New OpenAI Product Fails Spectacularly During Live On-Stage Demo

Would any live product demo be complete without a few technical difficulties?

At its annual DevDay conference in San Francisco on Tuesday, OpenAI unveiled its new AI agent called Dots — which, like Meta’s Muse, comes with a cute mascot. Powered by its GPT-6 Astra model, the agents are intended to be “always-on,” doing stuff for you in the background constantly.

Dots didn’t seem quite switched on, though, when OpenAI product team member Holly Li tried to show off its capabilities.

Li calls the AI on her phone like you might a human secretary, and a few rings later the agent, which she’s nicknamed “Dottie,” picks up.

“Can you catch me up on just user testing from last night?” Li asks.

“Checking that now,” says Dots.

But the AI never replies. An agonizing ten seconds go by without an answer, the crowd suspended in an awkward silence. Li, playing it off, asks the AI how it’s going, and it curtly replies that it’s “still checking.”

“I guess Dottie’s having a slow morning,” Li jokes.

When Dot’s having quicker starts to its day, OpenAI says it’ll be able to handle tasks in all sorts of professional settings. Per a blog post about the agent, an app developer could use it to monitor user feedback and even create bugfixes on its own. A scientist could have their personal Dot learn their research and run new analyses as fresh data comes in — or even automatically plug the new figures into a paper, in a move that we can’t imagine backfiring spectacularly.