Public transcript

Greg Brockman on Astra and the Future of OpenAI

By: TBPN
Length: 25 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

OpenAI co-founder and president Greg Brockman discusses the model's solution to the Navier-Stokes millennium problem, the launch of Astra and its computer use capabilities, and the shift from chat to agentic use cases. He covers ChatGPT's health strategy across consumer, clinician, and enterprise pillars, the lineage from Operator to Astra's computer use, and OpenAI's renewed startup-like focus. He argues AI should fade low-level details into the background and empower people to delegate work according to their goals.

Key points

OpenAI announced its model solved the Navier-Stokes problem, one of the seven millennium problems, by finding a counter example showing the equations break down under certain circumstances.
Astra represents a new threshold of computer use, with users creating 3D models in Blender, mapping physical locations, and designing physical parts.
ChatGPT has well over a billion users every week, including 300 million people per week with health queries.
Brockman says the future is unifying chat and agentic modes so people can delegate work proactively without managing context windows or model selection.
Health has three pillars: consumer queries, a clinician-tuned product with direct citations to medical literature, and enterprise sales to hospitals including an Epic integration.
Operator was below threshold in model capabilities, but the team's long-term investment in computer use delivered with Astra.
Brockman contrasts esoteric use cases with concrete ones, saying people don't generalize capability from one demo to their own problems.
He describes OpenAI switching back into startup mode this year, with the whole team rowing in the same direction.

Questions it answers

00:00Navier-Stokes solution

What did OpenAI announce about the Navier-Stokes problem and why does it matter?

The model solved Navier-Stokes, one of the seven millennium problems, by finding a counter example showing the equations break down under certain circumstances. Brockman says the proof is elegant and that it shows models can generate new knowledge for humanity, making ambitions like curing diseases more real.

The result is new knowledge for humanity with applications in fluid dynamics, ocean currents, and turbulence.
The bigger significance is model capability: solving problems otherwise outside human reach.

03:21Astra launch reaction

How has the community responded to Astra?

Brockman says he was blown away by the reaction, with people creating 3D models in Blender, mapping physical locations, and designing manufacturable physical parts like a shower-drain hair catcher. He says Astra reached a new threshold of computer use and empowers individuals in everyday problems.

Astra works with different applications in a way not previously possible.
The focus is empowering individuals, not just grand challenges.

05:23Chat and agents merging

How do chat and agentic use cases come together?

Chat is alive and well with over a billion weekly users, while agentic adoption has grown almost vertically since ChatGPT Work launched. Brockman says these two modes are being unified into a new form factor where you delegate work to something proactive and persistent, without managing context windows or thinking strength.

The promised future was never a low-level language model with manual settings.
Computers should operate according to your goals, developed safely.

09:21Discovery and marketing

How does OpenAI solve the discovery problem for new capabilities?

Brockman says the model should proactively tell users what else it can do if they add a connector or authorize it. He also learned that concrete use cases, like the ChatGPT Images 2.5 launch video showing a candle holder sketch becoming a physical object, work better than esoteric ones because people don't generalize capability.

Over a billion users weekly is a unique advantage for re-engaging people.
300 million people use ChatGPT for health queries every week.

14:53Image models

Where do image models go as a category?

Brockman says new capability thresholds unlock fundamentally new applications, and images need to be above a quality threshold for professional use. Precise edit control, speed, creativity, and good interplay unlock use cases like slide creation and website building, which he sees as unique to OpenAI's package of images, voice, and coding.

Below the quality threshold, output is a cool concept but unusable.
Images, voice, and coding are viewed as one package that empowers creation.

16:51ChatGPT Health

What is OpenAI's health strategy?

Three pillars: consumer health queries (300 million weekly), a clinician-tuned product with direct citations to medical literature, and enterprise sales to hospitals including an Epic integration. Brockman says synergies across them could transform healthcare through record sharing across providers and better clinical trial enrollment.

Patients today carry records between specialists and are on the hook for decisions.
Brockman's wife's five-year diagnosis journey shows how AI could connect symptoms specialists miss.

21:19Operator to Astra

Was Operator a failure, and how did computer use get solved?

Brockman frames Operator as a timing and iterative deployment story: it was below threshold, slow, and painful, but the team made a long-term investment and burned down a long list of issues to deliver computer use with Astra. He says humans can already use everything on a computer, so an AI that operates that way can help across everything.

Operator was a cloud-based computer use system that some people got value from.
Astra is a true milestone from long-term focus and not giving up.

23:54Startup focus

How has OpenAI changed how it ships this year?

The hosts describe OpenAI switching from operating like a big company with a portfolio of bets back into startup mode with intense focus. Brockman agrees it took real effort from many people and says the company is laser focused on its mission.

Focus and the whole team rowing in the same direction drove momentum and growth.

Notes

Navier-Stokes breakthrough

OpenAI announced its model solved the Navier-Stokes problem, one of the seven millennium problems, by finding a counter example showing the equations have a singularity or break down under certain circumstances.
Brockman calls the proof elegant and beautiful, with applications in fluid dynamics, ocean currents, air flow around aircraft, and turbulence.
The deeper significance is that models can now generate new knowledge for humanity and accelerate scientific research, making ambitions like curing diseases and developing new medicines more real.

Astra and computer use

Astra reached a new threshold of computer use, and the community reaction blew Brockman away: 3D creations in Blender, mapping physical locations, and designing physical parts like a shower-drain hair catcher that could actually be manufactured.
The vision is for tools like Blender to fade into the background so humans set goals and objectives while still providing oversight and feeling accountable for outcomes.
This year saw a shift from pure chat to agentic use cases, with a near vertical wall of agentic adoption since ChatGPT Work launched, though chat is alive and well with over a billion weekly users.

Discovery and marketing

Brockman sees a first-mover disadvantage: people tried ChatGPT early, formed an impression, and capabilities have since changed.
The solution is self-knowledge: the model should proactively suggest connectors, authorizations, or better ways to ask so users discover capabilities.
He learned that highlighting esoteric use cases doesn't help people generalize; concrete use cases like the ChatGPT Images 2.5 launch video, showing a candle holder sketch becoming a physical object, work better.

Health

Three pillars: consumer (300 million weekly health queries), clinician (tuned with direct citations to medical literature), and enterprise (selling to hospitals, with an Epic integration).
Synergies across pillars could enable sharing of medical records across providers and better clinical trial enrollment, a huge bottleneck to drug development.
Brockman shares his wife's five-year journey across specialists before an allergist connected her symptoms to a genetic condition, illustrating how AI could spot patterns humans miss.

Operator to Astra

Operator was a cloud-based computer use system that was slow, inaccurate, and below threshold, but OpenAI's long-term investment and grinding through a long list of issues delivered computer use with Astra.
He frames the year as OpenAI switching back into startup mode: focus, the whole team rowing in the same direction.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.