From predictive models to agentic AI in drug discovery
AI agents in drug discovery are software systems that not only predict molecular behavior but also orchestrate tools, data and lab automation to execute multi-step scientific workflows with limited human intervention, turning language instructions into real experimental and clinical actions. For a decade, AI in pharma focused on prediction: models that fold proteins, score ligands or rank trial sites. The new wave is agentic AI pharma, where agents plan, call specialized tools and update downstream systems. This shift is visible across the pipeline: AI agents drug discovery platforms now talk directly to chemistry services, autonomous lab automation stacks and clinical data standards. Instead of serving as smart calculators, models are becoming decision-making teammates that coordinate AI retrosynthesis chemistry, run virtual screens and structure trial protocols so that changes cascade automatically. The result is a gradual transfer of routine scientific work from people to agents.
NVIDIA’s BioNeMo Agent Toolkit: tools for agents, not a single agent
NVIDIA’s BioNeMo agent toolkit shows how infrastructure for AI agents drug discovery is maturing. The platform packages protein-structure prediction, molecular docking, generative chemistry and genomic models as callable “skills” that any compatible agent can use. NVIDIA calls frontier language models the “brains” and BioNeMo the “scientific toolbox,” emphasizing that general models need domain tools and a harness to carry out real workflows. According to NVIDIA, the toolkit has early uptake from nearly 50 partners, including Eli Lilly, Thermo Fisher Scientific and Dassault Systèmes. Because it is agent-agnostic, developers can plug it into diverse operating environments while keeping governance and audit controls. This moves beyond AI that describes chemistry toward systems that can design, score and prioritize candidates as part of repeatable pipelines. As agentic AI pharma platforms standardize on such toolkits, they gain a consistent way to connect language interfaces with specialized life-science computation.
Boltz’s agent-ready API: drug screening through natural language
While BioNeMo supplies the toolbox, Boltz is building a front door for coding agents. Its drug-discovery API exposes BoltzProt-1 for protein design and BoltzMol-1 for small-molecule hit discovery, and the company engineered it “for agents as much as for people.” Internal teams already use coding agents to reach these models, and the public release includes SDKs and integrations with tools such as Claude Code and Gemini command-line interfaces. Experimental validation of BoltzMol-1 spans 10 targets, including GPCRs, kinases, ion channels and protein–protein interactions, with confirmed hits across these categories. By allowing an agent to call a single endpoint and receive ranked candidates, Boltz effectively bridges coding and chemistry. In practice, an autonomous workflow can move from a natural-language request—“find binders for this target”—to AI-generated molecules, ready for follow-up in wet-lab pipelines or further AI retrosynthesis chemistry planning.
MilliporeSigma’s SYNTHIA and the rise of autonomous lab automation
On the lab floor, MilliporeSigma is tying AI retrosynthesis chemistry to autonomous lab automation. Its SYNTHIA platform connects AI-based retrosynthesis with automation hardware and data systems, helping translate digital designs into executable chemistry. The broader portfolio includes AIDDISON for AI-powered drug discovery, the Opentrons-based AAW Automated Assay Workstation for routine assays and green solvent replacements aimed at more sustainable workflows. CTO Karen Madden describes the challenge of focusing investment across research chemicals, reagents, filtration, water systems, software, automation and bioprocessing. The same breadth, however, lets the company assemble an end-to-end “autonomous lab stack” that spans design, execution, data capture and scale-up. When AI agents plug into stacks like this, they can move beyond in-silico tasks and trigger real experiments, turning virtual hits from platforms such as Boltz into physically synthesized and tested compounds with minimal manual intervention.
Medable’s Digital Data Flow Agent: structuring clinical trials for continuous review
Agentic systems are also reshaping clinical development. Medable’s Digital Data Flow Agent tackles one of the most tedious startup steps: translating static clinical trial protocols into structured, machine-readable formats. The agent converts a protocol into CDISC USDM 4.0 JSON so that amendments propagate automatically across connected documents and systems instead of being re-entered by hand. Earlier, Medable used AI to translate a study protocol into a configured eCOA app, including workflows and about 25 language variants, and reported that this could cut deployment timelines in half. Now the company describes its platform as “agentic clinical development,” with agents handling tactical and administrative burdens while experts focus on complex decisions. Framed against regulators’ push for real-time clinical trials, such machine-readable protocols become infrastructure for continuous data review, closing the loop between AI agents drug discovery upstream and structured clinical execution downstream.
