Scaling Git at OpenAI

The discussion centers on OpenAI’s Pando system for handling Git at the company’s scale, amid similar efforts at Cursor, GitHub, and Anthropic. Some commenters praised the post’s engineering detail and straightforward tone, and were curious whether ideas are being shared across companies or might eventually influence everyday development. Others noted that Google and Meta have long used centralized or custom source-control systems, so they viewed the approach as established rather than novel. The claim that Pando was built quickly with Codex impressed some, while others saw the announcement as aimed more at business audiences than engineers. They wanted concrete throughput and concurrency benchmarks, and questioned whether streaming incoming packfiles directly to blob storage provides adequate security.

The broader divide concerned why code volume is rising and whether that growth is valuable. Some commenters described agent-driven commits and parallel coding sessions as a new operational reality that makes Git infrastructure a bottleneck. Others asked what so much code is doing, and warned that rapid AI-generated changes could produce bloated, poorly organized software and a larger attack surface. One argued that a company at OpenAI’s scale should adopt a centralized model rather than preserve Git’s distributed design; another pointed to global caching as a possible alternative. Sarcastic comments mocked the hype around agentic productivity, while supporters saw AI-assisted development as enabling genuinely new software and faster iteration. Several commenters also framed scaling beyond Git as a milestone for large technology companies, though they differed on whether it represents progress or accumulating liability.