OpenAI has published a case study describing how Perplexity, the AI-powered search company, is using its GPT-6 Astra model across several internal operational functions. According to the document, Perplexity now relies on the model to draft communications, make changes to its software, and monitor its production systems.
The most notable detail in the case study concerns the frequency of human oversight. OpenAI states that Perplexity's teams check in on Astra's work considerably less often than they did with earlier generations of models. This reduced need for manual review points to a higher degree of confidence in the model's reliability when handling sensitive technical tasks, such as code changes or production monitoring.
This kind of deployment reflects a broader pattern among companies building products on large language models: the gradual automation of engineering and operations work once handled almost entirely by human teams. For a company like Perplexity, whose core business already depends on generative AI, deeper integration of Astra into internal workflows marks a further step toward end-to-end automation of software operations.
Still, the details shared remain limited, and some caution is warranted about the actual scope of this autonomy. Fewer human check-ins do not necessarily mean oversight has been eliminated, and managing the risk of undetected errors in production environments remains a central concern regardless of how capable the underlying model is claimed to be.