A recent experiment by the Cursor team, which involved using AI agents to build a fully functioning browser from scratch, is analyzed. The initial flat structure of AI agents led to problems like agents holding locks for too long and avoiding difficult tasks. A hierarchical structure of planners and workers was then implemented, mirroring a corporate hierarchy. Despite writing 1 million lines of code at a cost of millions of dollars, the resulting browser was non-functional and failed to compile. Claims of AI agents writing HTML, CSS, and JavaScript parsers from scratch were also found to be false, as the project imported existing libraries. Gregory Turson, an architect of Servo, criticized the project's design and implementation, describing it as a "tangle of spaghetti" and "AI hallucinated BS." The experiment is ultimately viewed as a marketing stunt.
Sign in to continue reading, translating and more.
Continue