Scaling long-running autonomous coding
Summary
The post analyzes Cursor's experiment of running hundreds of autonomous coding agents to write a large-scale browser project, using planners, sub-planners, workers, and a judge to decide completion. It discusses build attempts, observations of rendering results, and the role of conformance references like WhatWG specs. The piece reflects on the feasibility, challenges, and near-term potential of AI-assisted coding at scale.