OpenAI Says Its AI Solved a $1 Million Math Problem. The People Behind It Disagree.

A machine didn’t just outsmart mathematicians here. Let’s talk about what headlines like this are built to make us believe, widening the gap between AI hype and reality, and what it actually takes to navigate today’s environment.

By Drew Mathias · Heartfelt Strategy · September 12, 2026

My husband and I were talking about this over morning coffee the day OpenAI made their announcement. He works in tech, at a company where AI is deeply embedded in everything they do, and this week's headline completely gripped him.

On September 8, *OpenAI announced that a swarm of its AI agents had cracked the Navier-Stokes existence and smoothness question, one of seven Millennium Prize Problems that have carried a $1 million bounty since 2000

While he was sitting with some big (and, if I'm being honest, unsettling) questions about what it means when a machine does this—If machines can do this now, where does this go?!—I found myself asking: Wait. What, exactly, did it take to get here? How could a headline credit a machine without mentioning the people whose work built the path?!

And here’s what irked us the most... headlines love to act like these AI models just woke up, stretched, and decided to solve an impossible math problem all on their own (as if there weren't thousands of human decisions behind the curtain!). AI doesn't work unprompted or magically in a vacuum. Real human beings had to frame the question, design the research plan, set up the agent harness, and guide the entire process. Pretending the machine did this single-handedly completely ignores all the actual work people had to put in to make it happen.

What Actually Happened

The 3D Navier-Stokes equations describe fluid motion, and mathematicians have long questioned whether they can produce infinite speeds in finite time (a "singularity"). OpenAI deployed 10,000 AI agents, *finding a breakdown in 88 hours and spent 17 more formalizing the proof in Lean. While a formally verified proof for a 90-year-old problem is impressive, it didn't happen in isolation

Show Your Work

Mathematicians *Diego Córdoba and Luis Martínez-Zoroa first opened a subtle entry point into the problem. NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge picked up that exact thread, spending months closing in on a proof

Then OpenAI stepped in. Per *Buckmaster’s account, OpenAI heard Anthropic was close, started work just a week prior, and deployed an unreleased model using an estimated $22.5 million in compute. In Live Science, Buckmaster claims he was offered joint credit only if he dropped Alpöge’s name—and that an OpenAI researcher asked him: "Why would you ruin your career?

While OpenAI called the allegations false and I don't know how that corporate dispute ultimately resolves, what I keep asking myself is this: Does an agent swarm solve this problem without Córdoba and Martínez-Zoroa opening the route, or without Buckmaster and Alpöge proving it was worth walking down?

I don't think it does. "AI solved a Millennium Prize problem" is a lot easier to write when you leave out the real human effort that happened over many years.

Two Different Realities

Ten thousand agents and tens of millions of dollars in compute, built on years of human mathematical intuition, that’s not a cynical footnote. That IS the story.

Almost nobody has 10,000 agents or $20 million to spare. Most leaders face far more practical questions such as: Can employees safely use LLMs? Why did we buy unused software? Why are only a few people experimenting while the rest wait for a plan?

When headlines collapse those two realities into one, leaders can end up paralyzed because they either become dazzled by capabilities their teams aren't prepared for or terrified that the distance between where they are and what’s "possible" just became an insurmountable chasm.

What Should We Do About It

Sitting on your hands isn't a strategy (of course it isn't ;)), and neither is waiting for the frontier to sort itself out (though if I'm being honest, most of us do some version of both).

If there's one thing we’ve become convinced of in the work we do, it's that organizational change moves nowhere near as fast as headlines suggest. Take *Microsoft's recent 2026 Work Trend Index Annual Report, which found that organizational factors including: culture, manager support, and role structure, account for twice the AI impact of individual effort alone (67% vs. 32%)

Data Source: Microsoft 2026 Work Trend Index Report, page 16.

The report calls out a brilliant finding it terms the *Transformation Paradox, where 65% of AI users fear falling behind if they don't adapt quickly, yet 45% say it actually feels safer to stick to their current goals than to try and redesign their work. And a meager 13% feel they are rewarded for reinvention even if the results aren't immediately met

The conclusion is that workers are ready, but their organizations aren't. Employees want to reinvent how they work, but the systems around them—metrics, incentives, and norms—continue to reinforce the old way. (How many of us feel that way?!).

But when leadership actually steps up to redesign those systems, everything changes. The data shows that when managers modeled using AI themselves and created psychological safety around experimentation, their teams reported up to a 20-point jump in AI readiness and value, and were 1.4x more likely to become high-frequency users of agentic AI.

The organizations we've watched make real progress didn't look like frontier labs. They just had someone senior enough who decided this mattered enough to change how work actually got done:

• Education: One county education entity turned a process with nearly 90 manual steps into a single automated workflow, freeing up nearly 60 hours a month per person.

• Consulting: A specialized consultancy found workflow redesign enhanced team capacity by 1.3 FTEs a year—returning nearly $400K in value to their organization’s bottom line.

• Agency: A communications agency built a media monitoring and sentiment tool of their own rather than buying an off-the-shelf version that would have cost tens of thousands of dollars a year.

  
  

But none of those hours or money saved will matter if they are not put to good use!! Getting time back is the low-hanging fruit. Deciding what human work, strategic thinking, or relationship-building you're going to grow with it is the harder, much more interesting problem. These are the problems that people need to solve. (As the report explicitly notes, the most advanced professionals using AI absolutely refuse to outsource their critical thinking).

My business partner, Erin, often uses this example...think about how factories transitioned from steam power to electricity a century ago. Early adopters simply bolted electric motors onto old steam-shaft layouts and wondered why productivity barely budged! It wasn't until factory managers completely rethought the layout of the floor that the real breakthrough happened.

Right now, history is repeating itself. Most organizations are spending their energy bolting AI tools onto broken, manual workflows and expecting a miracle. The right thing to be focused on is this: what can we do now that we couldn’t do before?

As Harvard Business School's Dr. Karim Lakhani points out in the *Microsoft report's foreword: "The central task of leadership, therefore, is shifting from deploying technology to leading and enabling their teams to redesign work and processes.

The Questions I'd Be Asking

If you're trying to figure out where your organization actually stands today, here is what I'd be asking:

• Leadership: Do you have a clear decision-maker driving where AI can genuinely make work better?

• Costs & Incentives: Who owns the budget, and is that the same person pushing adoption?

• Metrics: Are you measuring whether anything actually got better, or just how many people use it?

• Expertise: Whose domain knowledge are your tools leaning on, and are those people being credited, encouraged, and supported?

• Intentionality: If you freed up time for your team tomorrow... what would you actually have them do with it?

• The Bottleneck: Has "figure out our AI strategy" been sitting on your list for months (or years)? What's really stopping you? Is it the technology, or is it that nobody's job or accountability actually depends on getting this done yet?

  
  

Being on the "AI Frontier" Isn't the Strategy

Most of us don't need OpenAI's compute budget to build something meaningful, and racing them was never the point. The organizations pulling ahead over the next few years won't be the ones with the most agents or the newest model. They'll be the ones willing to rethink how work actually gets done from the top-down and the bottom-up, on their own terms. Being first on the frontier isn't the only path to success. Look at Apple: they announced a foldable phone nearly a decade after Samsung, yet framed it as the iPhone's most significant innovation in years. Addressing a real human need, timing, and execution often matter far more than initial novelty.

This is the exact gap we focus on at *Heartfelt Strategy: acting as a friendly outside forcing function to sit with your people, find the real friction, redesign roles, and make accountability for change actually stick using the newest tools at your disposal

The frontier will keep moving no matter what any of us do. The real question is whether you have a plan, and the right partners, to move with it.

Ready to figure out what your organization can do now that you couldn't do before Reach out to Heartfelt Strategy to start redesigning how your team works with AI.*

P.S. On practicing what we preach: This post started as a morning coffee debate with my husband, was co-edited and shaped alongside both him and my business partner, and went through a final review and edit with our AI tool partners. Using human connection and shared expertise to set the intent, and using AI to help refine and execute, is exactly how we believe real work gets done.

Editor's note: Both the Millennium Prize proof and the credit dispute described here reflect reporting as of September 12, 2026, and are still unfolding as more details emerge.