870: Windsurf forked VS Code to compete with Cursor. Talking the future of AI + Coding

Episode Summary

Executive Summary: The episode centers on Windsurf’s vision for AI-native coding: not replacing developers, but making them dramatically more productive through a new editor UX, deeper context retrieval, and agentic workflows. Kevin and Varun argue that the interface layer—not just the model—is the bottleneck, and that the future of software development is more iterative, collaborative, and accessible to more people while still requiring skilled engineers.

Main Topics: Why Windsurf forked VS Code: Varun explains that Codium began as a free autocomplete extension across many IDEs, but Windsurf was created to unlock a better native experience for AI-driven development that VS Code’s extension model couldn’t fully support. User experience as the key differentiator: The guests emphasize that AI coding tools need interfaces designed around human-AI collaboration, including features like Cascade and SuperComplete, rather than bolting AI onto traditional editors. Agentic context retrieval and workflows: They describe Windsurf as retrieving context like a human developer: reading docs, grepping the codebase, checking configs, and taking actions step-by-step with tools like terminal commands and file edits. The future of developers and software work: The conversation argues that AI will not eliminate developers, but will raise productivity, broaden who can build software, and increase demand for more software overall. Rules, memories, and personalization: Windsurf uses memories and intended rules to remember style, language, and project-specific conventions, with plans for AI-generated dynamic memories. Large codebases, version shifts, and staying current: They discuss how agentic systems can handle large repositories, adapt to changing libraries like React and Svelte, and use web search, memory, and trajectory tracking to reduce stale suggestions. Pricing, usage limits, and enterprise strategy: The pricing discussion explains why Windsurf remains relatively low-cost for individuals but leans on enterprise adoption to sustain heavy model usage and infrastructure costs.

Key Arguments: The editor UX matters as much as the model; AI coding tools need native interfaces that match the capabilities of modern LLMs. Windsurf’s agentic approach retrieves context the way a human developer would, using docs, examples, configs, search, and codebase exploration. AI coding tools should reduce boilerplate prompting by remembering project conventions and user preferences through memories. Developers are not being replaced; instead, they become far more productive, and companies will build more software because ROI improves. The market for simple, sellable utility apps may shrink, but personal software creation and internal business automation will expand significantly. Large codebases are manageable because the system can iteratively inspect, synthesize, and act rather than depending on one giant context dump. Frequent version changes require freshness mechanisms such as web retrieval, memories, and trajectory awareness so the AI stays aligned with the current stack. Enterprise usage is the economic engine that supports low individual pricing because power users consume substantial model credits and trigger high provider costs.

Data Points: Time since Windsurf launch: last two months - Varun says Windsurf was launched in the last two months after Codium’s extension-era product. Company age: close to four years ago - Varun says the company started nearly four years ago as a GPU virtualization company. Supported IDEs: VS Code, IntelliJ, PyCharm, Eclipse, Visual Studio, and more - Codium was built as an extension for many major IDEs. Pricing launch point: $10/month - The hosts reference Windsurf’s initial paid tier at launch. Current price discussed: $15/month - The episode discusses the updated monthly price after heavy usage by power users. Product usage increase: multiple factors - Varun says native implementation in Windsurf increased acceptance rate by multiple factors versus the VS Code extension. Internal code contribution: about 50% - Kevin says roughly 50% of the code in their own work is written by Windsurf or guided by it. Enterprise deployment scale: thousands to 10,000+ developers - They cite JPMorgan Chase and Dell using the product internally at large scale. Usage limits hit: highest of highest levels - They say they have hit rate limits at the top levels across providers due to heavy usage. Context algorithm complexity: quadratic - They note attention-based context use is quadratic, making very large token counts expensive. Failure rate target: over 50% - They say the company expects to fail on more than half of the things it tries as part of ambitious experimentation. New features killed internally: five - Kevin notes there is a 'graveyard' of about five features killed during experimentation.

Pivotal Quotes: "if we are not wrong a lot, that probably means we're not trying enough ambitious things" — Kevin: On the company’s experimentation philosophy and willingness to fail frequently. "if you're not using a tool like WinSurf, I think you are just like significantly less productive as a developer" — Varun: On the near-future impact of AI coding tools on developer productivity. "the dream state is actually the human and the AI are operating on the same version of where they think the software is going" — Varun: On the ideal human-AI collaboration model inside the editor.

Implications: AI coding is shifting from chat-style prompt dumping to agentic, interface-led collaboration. Developers who adopt these tools will likely outpace those who don’t, while teams and enterprises will build more software with less friction and more personalization.