Reading Room

OpenAI describes how its storage platform, called Habitat, grew from a small Python library into infrastructure that now handles 22 million requests a second for a billion ChatGPT users. It is a rare inside look at what real scale requires, useful context next time you are reasoning about infrastructure limits.

Perplexity says it now lets the model Astra write its own communications, change its own software, and watch its production systems, checking in far less often than it did with earlier models. It is a concrete data point on how much autonomy a real company is now willing to hand an AI model.