Perplexity says it trusts OpenAI’s GPT‑6 Astra to work on entire systems while employees check in less frequently. That is a bigger responsibility than helping someone write a function or fix a bug, and I would like to know what happens when it gets something wrong.
In an OpenAI customer testimonial, Perplexity cofounder and Chief Strategy Officer Johnny Ho describes using the model for communications, software changes, and production monitoring. “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.”
The clearest example involves testing applications. When Ho has limited time for manual testing, he asks Astra to build a small testing program that generates responses resembling those from an outside service, such as a language model API or connector. That lets it check how an application handles a workflow without needing every external service involved in the test.
That sounds useful, particularly when developers need to test different responses from services they do not control. However, the write-up provides no measurements showing how thoroughly these tests cover potential failures or how often they catch bugs.
Ho makes a broader claim about the company’s confidence in the model. “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.”
Unfortunately, OpenAI does not explain what that trust looks like in practice. We do not learn which changes require approval, what access Astra receives, or how Perplexity recovers from mistakes. The statement also does not establish that the model independently deploys changes to production.
Ho connects better coding capabilities with improvements to Perplexity’s search engine, too. His explanation is that stronger models can write better programs to search the web and internal information, then summarize the results. The page offers no search accuracy figures to demonstrate that improvement.
The business relationship is interesting as well. Perplexity competes with ChatGPT for people looking for answers, yet it is endorsing OpenAI’s technology for its own operations. Competition for users clearly does not prevent the company from finding value in a rival’s models.
Oddly, the page carries a September 14, 2026 publication date despite being accessible before then. That could be a publishing mistake or a page released ahead of schedule, but the date alone does not prove a leak.
Ultimately, this is a customer testimonial published by the company selling the model. Perplexity’s willingness to reduce human oversight is newsworthy, but whether that confidence is justified deserves more than an executive endorsement. Show us the safeguards and the results.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz