AI Just Solved Problems Scientists Couldn’t Crack for Decades. Here’s Why Business Leaders Should Pay Attention.
On 1 August 2026, OpenAI quietly released something significant.
An internal version of its next model, called Astra, solved ten major open problems in mathematics and theoretical computer science. These were not routine exercises. They were problems that professional researchers had made little or no progress on for 10 to 27 years.
The model generated the core arguments. Humans helped shape them into papers. The model then formalised every proof in Lean — a system that allows mathematical proofs to be machine-checked for correctness. The total compute cost for all ten solutions was roughly $2,000.
Elon Musk’s response was characteristically direct:
“Welcome to the Singularity. How’s the temperature?”
A few days earlier, OpenAI CEO Sam Altman had already stated that “we are now in the singularity.”
Why This Matters Beyond the Research Community
Most business leaders still experience AI as a productivity tool. It writes faster, analyses data quicker, codes more efficiently, and summarises documents in seconds. That is useful, but it is still an extension of existing work.
What happened with Astra is different.
For the first time at this scale, AI has produced original, verifiable research-level results in domains that previously required years of specialised human effort — and it did so at very low cost. This is no longer just automation of known tasks. It is acceleration of the discovery process itself.
When the cost of generating new knowledge drops this sharply, the implications ripple outward:
- Innovation cycles compress. Companies that once needed large research teams or long timelines to explore complex technical questions may now move significantly faster.
- Competitive advantage shifts. The ability to ask better questions and integrate AI-generated insights will matter more than simply having more people or more data.
- Talent models change. The bottleneck moves from “Can we hire enough specialists?” to “Do we have people who can frame the right problems and critically evaluate AI outputs?”
- Planning horizons shorten. Five-year strategies built on linear assumptions about technological progress are becoming riskier.
A Clearer Signal, Not Just Noise
It is easy to dismiss singularity language as hype. In its strictest form, the term refers to a point where AI improves itself so rapidly that human understanding and control become difficult. We are not there yet.
What we are seeing is a meaningful step-change. AI systems are beginning to contribute original intellectual work in hard scientific domains, not merely recombine existing knowledge. The formal Lean proofs make these claims unusually open to scrutiny — which is important. Claims without verification are just marketing. Claims with machine-checkable proofs demand attention.
For business leaders, the precise definition of “singularity” matters less than the practical reality: the rate at which new capabilities appear is increasing, and the cost of generating sophisticated outputs is falling.
What Leaders Should Do Now
You do not need to become a mathematician. You do need to update your mental model.
Treat AI less as a set of tools and more as a continuous capability shift. Ask sharper questions inside your organisation:
- Where would a significant acceleration in insight generation create the most advantage?
- Which of our current processes still assume that deep expertise is scarce and slow?
- Are we building the internal ability to evaluate and act on AI-generated work, or are we still primarily focused on adoption theatre?
The companies that adjust their operating rhythm and decision-making speed will pull ahead. Those that continue to treat this as incremental will find the gap widening faster than expected.
The temperature is rising. The question is whether your organisation is paying attention.
Further reading: When AI Agents Escape, on the risk side of the same shift; my book AI for Business Leaders; and the AI Leadership Course, a six-week cohort for founders and senior leaders.
Source: OpenAI research note on the ten advances — https://openai.com/index/ten-advances-in-mathematics/
