AI-Assisted Coding: The Cursor Team on Rethinking the Code Editor and the Future of Programming

This podcast episode features Lex Fridman in conversation with Michael Truell, Swall Oif, Arvid Nørmark, and Aman Sanger, the founding members of Cursor. The discussion centers on Cursor, a code editor forked from VS Code, and its role in the evolving landscape of AI-assisted programming. The team explains that a code editor, traditionally a "souped-up word processor" for programmers, is undergoing a radical transformation due to advancements in AI. They highlight the shift from simple text editing to a deeply integrated AI environment that anticipates and assists the programmer's next actions, aiming to make the coding process more efficient and enjoyable.

A key distinction made is between Cursor's approach and existing AI tools like GitHub Co-pilot. While Co-pilot was a "killer app" for early LLMs, the Cursor team felt it became "stale" as model capabilities rapidly advanced. Their decision to fork VS Code rather than build an extension stems from the need for deeper control over the editing environment to fully leverage AI's potential. They emphasize that the rapid pace of AI innovation, driven by "scaling laws," demands a programming environment that can quickly integrate new model capabilities and rethink the entire human-AI interaction paradigm.

The conversation delves into Cursor's flagship "tab" feature, which goes beyond simple autocomplete to predict and execute "next actions" – from generating multi-line code and editing existing blocks to navigating across files and even suggesting terminal commands. This is achieved through technical innovations like Mixture of Experts (MoE) models, speculative decoding, and intelligent caching, all designed for low latency and high accuracy. A significant challenge discussed is the "verification problem" for large AI-generated code changes, leading to a focus on ergonomic "diff" interfaces that intelligently guide the human reviewer, highlighting critical information and graying out low-entropy sections.

The broader implications of Cursor's work extend to a future where human-AI collaboration fundamentally reshapes software engineering. The goal is to eliminate "low entropy actions" and allow programmers to focus on higher-level intent, with AI handling the predictable, repetitive tasks. This vision suggests a future where the act of building software becomes radically different, more productive, and even "fun," demanding continuous innovation in both AI models and the user interfaces that harness them. The team believes that the speed of innovation in AI programming will make today's tools obsolete within a year, underscoring the need for agile, research-driven development.

the code editor is largely the place where you build software and today or for a long time that's meant the place where you text edit uh a formal programming language

I think also code edor should just be fun

copile was kind of like the first killer app for LMS

scaling laws papers came out from from open Ai and that was a moment where this looked like clear predictable progress for the field

the Step Up in gbd4 felt like look that really made concrete the theoretical gains that um we had predicted before felt like you could build a lot more just immediately at that point in time

this wasn't just going to be a point solution thing this was going to be all of programming was going to flow through these models it felt like that demanded a different type of programming environment to different type of programming

I think the cursor a year from now will need to make the cursor of today look Obsolete

the gold cursor tab is let's eliminate all the low entropy actions you take inside of the editor when the intent is effectively determined let's just jump you forward in time skip you forward

these diffs are great for small edits uh for large edits or like when it's multiple files or something it's um actually a little bit prohibitive to to review these diffs

you don't want to have to figure out that yourself you want a model to guide you through the thing