AI co-scientists take the lab: agentic AI and BioNeMo toolkit redefine life sciences, freeing researchers for big ideas

Agentic AI systems have redefined life sciences research by autonomously generating hypotheses, designing experiments, and orchestrating workflows, exemplified by DeepMind’s AI Co-Scientist which organizes competing agents to evolve and reflect on research ideas. This shift repositions human scientists from manual hypothesis generators to expert 'prompt engineers,' as Pushmeet Kohli notes, emphasizing the growing importance of formulating precise research questions to effectively guide AI collaborators.

The transformative power of agentic AI is vividly illustrated by DeepMind’s Co-Scientist predicting a bacterial gene transfer mechanism in just two days—a discovery that took human researchers a decade—highlighting AI’s ability to rapidly generate multiple plausible and novel hypotheses. This acceleration is further supported by platforms like Harvard’s ToolUniverse, which between late 2025 and early 2026 facilitated over 700,000 analyses and saw 190,000 installations in a single week, demonstrating widespread adoption of AI-enabled research infrastructure.

Beyond hypothesis generation, agentic AI systems are revolutionizing translational biology by integrating specialized AI platforms with robotic labs to automate complex workflows, thereby reducing manual bottlenecks and freeing researchers for higher-level analysis. As Union explains, these platforms combine tools like ChatGPT for literature synthesis and Codex for long agentic workflows, with over 50 tailored skills built specifically for translational biology, enabling AI to deliberate over complex problems for days to reach deeper conclusions, fundamentally changing the practice of biology.

The deployment of agentic AI in life sciences is carefully balanced with robust governance and biosecurity measures, including model refusals, layered mitigations, and differentiated access for verified institutional users, addressing concerns that beneficial and malicious actors may employ similar tactics. Looking ahead, thought leaders like Joy Zhao envision a decade where autonomous labs, continuous environmental surveillance, and democratized access to expert-level biology accelerate drug repurposing and personalized therapeutics, expanding who can do biology while ensuring responsible innovation.

By mid-2026, NVIDIA’s BioNeMo Agent Toolkit emerged as a transformative domain-specific AI platform that integrates over a decade of NVIDIA’s life sciences libraries and models—such as Parabricks, NeMo, and Nemotron—to accelerate workflows in biology, chemistry, genomics, and drug discovery. Collaborations with leading institutions like the University of Washington’s Institute for Protein Design demonstrated concrete performance gains, achieving 2x faster runtimes for protein structure prediction models like RosettaFold3, underscoring BioNeMo’s capability to supercharge complex scientific computations.

BioNeMo enables rapid specialization of general-purpose AI agents into life sciences experts, significantly compressing research timelines and reducing costs by transforming large language models into domain-specific agents capable of executing end-to-end workflows—from literature review to lab automation. This flexibility is enhanced by its agent-agnostic design, supporting integrations with OpenAI, Anthropic, in-house LLMs, or NVIDIA’s Nemotron models, allowing life sciences companies to avoid vendor lock-in and tailor AI-driven drug discovery pipelines to their needs.

The toolkit’s architecture packages proven life sciences models—covering protein folding, molecular docking, generative chemistry, genomics, and imaging—into agent-callable tools with clear schemas, exposed via NVIDIA’s NIM microservices that can be deployed on-premises, cloud, or hybrid environments. This design not only optimizes token efficiency and computational costs by reducing LLM calls and GPU usage but also enables seamless integration with existing laboratory systems from partners like Tecan and Simulations Plus, positioning BioNeMo as a practical instrument bridging high-performance computing and physical lab automation.

NVIDIA CEO Jensen Huang encapsulated BioNeMo’s impact by likening AI agents empowered by the toolkit to possessing 'the skills of a PhD research assistant and the speed of a supercomputer,' signaling a paradigm shift in scientific workflows. This vision frames agentic AI as the next-generation scientific instrument—akin to the microscope or gene sequencer—that autonomously reasons, plans, and acts across digital and physical labs, thereby accelerating discovery timelines and reducing costs for over 50 leading life sciences companies already adopting the platform.

By mid-2026, Bio Protocol’s BioAgents platform emerged as a pioneering decentralized AI co-scientist system, integrating flexible deployment across user hardware and cloud environments with popular platforms like Discord and GitHub. Central to this ecosystem is the Gaia AI intelligence layer, which enforces a robust governance framework spanning six accountability dimensions—Delivery, Transparency, Momentum, Stewardship, Legitimacy, and Impact—ensuring that AI-generated scientific outputs maintain rigor and trustworthiness. Complementing this infrastructure, Bio Protocol’s OpenLabs foster human-agent collaboration by combining human oversight with blockchain-verified outputs, while the BIO token incentivizes participation through staking and governance, effectively aligning economic and scientific incentives within decentralized biomedical research workflows.

Certara’s adoption of NVIDIA’s BioNeMo Agent Toolkit exemplifies the transformative impact of agentic AI in drug discovery, enabling autonomous AI scientists to interpret high-level goals and independently orchestrate complex multi-step research workflows. CEO Jon Resnick highlights this 'scientist-in-the-loop' paradigm, where AI accelerates computationally intensive tasks—achieving over 10x speedups in genomic analysis and 3,000x in cheminformatics—while preserving human expertise and judgment. This approach has gained rapid traction industry-wide, with over 50 companies and research institutions, including Eli Lilly and Databricks, integrating BioNeMo to enhance model-informed drug development processes.

Biomni stands out as a comprehensive AI-powered biomedical co-scientist that autonomously designs and executes complex research workflows by integrating an expansive biomedical action space comprising 150 specialized tools, 105 software packages, and 59 databases across 25 subdomains. Its large language model-driven architecture decomposes user queries into executable multi-step plans, enabling it to perform tasks—from literature review to experiment design—with expert-level accuracy while drastically reducing research time; for instance, analyzing over 450 files in 40 minutes, a task estimated to take humans more than 60 hours. With adoption by over 10,000 academic and industry labs, Biomni exemplifies a paradigm shift that augments human researchers by automating mechanical burdens, thereby freeing scientists to focus on ideation and interdisciplinary collaboration.

While Biomni’s modular and flexible design allows continuous integration of new data sources and analytical tools, experts caution that its ultimate success hinges on the quality of input data and the effective integration of research infrastructure rather than solely on foundational AI models. As Lim Soo-jong notes, the core innovation lies in the LLM’s ability to orchestrate the entire research process as a unified workflow, connecting diverse biomedical resources seamlessly. However, Park Jong-hwa warns that even the smartest AI cannot overcome poor data quality, emphasizing the necessity for ultra-precise datasets, such as aging-related multi-omics data, to unlock truly creative hypothesis generation and advance AI-assisted life sciences research.

By mid-2026, strategic partnerships between technology leaders like Snowflake and NVIDIA have pioneered the integration of governed agentic AI workflows directly within life sciences data environments, enabling pharmaceutical R&D teams to conversationally orchestrate complex drug discovery pipelines. This collaboration, exemplified by the fusion of Snowflake’s Cortex Agents with NVIDIA’s BioNeMo Agent Toolkit, not only accelerates multi-step AI-driven workflows such as target retrieval and compound generation but also establishes a secure, scalable control plane that unifies data, AI, and domain expertise—ushering in enterprise-ready AI that transforms scientific discovery through trusted, operationalized workflows.

NVIDIA’s BioNeMo Agent Toolkit has rapidly become a cornerstone technology bridging high-performance computing with lab automation and biotech workflows, as evidenced by integrations with firms like Tecan, Simulations Plus, and Sigmatic Sciences. This positions NVIDIA not just as an AI infrastructure provider but as a pivotal enabler embedding agentic AI into pharmaceutical R&D and automated laboratory systems, broadening the scope and efficiency of AI-driven research across the industry.

Innovations in decentralized science are also being propelled by agentic AI, with companies like Bio Protocol deploying AI co-scientists through their Gaia intelligence layer to automate hypothesis generation and coordinate research workflows. Their ecosystem, strengthened by partnerships with VitaDAO and Molecule, leverages blockchain governance and token incentives—such as the BIO token backed by a $6.9 million funding round—to foster transparency, accountability, and community engagement, marking a novel fusion of AI and decentralized funding models in life sciences R&D.

Certara’s collaboration with NVIDIA to embed the BioNeMo Agent Toolkit into its AI platform exemplifies the industry-wide adoption of agentic AI, where autonomous 'AI scientists' accelerate drug discovery by simulating human biology and automating complex workflows with GPU-accelerated speedups ranging from 10 to 3,000 times. CEO Jon Resnick highlights a 'scientist-in-the-loop' approach that augments human expertise rather than replaces it, reflecting a broader trend with over 50 companies—including Eli Lilly, Databricks, and Snowflake—embracing similar AI toolkits to enhance reproducibility, scale, and strategic decision-making in pharmaceutical R&D.

By mid-2026, agentic AI systems like Biomni have fundamentally transformed scientific roles in life sciences by automating complex, repetitive tasks such as routine pipetting, literature review, and multi-step biomedical workflows. This automation liberates researchers from laborious mechanics, enabling them to focus on creative problem-solving, strategic analysis, and interdisciplinary collaboration. As researcher Jure Leskovec observes, 'The hurdle in biomedical science is not intelligence or ideas; it is mechanics... Biomni can do this work in minutes,' highlighting how AI shifts human effort toward innovation rather than manual execution.

Agentic AI platforms enhance efficiency, accountability, and reproducibility by integrating advanced computational models with extensive validation pipelines, including synthetic data testing and wet lab experiments. Biomni exemplifies this by autonomously decomposing user queries into actionable subtasks and leveraging a comprehensive biomedical action space of over 150 specialized tools, 105 software packages, and 59 databases to deliver expert-level accuracy while drastically reducing time investment—cutting tasks from 60+ hours to mere minutes. This modular design also ensures continuous adaptation to evolving research needs, fostering rigorous, traceable science with full citation tracking.

The integration of large language models (LLMs) within agentic AI systems revolutionizes research workflows by orchestrating complex tools and interpreting intricate outputs like predicted crystal structures, thereby reducing the cognitive load on scientists. As one expert notes, 'You don't need to deeply understand all of the hyperparameters... you can just let this thing go and figure out and solve the problems.' This collaborative design, featuring over 50 specialized skills and interactive visualizations, transforms team dynamics and democratizes access to advanced computational expertise, positioning AI as a powerful partner rather than a replacement in scientific discovery.