Wilson Lin on FastRender: a browser built by parallel agents
3 months ago
- #AI
- #Autonomous Agents
- #Browser Development
- FastRender is a web browser built from scratch using autonomous coding agent swarms.
- The project started as a personal side-project by Wilson Lin to explore frontier models like Claude Opus 4.5, GPT-5.1, and GPT-5.2.
- At its peak, FastRender had approximately 2,000 agents running concurrently, making thousands of commits per hour.
- The browser can currently load pages like GitHub, Wikipedia, and CNN, though JavaScript is disabled due to ongoing engine implementation.
- The project uses a tree structure for agent coordination, with planning agents delegating tasks to worker agents.
- Feedback loops, including specifications and visual comparisons, are crucial for autonomous agent operation.
- Agents selected dependencies like Skia and HarfBuzz, sometimes opting for third-party libraries over from-scratch implementations.
- The system allows intermittent small errors to maintain high throughput, fixing them quickly in subsequent commits.
- FastRender represents over a million lines of Rust code, written in a few weeks, serving as a research example for multi-agent coordination.