Lecture, Conference Presentation
Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, AI in Life Sciences
- Eric Abrams projects that long-term lab work will shift predominantly to computer-based design within the medium term, with "zero-shot" antibody molecule design identified as a near-term goal.
- The plan at Anthropic involves training the Claude model to execute the full end-to-end R&D process, spanning basic research, drug development, clinical trials, regulatory steps, and manufacturing transfers.
- Anthropic aims to achieve an order-of-magnitude acceleration in the life sciences by integrating full-stack model training with accessible product workflows.
- Current median drug development timelines of 10 to 15 years are forecast to potentially decrease to the five-year range, with a theoretical lower bound of a few years constrained by clinical trial logistics.
- The current discovery rate of approximately 30 net new targets annually is deemed insufficient, prompting a focus on scaling high-quality target discovery to unlock thousands of potential targets.
- Anthropic intends to establish a wet lab sandbox for testing tools and accelerating metagenomics discovery while explicitly stating they do not plan to monetize drugs directly.
- Integration of AI into lab instruments is expected to be a "year or two" step within a broader multi-year journey toward AI-controlled autonomous wet labs.
- Josh Tam predicts a "Jevons paradox" where AI-driven efficiency increases the ROI of experiments, leading to a renaissance of physical lab activity rather than a reduction.
- Purely software-based bio tools lacking frontier-level capabilities are forecast to lose market share to AI methods that surpass the physics-based or computational techniques developed over the last 30 years.
- The biotech landscape is anticipated to evolve exponentially over the next couple of years through faster iteration cycles between AI design and physical lab testing.
- Consumer-focused markets for sleep and lean muscle mass are expected to grow significantly, mirroring the trajectory of GLP-1 drugs for obesity.
- Josh Tam expresses skepticism regarding business models based solely on tool sales, predicting that dropping barriers to entry will enable "pipeline in a person" models and challenge traditional tool-only providers.
- The democratization of drug development is expected to lower entry barriers, allowing small teams to run clinical-stage programs and increasing competition for traditional tool-only businesses.
- Josh Tam is bullish on pharmaceutical companies evolving into capital aggregators that reinvest revenues more effectively due to AI-driven discovery efficiency.
- Josh Tam believes China's discovery efficiency cannot outpace AI, positioning AI as a necessary democratizing function for the US to compete globally.
- The convergence of large language models as "outer loops" and specialized foundation models is predicted to enable fewer iteration cycles and potentially "zero-shot" drug discovery.
- Josh Tam expects models to eventually reach a performance threshold where they can run autonomous drug programs and achieve "best in class" results.
- Short feedback loops in discovery versus long loops in clinical development currently concentrate AI attention on the discovery phase rather than development.
- The integration of AI is expected to alleviate "target crowding" around the ~30 annual new targets by unlocking vast potential from the human genome.
- Future capabilities include designing "more sophisticated medicines" with atomic-level precision to expand possibilities beyond current off-target effect constraints.
- Scaling laws and expanding datasets are predicted to enable "single-shotting" complex biological systems, challenging the assumption that zero-shot biology is impossible.
- While AI may not cure all diseases within the next five years, it is expected to deflect the trajectory of disease burden and aging toward previously unrealistic outcomes.