Cursor Just Killed Every AI Coding Tool

Quick Overview

Cursor has significantly advanced its AI coding capabilities, particularly with a new feature that allows it to run development tasks for over 10 hours straight, autonomously generate code, and then create and show a demonstration video of the new feature working, effectively challenging existing AI coding tools like Google's Anti-Gravity.

Key Points: Cursor can run development tasks for over 10 hours straight, which the speaker notes is a key differentiator. The new feature allows Cursor to not only create code but also demo the code by recording a video showing the feature actually working, utilizing Remotion for video creation. The long-running capability is likened to the speaker's concept of 'Aries,' an autonomous running software engineer, capable of grinding for 3, 5, or 10 hours. Setup involves integrating Cursor into GitHub, choosing a high-level model like Codeex 5.3 high or Opus 4.6 high, and enabling the 'long running' mode. Effective prompting involves using terms like 'open source UI that looks like XYZ' and concluding the prompt with instructions like 'then do a demo of each feature proving it works.' Cursor builds a sandboxed environment that installs dependencies and securely loads environmental variables like API keys, allowing long runs on a pre-built application structure. The speaker demonstrated this by having Cursor code an entire media website, techniff.com, generating 10,000 lines of new code while the speaker was cooking steak.

Context: The video focuses on a major new update to the AI-powered code editor, Cursor, which the speaker believes has 'killed every single AI coding tool in the market.' The context centers around the demonstration of Cursor's enhanced autonomy, specifically its ability to execute long-running code generation tasks and automatically create proof-of-concept demo videos for the resulting features, a capability the speaker was surprised Google's Anti-Gravity IDE had not implemented first.

Detailed Analysis

Cursor's latest update introduces game-changing functionality that enables long-running AI agent execution, potentially for over 10 hours, which results in fully built application features accompanied by automated demo videos. The speaker highlights that this combination surpasses expectations, noting that the AI can code a feature, encounter an error, self-correct the code file, and then generate a video demonstration using Remotion, an API for creating videos, all without continuous human oversight. To utilize this, users integrate Cursor with GitHub, select a powerful reasoning model (like Opus 4.6i), set the desired run time, and employ specific prompting techniques that leverage open-source frameworks for UI context and explicitly request a demo at the end. Furthermore, Cursor establishes a persistent, sandboxed environment associated with the Git repository, pre-loading dependencies and sensitive API keys, which acts like a pre-built house where the AI agent only needs to add new extensions rather than rebuild from the foundation every time. The speaker validated this by showing techniff.com, a media website comprising 10,000 lines of new code generated by Cursor during a long run while the speaker was occupied, showcasing the profound impact on development speed and review processes, as teammates can receive video proof of functionality instead of running the code locally.