Japan's pharmaceutical leaders are uniting around agentic AI frameworks to reshape drug discovery, with major companies deploying autonomous AI agents that function as virtual scientists. This shift represents a fundamental change in how the industry approaches molecular research, moving beyond passive data analysis to active, reasoning-based discovery systems that can evaluate, design and optimize drug candidates with minimal human intervention.

What Are Agentic AI Frameworks in Drug Discovery?

Agentic AI frameworks are software systems that give artificial intelligence agents the ability to understand complex problems, plan solutions and take autonomous actions across multiple steps. In drug discovery, these frameworks allow AI agents to access specialized tools and databases, reason about molecular structures and propose new compounds without waiting for human researchers to manually guide each step. Unlike traditional AI systems that simply predict or classify data, agentic systems can orchestrate entire workflows, from screening millions of molecular candidates to designing novel compounds that meet specific therapeutic criteria.

The key innovation is that these agents can call upon specialized AI models and databases as needed, much like a human scientist consulting reference materials and lab equipment. This capability dramatically accelerates the discovery process by eliminating bottlenecks where researchers would normally need to interpret results and manually set up the next experiment.

Which Japanese Pharma Companies Are Leading This Shift?

A consortium called Tokyo-1, operated by Xeureka, has become the epicenter of agentic AI adoption in Japanese drug discovery. The platform brings together some of Japan's most respected pharmaceutical names, each deploying agentic frameworks in different ways to match their research priorities.

• Astellas Pharma: Has deployed nearly all BioNeMo NIM microservices within NVIDIA's digital biology portfolio and is actively running BioNeMo Agent Toolkit, giving its AI agents immediate access to NVIDIA's full life sciences stack for autonomous research.
• Eisai: Joined the Tokyo-1 consortium in April 2026, bringing its research capabilities into the collaborative agentic AI ecosystem alongside other major pharmaceutical players.
• Ono Pharmaceuticals: Is using the Boltz-2 NIM microservice to streamline and accelerate internal drug discovery processes through specialized molecular modeling.
• Daiichi Sankyo: Is conducting ultralarge-scale virtual screening on the Tokyo-1 platform and leveraging NVIDIA RAPIDS to accelerate large-scale data processing for candidate identification.
• Xeureka: Operates the Tokyo-1 platform itself and uses NVIDIA BioNeMo to power its AI-driven drug discovery efforts, enabling researchers to use the most appropriate models and tools across diverse discovery programs.

This consortium approach allows competing pharmaceutical companies to share infrastructure and best practices while maintaining proprietary research, creating a shared ecosystem that benefits the entire Japanese biotech sector.

How to Implement Agentic AI in Your Drug Discovery Workflow

• Start with Foundation Models: Deploy specialized foundation models like NVIDIA's BioNeMo that convert molecular structures into data formats AI can understand and reason about, enabling agents to evaluate and design compounds.
• Integrate Generative Capabilities: Use molecular generative models such as KOYA, which can design novel, high-affinity ligands for target proteins while reflecting researcher intent, allowing agents to propose new candidates autonomously.
• Connect to Screening Databases: Link agentic systems to virtual screening platforms and molecular databases so agents can evaluate millions of candidates and identify promising leads without manual intervention.
• Orchestrate Complex Experiments: Leverage agent toolkits to autonomously propose and orchestrate complex virtual experiments, as demonstrated by Biomy's use of NVIDIA Nemotron-powered agents for drug development.
• Optimize Data Processing: Implement accelerated computing libraries like NVIDIA RAPIDS to handle large-scale molecular data processing, enabling agents to work with massive datasets in real time.

The practical impact is substantial. SyntheticGestalt's molecular AI foundation model ZAO, which can be called directly from the NVIDIA BioNeMo Agent Toolkit, ranked number one on nine public drug-discovery benchmark tasks, achieving the world's best performance in converting small molecules into data formats that AI agents can reason about. This demonstrates that agentic frameworks aren't theoretical; they're already delivering measurable improvements in discovery accuracy.

What Makes These AI Agents More Effective Than Traditional Approaches?

Traditional drug discovery relies on researchers manually designing experiments, running them, interpreting results and then designing the next experiment. This cycle can take months or years. Agentic AI frameworks compress this timeline by allowing autonomous agents to propose experiments, evaluate results and iterate continuously. Biomy's work with virtual cell models illustrates this advantage; the company achieved 90 percent faster spatial transcriptomics analysis using NVIDIA single-cell RAPIDS, and plans to use NVIDIA Nemotron-powered agents to autonomously propose and orchestrate complex virtual experiments for drug development.

Another advantage is scale. Human researchers can evaluate perhaps dozens or hundreds of molecular candidates manually. Agentic systems can evaluate millions in the same timeframe, identifying promising leads that humans might have missed. Daiichi Sankyo's ultralarge-scale virtual screening on Tokyo-1 demonstrates this capability, allowing the company to search through vastly larger chemical spaces than would be feasible with traditional methods.

The framework also enables specialization. Different AI agents can be optimized for different tasks. Takeda's recent collaboration with Boltz to deploy BoltzMol-1 and BoltzProt-1 biomolecular models shows how companies can integrate specialized agents for structure prediction, affinity estimation and generative design directly into existing discovery workflows. This modular approach means researchers can combine the best tools for each step rather than forcing all work through a single system.

What's the Broader Significance for Global Pharma?

Japan's embrace of agentic AI frameworks signals that the technology has moved beyond experimental status into production use. These aren't pilot projects or proofs of concept; they're major pharmaceutical companies deploying autonomous agents to conduct actual drug discovery research. The Tokyo-1 consortium's expansion, with Eisai joining in April 2026, suggests that competitive pressure is driving adoption across the industry.

For the global pharmaceutical sector, this represents a potential shift in competitive advantage. Countries and companies that master agentic AI frameworks may be able to discover drugs faster and more cost-effectively than competitors still relying on traditional methods. Japan's historical strength in precision manufacturing and robotics appears to be translating into leadership in agentic AI systems, where the ability to orchestrate complex, multi-step processes is essential.

The convergence of agentic frameworks with specialized molecular AI models also suggests that future drug discovery will be increasingly automated. Rather than asking "Can we build an AI system to help with drug discovery?" the question becomes "How do we design agents that can autonomously conduct the entire discovery process while keeping human researchers in the loop for strategic decisions?" Japan's pharmaceutical leaders are already answering that question in production environments.