Cursor's recent blog post describes an ambitious experiment where its autonomous coding agents spent nearly a week writing over a million lines of code for a new web browser. The company framed this as a test of scaling agentic coding for large projects typically requiring human teams. However, the post provides no working demo or executable code.

The key issue is reproducibility. While the company claimed 'meaningful progress' and minimal agent conflicts, the public GitHub repository fails to compile. Independent build attempts reveal dozens of compiler errors, and the codebase shows signs of 'AI slop'—output without functional intent. Cursor never explicitly claimed the browser works, but the presentation strongly implies a successful prototype.

This highlights a growing tension in AI development. Companies often showcase impressive token output and scale, but the real test is whether the resulting code is functional and maintainable. For autonomous systems, producing compilable, working software remains a fundamental hurdle that this experiment appears to have missed entirely.