In Brief
OpenAI says roughly 10,000 coordinated AI agents produced a proposed solution to a longstanding mathematics problem in 88 hours.
The announcement came amid an authorship dispute and raised a question that reaches well beyond mathematics: What happens to human expertise when AI produces answers faster than people can understand them?
What Happened
OpenAI released a proposed solution to the Navier-Stokes Millennium Prize Problem, a question about whether equations describing fluid motion can break down. Those equations support work in aircraft design, weather forecasting, and blood flow.
The company released a written proof and a version in Lean, software that checks mathematical arguments. It says they used an unreleased model significantly more capable than GPT-6 Astra, with thousands of AI agents working together, to reach the proposed solution. OpenAI says it does not intend to claim the Millennium Prize.
The announcement also came amid a dispute with NYU mathematician Tristan Buckmaster and Anthropic employee Levent Alpöge, who had pursued related work as a personal collaboration. Buckmaster says OpenAI researcher Sebastien Bubeck proposed a paper without Alpöge as an author because of his employer. Bubeck says the proposal concerned rewriting OpenAI's work, not removing Alpöge from his own research, and apologized for his choice of words.
On Episode 239 of The Artificial Intelligence Show, SmarterX founder and CEO Paul Roetzer examined the capability gains and the human consequences .
The Key Numbers
~10,000 - Concurrent AI agents from OpenAI that worked to solve the math problem
88 - Hours it took to reach the proposed solution, according to OpenAI
17 - Additional hours needed for Lean formalization and verification
130 billion - Output tokens, or chunks of text, for the Navier-Stokes effort
Why Faster Answers Challenge Human Expertise
The speed changes expectations. Roetzer compared the announcement with the recent past when models used to struggle with basic tasks such as counting letters. "So the speed of the improvement of the models to me is maybe the biggest story of all here," he says. He sees the result as a reason to reconsider what AI might eventually do in fields such as law and finance. That is a question about future capability, not evidence that those professions can already be automated.
Producing an answer can leave understanding behind. Roetzer discussed a declaration from leading mathematicians arguing that solving problems serves a larger purpose: developing insight and the ability to ask new questions. Their concern extends to other intellectual professions where the process of doing the work develops expertise.
"We're all going to come to this point where it's like, 'Damn, it's better than me at the thing I spent my life doing. Now what?'"
— Paul Roetzer, founder and CEO of SmarterX on Episode 239 of The Artificial Intellgence Show
Trust in the tools matters, too. Buckmaster questioned whether drafts entered into Codex, OpenAI's coding tool, influenced OpenAI's result. OpenAI says his prompts could not have influenced the system, including through training. Roetzer raised the broader concern for companies uploading intellectual property: "Is it really walled off?" The dispute does not establish that OpenAI used those drafts.
SmarterX Take
Business leaders should separate the speed of an AI result from their ability to evaluate and use it. OpenAI's math effort included a dedicated verification stage. In everyday knowledge work, teams still need to decide what a sound answer looks like and who has the expertise to vet it.
There is also a human responsibility. Roetzer pushed back on dismissing mathematicians' concerns as resistance to progress. If AI changes the work someone has spent a career mastering, the response needs to include how that person keeps learning, contributing, and finding purpose.
What to Watch
Human understanding of the result remains central. Follow how mathematicians explain the proof and what they learn from it. A checked argument and a useful body of shared knowledge serve related purposes, but developing the latter takes human work.
Broader capability and future solutions remain an open question. Roetzer wondered whether comparable reasoning could accelerate other scientific challenges, while acknowledging that mathematics offers a provable target. This announcement alone does not establish a timetable for solving diseases or transforming entire industries.
Are Professionals Ready to Work With AI Agents?
Using AI agents is a requested training topic for 51% of respondents to the 2026 State of AI for Business Report. OpenAI's research shows why professionals want to understand this capability. Teams need the knowledge to decide where agents fit, how to evaluate their work, and what human expertise must remain part of the process.
Based on more than 2,100 responses from professionals across roles, functions, and industries, the report examines AI adoption, training, and organizational readiness. It offers a benchmark for identifying the education your people need as AI capabilities advance. Read the full report →
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.