single line of code.
One of the interview’s most compelling themes is Windsurf’s redefinition of who gets to build software. Mohan argues that we’re witnessing the end of the “developer as gatekeeper” era. “Everyone is going to be a builder,” he says. Whether you're modifying Java in IntelliJ or generating a marketing dashboard through natural language, the barrier to entry is collapsing.
It’s not a vision of no-code, but rather post-code: where AI systems understand your intent, manipulate large-scale codebases, and evolve with you.
Mohan’s philosophy is clear: building a durable company in AI isn’t about nailing a single technical insight—it’s about constantly compounding them. Every breakthrough, every architectural decision (like parsing ASTs or combining RAG with syntax-aware search), builds momentum—but only if you're constantly moving.
In their view, even winning products like GitHub Copilot or Cursor are beatable, because their lead is perishable. Windsurf’s secret? A feedback loop of aggressive experimentation, ruthless evaluation metrics, and irrational optimism paired with uncompromising realism.
The next frontier for Windsurf? Turning coding into something closer to research. With AI taking care of the mundane, Mohan envisions engineering becoming a rapid hypothesis-testing discipline. Developers will prototype, evaluate, and discard faster—freeing them to chase bigger ideas.
They're also looking beyond code generation toward “just-in-time software”—apps generated on demand, deeply personal, and contextually aware. Think agents that spin up custom interfaces based on real-time data from your environment, and retire them when no longer needed.
In this world, “software” becomes ambient. And Windsurf intends to be the default shell in which this post-software software is built.
