An opinion essay argues that AI agents—AI systems with access to tools and the ability to reason through problems—will accelerate scientific discovery more broadly than approaches like AlphaFold, which rely on vast, meticulously assembled datasets. While AlphaFold succeeded because it could train on 170,000 experimentally validated protein structures built over 53 years, most fields lack such perfect data.
AI agents instead replicate how scientists actually work: combining multiple imperfect tools, reasoning under uncertainty, and iterating as evidence comes in.
Google's AI Co-Scientist reached in hours a conclusion about antibiotic resistance that researchers spent a decade deriving through laboratory work, suggesting agents could lower the cost of experimentation and unlock bolder research.
What happened
An opinion essay argues that AI agents—reasoning systems given access to tools—represent a better path for scientific discovery than the AlphaFold template of massive datasets and neural networks. Google's AI Co-Scientist, announced in May, demonstrated this by independently reaching the same conclusion about antibiotic resistance that Imperial College London researchers took a decade to derive through wet-lab work.
Why it matters
AlphaFold's success required 53 years of international effort and roughly $21 billion in experimental work to build the Protein Data Bank, conditions rarely repeated elsewhere in science. Most experimental fields face inconsistent results: cell lines drift, chemicals have contaminants, lab conditions vary. AI agents sidestep this bottleneck by mimicking how scientists actually work—combining multiple tools, reasoning under uncertainty, and iterating as evidence arrives—rather than requiring perfect, standardized datasets.
What to watch
AI agents still face real barriers: they hallucinate, their judgment is inconsistent, and they have memory and input constraints limiting autonomous operation. If those technical barriers fall away, agents could reshape science by automatically logging every experimental step (solving the reproducibility crisis), preserving institutional knowledge in standardized repositories, and dramatically accelerating the pace at which researchers can test ideas.
Ask the AI about this article →
The essay pivots away from a seductive narrative: that AI, embodied by AlphaFold's 2024 Nobel Prize, will simply accelerate all of science at digital speed. AlphaFold's achievement was real and rare, but it rested on an exceptional foundation—the Protein Data Bank, built through 53 years of international cooperation and roughly $21 billion in experimental work. The author argues this is not a replicable template. Most of experimental science does not operate under such controlled, standardized conditions. Cell lines drift, chemicals contain trace impurities, lab humidity varies. Assembling datasets consistent enough, accurate enough, precise enough, and scalable enough to train modern neural networks in biology or chemistry would require new measurement approaches that do not yet exist and will not be ready for decades.
Instead, the essay advances AI agents as the transformative tool: systems that combine reasoning with access to multiple tools (digital or physical) and can iterate under uncertainty, just as working scientists do. Unlike AlphaFold, which applies brute force to a narrow, well-defined problem, agents are generalists. They do not invent a new way to do science; they digitally model how humans actually do it—combining docking calculations, running binding assays, weighing the strengths and failures of each method, and revising conclusions as evidence arrives. Google's AI Co-Scientist, announced in May, reached a correct conclusion about antibiotic resistance in hours that took Imperial College researchers a decade of wet-lab work to derive independently.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Analyst Ming-Chi Kuo says Nvidia has revived the Rubin CPX AI accelerator with a substantially redesigned arch…
A UK study by UK AI Security Institute and Limbic AI surveyed 6,474 British adults
Broadcom's Clayton Donley says companies are doing mission-critical work with AI agents quickly, but without t…
OpenAI released a new evaluation framework on July 17, 2026, urging companies to measure AI ROI by 'useful out…
As AI agents perform real business tasks, 'Agentic Identity' (giving each AI a unique employee-like ID) and 'D…
The European Union is expanding regulation of ChatGPT and will mandate protections for minors
