Wilson Lin, an engineer at Cursor, developed FastRender, a web browser constructed using thousands of parallel agents. This innovative project began as a personal side-project to explore the capabilities of frontier models like Claude Opus 4.5 and GPT-5.1. FastRender's architecture allows it to load and render pages like Github, Wikipedia, and CNN, showcasing its progress despite JavaScript not yet being fully functional.

The project transitioned from a side-project to a core research initiative at Cursor when Lin observed promising results from single agents. This shift allowed for increased resources and the exploration of multiple agents working together. FastRender serves as a testbed for observing agent behaviors at scale, with plans to tackle complex web technologies like JavaScript, WebAssembly, and WebGPU in the future.

One of the most impressive aspects of FastRender is its infrastructure, which can run approximately 2,000 agents concurrently. These agents are arranged in a tree structure, with planning agents delegating tasks to worker agents. The system can make thousands of commits per hour, demonstrating the efficiency of parallel agent work. Despite the high number of agents, the project has managed to minimize merge conflicts by effectively dividing tasks.

FastRender's design includes feedback loops, using specifications like csswg-drafts and tc39-ecma262 to guide agent decisions. The project also leverages the strictness of the Rust compiler for verification. This approach allows agents to work autonomously for extended periods, making FastRender a significant advancement in the field of AI-driven software development.