In a move that signals a paradigm shift for the pharmaceutical industry, the veteran scientific software firm Schrödinger has announced a high-stakes strategic collaboration with Bristol Myers Squibb (BMS). At the heart of this partnership is "Bunsen," Schrödinger’s newly unveiled agentic AI co-scientist—a platform designed not merely to assist in research, but to function as an autonomous execution engine for complex computational chemistry workflows.
As the drug discovery pipeline faces mounting pressure to deliver higher-quality candidates in shorter timeframes, the integration of Bunsen represents a transition from traditional, siloed digital tools to an interconnected, agent-driven research environment. By combining Schrödinger’s legacy in physics-based simulation with the emergent capabilities of generative AI, the companies aim to navigate the "trillion-scale" molecular landscape that is rapidly becoming the new frontier of pharmaceutical R&D.
The Chronology of Computational Evolution
To understand the significance of the Bunsen collaboration, one must look at the trajectory of computational chemistry over the last three decades. Founded in 1990, Schrödinger has spent years building the foundation for this moment.
In its early years, the company focused on physics-based simulations, providing the theoretical rigor necessary to model molecular interactions. The landscape shifted dramatically in the early 2010s with the advent of GPU-accelerated computing, which allowed for the scaling of computationally intensive methods such as Free Energy Perturbation (FEP). This allowed researchers to predict the potency and selectivity of compounds with unprecedented accuracy.
By 2020, Schrödinger’s IPO filing codified its "Large-Scale Molecule Exploration" and "Evaluation" frameworks. This was followed by the 2022 release of "AutoDesigner," which demonstrated the ability to generate and filter massive molecular libraries. By early 2024, the company began integrating these generative models with Active Learning FEP+ workflows, creating a closed-loop system that effectively "compiles" drug discovery tasks—much like an integrated development environment (IDE) for software engineering.
The final catalyst arrived in February 2025, with the limited research preview of "Claude Code." Observing how coding agents could write, compile, and iteratively debug code provided the final piece of the puzzle: the ability for an AI to act as an agent that understands scientific outcomes and iterates on them without constant manual intervention.
Redefining Scale: From Wind Tunnels to In Silico Funnels
Pat Lorton, Chief Technology Officer and Chief Operating Officer at Schrödinger, draws a poignant analogy between the evolution of pharmaceutical research and the history of aerospace engineering.
"In the 1940s and 50s, engineers literally built wind tunnels and carved balsa wood to determine aerodynamic drag," Lorton explains. "Decades later, that empirical, physical testing was largely replaced by computational fluid dynamics and supercomputing."
Schrödinger is currently leading a similar transition in pharma. Five years ago, a weekly cycle might have involved screening roughly 100 billion compounds. Today, that number has reached the trillions. However, scale is not merely about volume; it is about the "funnel." As researchers require higher quality in the final candidates—balancing potency, solubility, safety, and metabolic stability—the breadth of the initial search must expand proportionally.
"We need to keep increasing the size of the funnel at the top because we are screening for an increasingly complex array of properties," says Lorton. By utilizing Bunsen, researchers can perform these massive, multi-property screens with a level of selectivity that was previously impossible.
Supporting Data: The Infrastructure Arms Race
The collaboration with BMS arrives at a time when Big Pharma is undergoing a massive capital expenditure cycle centered on AI-ready infrastructure. The sheer computational requirements of running Bunsen and foundational models at scale have turned pharmaceutical giants into high-performance computing (HPC) powerhouses.
- Eli Lilly: Launched "LillyPod" in February, a massive NVIDIA DGX SuperPOD featuring 1,016 Blackwell GPUs.
- Roche: Announced an AI factory in March, utilizing over 3,500 Blackwell GPUs across cloud and on-premises environments.
- Bristol Myers Squibb: On July 20, BMS unveiled plans for one of the most powerful supercomputers in the industry, built on eight NVIDIA DGX Vera Rubin NVL72 systems.
BMS’s "Predict First" approach, which integrates AI-generated predictions into virtually every small-molecule program, serves as the perfect laboratory for Bunsen. As Erin Davis, BMS Vice President of Research Business Insights & Technology, noted, the company is already in production with large-scale predictions and is actively developing its own foundational models—a process that requires the exact type of hardware-agnostic, scalable software architecture that Bunsen provides.
Official Responses and Strategic Vision
The collaboration is built on a long-standing customer relationship, but this new agreement formalizes the development of new functionality within Bunsen and the RetroSynth platform.
Stephen Johnson, Vice President of Computational Sciences at BMS, stated in the official announcement that Bunsen allows their scientists to "think differently about how physics-based tools can be used to navigate molecular design space."
From Schrödinger’s perspective, the goal is to provide a "LLM-agnostic" platform. "The product we’ve built is designed to give customers total flexibility," Lorton explains. "It plugs into Claude and works amazingly, but if a company wants to run its own open-weight models locally on its own SuperPod, Bunsen is built to handle that just as well. We aren’t telling them which way to go; we are giving them the engine to drive their own research."
Implications: Addressing the Talent Bottleneck
Perhaps the most significant implication of the Bunsen era is how it addresses the chronic shortage of computational chemists. According to Lorton, many research programs are critically under-resourced, with some programs receiving only a fraction of a modeler’s time. These experts often spend the majority of their day on "plumbing"—assembling workflows, managing data, and setting up calculations—rather than true scientific discovery.
"For them, the arduous task of just putting these pieces together is a blocker," Lorton says. "It limits their impact. When they use Bunsen, it’s the equivalent of having a postdoc on the other side of the screen who can fill in all the details, handle the heavy lifting, and ask the right questions."
The Human-in-the-Loop Paradigm
Despite its autonomous capabilities, Bunsen is designed with "human-in-the-loop" safeguards. Scientists can choose to have the agent propose a plan without executing it, allowing for expert human review of the "thesis" before any computational resources are consumed.
Lorton emphasizes that the best results come from a hybrid approach. "You are going to have much better luck if you steer the work," he says. "If you know your target and you have experience in that space, you can direct Bunsen in ways it could never discover on its own. We encode as much generic scientific expertise as possible, but the ‘human touch’—the intuition born of years of medicinal chemistry experience—remains the differentiator."
Looking Ahead: The Future of Drug Discovery
As the industry moves toward this agentic future, the implications for drug discovery are profound. By offloading the operational burden of computational chemistry to an agent like Bunsen, organizations can potentially:
- Shorten Cycle Times: Reducing the time between design, simulation, and synthesis.
- Increase Success Rates: By screening trillions of molecules against multiple property filters, the probability of selecting a "drug-like" candidate increases.
- Democratize Expertise: Enabling scientists with less computational experience to leverage the full power of high-end physics-based simulations.
The collaboration between Schrödinger and Bristol Myers Squibb represents a watershed moment. It is the beginning of an era where the limit of drug discovery is no longer defined by human manual labor, but by the ability to effectively coordinate the vast, growing, and immensely powerful computational engines of the 21st century. As the "funnel" grows ever larger, the role of the scientist is evolving from a manual operator to an architect of discovery, directing the agentic workflows that will ultimately define the next generation of medicines.
About the Author
