No Priors Ep. 140 | With Benchling Co-Founder and CEO Sajith Wickramasekara
Sajith Wickramasekara, CEO of Benchling, discusses the broken economics of drug development, where high costs and failure rates stifle innovation. He explains how the biotech industry is facing a 'dot com' bust and new competition from China, and argues that AI agents and simulation can provide the solution by making science less artisanal and helping scientists create better molecules through faster, cheaper experiments.
Bringing a new medicine to market is an incredibly long, complex, and expensive process, taking over 10 years and more than $2 billion, with a very high probability of failure late in the process. The core challenge is that our ability to predict what will happen in a clinical setting is poor, and the cost to get that feedback is enormous. Medicines are magical—they represent a fantastic return on investment for healthcare and get cheaper over time as they go generic—so we should want more of them. The pressure on the biotech industry to become faster and cheaper is higher than ever.
The State of the Biotech Industry
The biotech sector is currently experiencing the equivalent of a “dot com bust” after a period of exuberance during the COVID-19 pandemic. This downturn is driven by a confluence of factors, including changes in interest rates, regulatory uncertainty, and a “trough of disillusionment” for promising technologies like gene editing and cell therapies that have been more expensive and taken longer to commercialize than investors expected.
A significant new pressure comes from the rise of Chinese biotech, which excels at speed and cost. Major pharmaceutical companies are increasingly going to China to acquire molecules that they historically would have bought from American biotechs, creating a new competitive dynamic.
The industry’s inefficiency is partly due to its “artisanal” nature. While the physical realm of science has been systematized with tools like DNA sequencers, the digital realm—how data is structured, how scientists collaborate, and how workflows are managed—remains bespoke and fragmented. Most biotech startups are playing a “one-time game” to get acquired, so they don’t invest in building scalable, durable systems for the long term.
This inefficiency is highlighted by the stories of major breakthroughs like GLP-1s and Keytruda, where the core science or the molecule itself existed for years but was nearly overlooked. It shows that “the returns to intelligence in scientific progress are very high.”
AI as the Path to Faster, Cheaper R&D
AI presents a massive opportunity to fix this broken system by enabling scientists to get more shots on goal, create better molecules, and bring them to the clinic faster and more safely. The focus should not just be on the high cost of clinical trials, but on improving the quality of molecules and compressing the years of pre-clinical work. The goal is to increase the share of experiments that have been touched by some kind of predictive capability or simulation.
Benchling AI is built on two main pillars to address this:
- Tools for Simulation: This involves integrating open-source, proprietary, and internal company models directly into the scientist’s workflow. The aim is to make powerful computational tools accessible to wet lab scientists at the right moment, helping them design better experiments and molecules without needing specialized computational skills.
- AI Agents for Automation: These agents automate scientific work by operating on the structured data within the Benchling platform. For example, a deep research agent can answer complex questions across an organization’s entire history of experimental data. In one case, a customer preparing for a costly, 8-month mouse study used the agent and discovered the experiment had already been done years prior by a scientist at an acquired company. This unlocked the organization’s “memory” and saved immense time and resources.
The Future Role of the AI Co-Scientist
In the near term (1-2 years), the most promising model is not a fully autonomous “AI scientist” but rather an AI co-pilot that augments human scientists. The goal is to use AI to compress the 7-10 year drug development timeline down to 2-3 years.
A major hurdle is building trust and creating the right user interface. The biotech world has powerful models (the “GPT”), but it’s still missing the intuitive, accessible interface (the “ChatGPT”) that makes them truly usable for the average scientist. There are significant concerns around accuracy, IP security, and legality that need to be addressed to drive widespread adoption. While large pharma companies are excited and running pilots, their R&D organizations have not yet been fundamentally transformed. Their primary advantage lies in their vast capacity for experimental data generation, which they can use to train unique and powerful internal models.
Lessons from Building an Interdisciplinary Company
Successfully bringing technology to a complex, regulated field like biotech requires building a deeply interdisciplinary culture that merges the worlds of science and software.
- What Biotech Can Learn from Tech: Storytelling. The tech industry excels at creating heroes’ journeys and communicating its mission. Biotech needs to do a better job of telling the stories of its scientists and the patients they impact to combat public misperceptions and highlight the incredible difficulty and importance of creating new medicines.
- What Tech Can Learn from Biotech: Rigor and safety. The “move fast and break things” ethos does not apply when patient lives are at stake. As tech’s ambitions grow into high-stakes domains, it can learn from pharma’s established processes for ensuring validity, accuracy, and safety.