Overview
Wilson Lin discusses FastRender, a web browser built from scratch using thousands of coordinated AI agents as a research experiment. The project started as a personal exploration of frontier model capabilities and evolved into an official Cursor research project to study how large swarms of autonomous coding agents can collaborate on complex software development tasks.
The Breakdown
- Autonomous agent coordination - Thousands of AI agents working in parallel built a functional web browser from scratch, demonstrating that agent swarms can tackle extremely complex software projects
- Self-governing development decisions - The agents made independent architectural choices like disabling JavaScript and adding feature flags, showing emergent project management behavior
- Research methodology validation - A browser was chosen as the test project because it’s both extremely ambitious and well-specified, allowing researchers to visually assess progress and agent capabilities
- Scalability exploration - The project graduated from single-agent experiments to multi-agent coordination to push the boundaries of what’s possible with frontier models like Claude Opus 4.5 and GPT-5.1
- Behavioral observation focus - The goal wasn’t to build a production browser but to study how agent harnesses can work at scale and collaborate on real-world software complexity