Demis Hassabis, CEO of Google DeepMind, declared from the Google I/O stage that humanity is "standing in the foothills of the singularity." The phrase, uttered as a video showed how WeatherNext anticipated Hurricane Melissa's landfall in Jamaica and potentially saved lives, reveals a growing tension inside the company: should Google keep investing in hyper-specialized AI tools for science, or bet everything on autonomous systems that will one day do science without human intervention?
The Big Picture
Hassabis's speech reflects a schism running through the entire industry. On one side, tools like WeatherNext or AlphaFold solve concrete problems with millimeter precision. On the other, agentic systems based on large language models (LLMs) promise to execute complete research projects, from formulating hypotheses to publishing results. This second vision fuels current enthusiasm, including the concept of recursive self-improvement: machines designing smarter machines in an accelerating cycle.
But concrete facts show a more complex reality. Just last week, Pushmeet Kohli, chief scientist at Google Cloud, published in the journal Daedalus that "we are moving toward AI that doesn't just facilitate science but begins to do science." Yet Google has not abandoned its specialized tools: AlphaGenome and AlphaEarth Foundations were released in summer 2025, and the latest WeatherNext version came out in November. Over three million researchers have used AlphaFold to predict protein structures, and Isomorphic Labs, Google's drug-discovery subsidiary, just closed a $2 billion Series B round.
“"Jumper's reassignment to coding suggests Google prioritizes agentic systems over specialized science tools"”
By the Numbers
- 3 million: Researchers who have used AlphaFold worldwide, according to Google data from 2025.
- $2 billion: Series B funding round raised by Isomorphic Labs, Google's AI drug-discovery spinoff.
- 1 Nobel Prize: Awarded to John Jumper for his work on AlphaFold, before being reassigned to AI coding projects.
- 1 disproven math conjecture: The most significant contribution of generative AI to mathematics so far, according to OpenAI, which used an agentic model.
- 0 evidence of singularity: Despite Hassabis's claims, WeatherNext's real impact is limited to weather prediction, not a qualitative leap in general intelligence.
Why It Matters
The reassignment of John Jumper to coding tasks is the clearest sign of a strategic shift at Google. The Los Angeles Times reported last month that the Nobel-winning scientist now works on improving Google's AI coding capabilities, an area where the company has lost ground to Anthropic and OpenAI. If the best minds are deployed to close that gap, it is inevitable to ask whether tools like AlphaFold will receive fewer resources in the future.
This move is not isolated. OpenAI announced this week that one of its models disproved an important mathematical conjecture, considered by some mathematicians as the most relevant contribution of generative AI to their field. The model used an agentic approach, working autonomously to explore and prove the conjecture false. Such achievements fuel the narrative that autonomous systems will soon surpass specialized tools.
However, the path to AI that does science on its own is fraught with obstacles. Current agentic systems require enormous amounts of data and computational power, and their reliability remains questionable. Moreover, the scientific community values transparency and reproducibility, qualities that black-box models do not always offer. The tension between both visions will not resolve overnight.
What This Means For You
If you are a tech investor, this pivot has direct implications. Companies betting on agentic systems could capture a disproportionate share of future value, but they also face regulatory and technical risks. Specialized tools, on the other hand, offer proven use cases and more predictable revenue streams.
- 1Investors: Diversify between companies developing specialized AI (e.g., drug discovery) and those building agentic systems. Do not assume one vision will annihilate the other in the short term.
- 2Scientists and researchers: Keep using tools like AlphaFold while they are available, but start familiarizing yourself with agentic systems. Human-machine collaboration will become increasingly common.
- 3Regulators: Prepare legal frameworks for autonomous AI in science, especially in sensitive areas like gene editing or weapons development. The pace of innovation will outrun legislation if action is not taken now.
What To Watch Next
The next milestone will be Alphabet's earnings call in July, where changes in resource allocation across AI projects could be announced. Also watch for publications by Pushmeet Kohli and other DeepMind leaders; their writings often foreshadow strategic moves.
On the regulatory front, the European Union is advancing its AI Act, which could classify agentic systems as high-risk. Any restrictions would directly affect the pace of adoption. Finally, keep an eye on OpenAI's achievements in mathematics and science; if they manage to disprove more conjectures or propose new theories, pressure on Google to accelerate its transition will increase.
The Bottom Line
Google is not abandoning AI for science, but it is reorienting its priorities. Jumper's move to coding and OpenAI's success in mathematics suggest agentic systems are gaining ground. However, specialized tools like AlphaFold and WeatherNext remain indispensable. The key will be how Google balances both visions over the next 12 months. For now, the singularity remains a metaphor, not a business plan.
Deeper Analysis: Implications for Investors and Operators
Google's decision to reassign Jumper is not merely technical; it reflects a strategic bet on agentic AI as a future growth driver. For investors, this suggests Alphabet may be willing to sacrifice near-term revenue from specialized tools to gain share in the autonomous agent market, which Gartner estimates will reach $50 billion by 2028. However, the risk is that OpenAI and Anthropic maintain their coding lead, leaving Google as a follower.
For operators of research labs, the signal is clear: while AlphaFold will continue to be maintained, major updates may slow. It is advisable for bioinformatics teams to start evaluating open-source alternatives like Meta's ESMFold, which already covers 80% of AlphaFold predictions according to recent benchmarks. Diversifying tooling will be key to avoiding vendor lock-in.
Moreover, OpenAI's success in mathematics raises an uncomfortable question: could agentic AI replace scientists in creative tasks? Although early, labs should invest in training their staff to use autonomous agents as research assistants, not replacements. Human-machine collaboration will be the norm, and those who adapt first will have a competitive edge.
