Perplexity Now Trusts GPT-6 Astra to Run Whole Production Systems!

Hey everyone, it's me! Today I found a story about just how much work companies are starting to hand off to AI, and I had to share it.

OpenAI News openai.com

What was announced?

OpenAI's News blog featured a story about how the search company Perplexity uses its new GPT-6 Astra model to draft communications, modify software, and even monitor production systems. Johnny Ho, Perplexity's Cofounder and Chief Strategy Officer, said: "We can have the model craft communications, edit real-world systems, and monitor our production software in a way that previous generations were not able to."

Why it matters

With earlier models, handing off critical work like code changes or production monitoring meant humans had to check in constantly. What stands out here is that Perplexity now trusts GPT-6 Astra enough to check in far less often. As Johnny Ho put it: "We're actually able to trust it with full end-to-end systems and check in on it much less frequently than previous generations of models."

What changes

As an AI-driven answer engine, Perplexity cares most about search precision and how well it processes information. According to Johnny Ho, better code-writing ability in the model directly improves Perplexity's own search quality — better querying of the web and internal data, with more concise summaries. It's a nice example of coding-model improvements feeding back into a completely different product experience.

Dive Deep

One area highlighted as especially useful is automated testing. GPT-6 Astra can generate realistic mock responses that simulate external services, like other language model APIs or connectors, which lets Perplexity test its applications end-to-end without manual intervention. Offloading even the work of faking realistic test data to the model is a small thing, but it can save a lot of developer time.

Wrap-up

• Perplexity uses GPT-6 Astra for drafting communications, modifying code, and monitoring production systems
• It now checks in less often than with earlier models, trusting the model with full end-to-end systems
• Better code-writing ability in the model is also improving Perplexity's own search precision and summarization
• The model also generates mock responses for external services, helping automate testing

If you're curious how far AI can go in running the operational side of a product, this is a useful data point.