Just this week, Pushmeet Kohli, Google Cloud’s chief scientist, published a piece in a special AI and science issue of the journal Daedalus, writing: “We are moving toward AI that doesn’t just facilitate science but begins to do science.” With autonomous AI scientists on the horizon, it’s getting harder to justify major efforts to build super-specialized tools, even one like AlphaFold, for which DeepMind scientists won a Nobel Prize, or a potentially life-saving system like WeatherNext. It also points to a much stranger future for science, where humans and AI systems work together as peers, or where AI might even make scientific progress on its own.
To be clear, Google doesn’t seem to be giving up on its specialized AI tools for science. AlphaGenome and AlphaEarth Foundations, trained for genetics and Earth science applications respectively, were launched last summer, and the newest version of WeatherNext appeared in November.
Even so, those tools remain very popular among scientists. Last year, for example, Google said that protein structure predictions from AlphaFold were used by more than 3 million researchers worldwide. And Isomorphic Labs, a Google subsidiary aiming to use AlphaFold and related technologies to create new drugs, just raised a $2 billion Series B funding round.
But there are clear signs of a shift, both in enthusiasm and in resources. Last month, the Los Angeles Times reported that Google fellow John Jumper, who won the Nobel for AlphaFold, is now focused on AI coding rather than science-specific AI tools. It’s not surprising that Google is putting its top talent on the coding challenge, since the company has recently taken a reputational hit because its coding tools don’t currently match those from Anthropic and OpenAI. Still, it might also suggest a growing focus on agentic science at Google, as coding skills are central to the success of some of these systems.
Across the industry, agentic researcher systems are showing real promise. This week, OpenAI announced that one of its models had disproven a major mathematics conjecture, perhaps the most significant contribution that generative AI has made to mathematics so far, according to some mathematicians.
Notably, the model used by OpenAI isn’t specialized for solving math problems or even for research; the company says it’s a general-purpose reasoning model similar to GPT-5.5. If general agents can make independent contributions to mathematical research, they might soon be able to do the same in science (though the fact that scientific ideas must be verified experimentally makes it a tougher field for AI).
ztoog.com