Axiom Math's Axplorer AI system solves complex mathematical problems in hours on a single machine, marking a transition from AI-assisted research to autonomous discovery.1 OpenAI chief scientist Jakub Pachocki expects models capable of working indefinitely "in a coherent way just like people do" by 2028, functioning as automated research interns.2
The technology is advancing faster than accessibility. Mathematicians express excitement about Google DeepMind's AlphaEvolve but face barriers—users must request DeepMind personnel to input problems directly.1 This closed-access model contrasts with Axiom Math's approach of making AI math tools more broadly available to researchers.
Pachocki acknowledges AI systems by 2028 won't match human intelligence across all dimensions. "You don't need to be as smart as people in all their ways in order to be very transformative," he notes.2 The focus is domain-specific capability rather than general intelligence, particularly in mathematical problem-solving and research workflows.
The low-hanging fruit remains abundant. "There are tons of problems that are open because nobody looked at them, and it's easy to find a few gems you can solve," says François Charton, highlighting opportunity in neglected research areas.1 AI systems can systematically explore these overlooked problems at scale, potentially accelerating discovery rates across scientific disciplines.
For technology investors, the implications extend beyond software. Compressed research cycles could shorten time-to-market for pharmaceuticals, materials science, and semiconductor design. Financial institutions face disruption as quantitative research, risk modeling, and algorithmic trading become increasingly automated.
Regulatory challenges loom large. Pachocki characterizes automated research capabilities as "a big challenge for governments to figure out," particularly around intellectual property, research validation, and economic displacement.2 The technology is advancing ahead of policy frameworks designed for human-paced innovation.
Current limitations persist in novel ideation—AI excels at solving defined problems but struggles with identifying breakthrough questions. This keeps human researchers central to agenda-setting even as execution becomes automated. The distinction matters for workforce planning and investment thesis development across research-intensive sectors.
