Google DeepMind is fundamentally reshaping its approach to AI-driven science, moving away from specialized tools designed for specific problems toward autonomous AI agents that could eventually conduct research with minimal human guidance. The shift became visible at Google I/O 2026, where CEO Demis Hassabis announced the company is "at the foothills of the singularity" while simultaneously unveiling Gemini for Science, a suite of agent-based research systems that represent a departure from the company's celebrated work on tools like AlphaFold.

Demis Hassabis

What's Changing in Google's AI Science Strategy?

For years, Google DeepMind built its reputation on creating highly specialized AI tools trained to solve specific scientific problems. AlphaFold, which predicted protein structures and earned Hassabis a Nobel Prize in chemistry, became the gold standard for this approach. WeatherNext, the company's weather prediction system that provided advance warning of Hurricane Melissa's landfall in Jamaica, exemplifies the real-world impact these tools can deliver.

But the company is now redirecting significant resources and personnel toward a different vision: autonomous AI agents that could function as research collaborators or even independent scientists. The centerpiece of this shift is Gemini for Science, which unites several LLM (large language model)-based research systems under one brand. These include the hypothesis-generating AI Co-Scientist and algorithm-optimizing AlphaEvolve, neither of which are yet publicly available but will soon be accessible to any researcher who applies.

The strategic realignment is evident in personnel moves as well. John Jumper, a Google fellow who won the Nobel Prize for his work on AlphaFold, is now focused on AI coding rather than developing science-specific tools. While Google attributes this partly to competitive pressure from Anthropic and OpenAI in coding capabilities, it also signals a prioritization of agentic systems, since coding abilities are essential to autonomous AI researchers.

Why Are Autonomous AI Agents Becoming the Priority?

The appeal of autonomous research agents lies in their potential generality. OpenAI recently demonstrated that one of its general-purpose reasoning models disproved an important mathematics conjecture without being specifically trained for mathematical research. This achievement suggests that flexible AI agents might eventually contribute to scientific discovery across multiple domains, potentially making massive investments in specialized tools harder to justify.

Hassabis frames this evolution as a natural progression. In an interview published in the journal Daedalus, he stated that "for the next decade or so, we should think about AI as this amazing tool to help scientists," but acknowledged that "beyond that timeframe, it is hard to say with any certainty, but perhaps these systems will become more like collaborators".

The company is also investing heavily in foundational capabilities that autonomous agents will need. Gemini Omni, marketed as a consumer product for turning real-life videos into surreal creations, is actually a step toward AI models that better understand the physical world. Hassabis noted that Waymo, Alphabet's self-driving division, is testing AI models that would give autonomous vehicles a kind of "imagination" to react to unpredictable situations. Text-to-video models could be key to general-purpose robotics and artificial general intelligence (AGI), he suggested, because "an AGI is going to have to understand the physical world".

How Is Google Positioning These New Research Systems?

• Branding Strategy: The deliberate choice to name the system "AI Co-Scientist" rather than "AI Scientist" reflects Google's careful positioning of these agents as accelerants for human researchers rather than replacements, at least for now.
• Researcher Access: Google is allowing any researcher to apply for access to Gemini for Science, potentially enabling wider adoption in the scientific community beyond early testers who have already expressed enthusiasm.
• Integration with Specialized Tools: Agentic systems can be designed to call on specialized tools like AlphaFold when needed, meaning the company is not entirely abandoning its earlier work, but rather subordinating it to a broader agent-based framework.

Early testers have been enthusiastic about the potential of these systems. Gary Peltz, a Stanford geneticist, compared using the AI Co-Scientist to "consulting the oracle of Delphi" in a Nature Medicine article.

"For the next decade or so, we should think about AI as this amazing tool to help scientists. Beyond that timeframe, it is hard to say with any certainty, but perhaps these systems will become more like collaborators," said Demis Hassabis.

Demis Hassabis, CEO of Google DeepMind

What Does This Mean for Specialized Science Tools?

Google is not abandoning specialized tools entirely. AlphaGenome and AlphaEarth Foundations, trained for genetics and Earth science applications respectively, were released last summer, and the newest version of WeatherNext launched in November. AlphaFold remains extremely popular; Google reported that protein structure predictions from the tool have been used by over three million researchers worldwide.

However, there are concrete signs of realignment in both enthusiasm and resources. The company's subsidiary Isomorphic Labs, which uses AlphaFold and related technologies to develop new drugs, recently raised a $2 billion Series B funding round, but the company's public messaging and personnel allocation suggest that autonomous agents are now the primary focus.

Hassabis emphasized that Isomorphic Labs is not a traditional biotech company focused on any single drug or disease. Instead, the goal is to create technology that could cure "hundreds of diseases," with current projects serving as test cases. The company's competitive advantage, he noted, is not proprietary datasets but rather "a frontier AGI-lab-quality machine learning research team. No other biotech or pharma has that".

Hassabis

What Does the Singularity Comment Really Mean?

Hassabis's statement that we are "at the foothills of the singularity" generated significant attention, but the context matters. The singularity refers to a theoretical future moment when AI rapidly exceeds human intelligence and transforms the world. Hassabis clarified in an interview that he was being "authentic about what I'm thinking with AGI," and that "the singularity, at least my interpretation of that word and that term, means the era that we're in".

Hassabis

He pointed to several developments as evidence of this transition: agents are becoming useful, coding capabilities are improving, and areas of science and math are being accelerated. The increase in machine autonomy, he suggested, represents one of the key steps toward the singularity.

The tension between Hassabis's lofty rhetoric and the practical achievements of tools like WeatherNext highlights a fundamental question facing the AI industry. While specialized tools deliver measurable, life-saving results, the industry's enthusiasm and resources are increasingly flowing toward autonomous agents with far more speculative potential. Google's strategic pivot suggests the company believes that autonomous AI researchers will eventually prove more valuable than even the most sophisticated specialized tools, even if that future remains uncertain.