Machines and Molecules

Mike Ryan is the founder of Bullet Point Network. Before that he spent two decades at Goldman Sachs, where he was a partner and co-head of Global Equities, and then sat on the investment committee at the Harvard Endowment, allocating $ 12B to outside funds and $ 6B directly into companies.The episode opens with a comparison the host can't reconcile: one of the best-known venture firms published its performance, and measured against a Nasdaq-100 index fund the average return came out slightly below the index — with much worse liquidity. That pattern isn't unique to one firm. So how do a hundred very smart analysts produce index-matching returns?Ryan's answer has two parts. The first is that public technology stocks have had an unusually strong run on a risk-adjusted basis over the last five, ten and seventeen years. The second is about what goes wrong inside investment firms, and it isn't what most people assume: the largest funds see nearly every deal, employ excellent people, and produce enormous amounts of analysis. What that analysis often contains, he argues, is a repackaging of information the company itself supplied. A firm can hire very smart people, build a beautiful process, and still deliver index-like returns or worse.He then pushes back on the premise. The last five years had a specific problem — funds haven't been able to exit and realise profits — and meanwhile the value creation has moved into private markets. Companies are now going from nothing to a hundred billion, and in a few cases close to a trillion, before they ever go public. He points out that Microsoft, one of the best companies of its era, did almost all of its growing as a public company, and says the new pattern is something he has not seen before in his lifetime.The host raises Ryan's undergraduate thesis, written at Yale under David Swensen, who created the endowment model that pushed universities toward illiquid assets, and asks whether he ever suspected the model had stopped fitting. Ryan treats the 2008 financial crisis as the real test of that conviction: institutions forced to sell into the liquidity crunch suffered badly, while those who held were rewarded over the following two decades. But he expects the next ten to twenty years to be weaker than the last, largely because interest rates are returning to normal levels, and he grants directly that venture returns over the past decade have not objectively beaten the Nasdaq. The first segment ends on what actually lets an investor hold through a downturn: the temperament to stay invested, and a structure with no capital calls to meet.From there the conversation moves to where fund size and fee income pull a manager's incentives away from the people whose money they manage, and why the current buildout may nonetheless require very large funds. Ryan is asked whether making a firm more rigorous can make it worse — whether a fund can pressure-test its way out of the one bet that would have returned it — and where he draws the line between the work he hands to AI systems and the work he insists a human keep. The last third turns to the audience's own problem: how a founder with no revenue and no product should present a real probability of failure to an investor, and what the best of them have stopped hiding.

Mark DePristo, former CEO of BigHat Biosciences, reveals why the best biotech innovation doesn't bolt on AI later but bakes it into research from day one. He shifts from scaling a clinical-stage antibody powerhouse to the nimble "zero-to-one" world of Ember AI Studios, where small teams leverage AI to solve overlooked deep-tech problems. Mark illustrates how embedding high-throughput experimental loops into design processes creates a rhythm that feels less like a lab experiment and more like a jazz band finding its collective, synchronized flow.

In this episode of the Machines and Molecules podcast, host Marcus Bühler, the McAfee Professor of Engineering at MIT, discusses the principle of "adaptability" as the most important universal principle of materials for newcomers to absorb. His work focuses on bridging atomistic simulation, multiscale mechanics, and AI-driven materials design, aiming to automate the scientific process. He explains how biological systems, like spider silk and bone, have a built-in ability to change and adapt over time.The conversation explores the limitations of brute-force molecular dynamics (MD) simulations and the shift toward using smarter, goal-directed computational experiments. Bühler details his lab's use of category theory to model the discovery process and train AI agents to find universal principles, create new hypotheses, and automate scientific research. Finally, we chat about the connections between music, sound, and the material world, which informs his work in AI and the search for unifying theories.

Chris Bahl is founder and CEO of AI Proteins. The conversation delves into the challenges faced by platform companies in the biotech sector, highlighting the financial pressures that lead to the cutting of innovative projects. It discusses how investors prioritize immediate monetization of assets over long-term development, often resulting in the abandonment of potentially successful platforms.TakeawaysIt's tough out there in the world today to be a platform company.The fate of most platforms in biotech is that it's a cool engine.Investors want to monetize that asset.Hundreds of millions of dollars are needed to get products across the finish line.Investors are not interested in products that are 10 years away.Platforms often become line items that get cut from budgets.Many great platforms are effectively thrown in the trash.Successful platforms are often discarded due to financial pressures.Monetization is key for biotech platforms.Innovation struggles under financial constraints.

Victor Guallar is an ICREA Professor and group leader of the EAPM at the Barcelona Supercomputing Center and Co-Founder of Nostrum Biodiscovery. With a joint PhD from the Autonomous University of Barcelona and UC Berkeley, followed by roles at Columbia University and Washington University, he has built extensive expertise in molecular modeling, enzyme engineering, and drug discovery. At the Barcelona Supercomputing Center, he leads the Atomic and Electronic Protein Modeling group, where his work integrates advanced simulations, machine learning, and quantum mechanics to solve challenges in biophysics and sustainability. Victor’s contributions have resulted in over 120 peer-reviewed publications and recognition through prestigious grants, including the ERC Advanced Grant. In this episode of Machines and Molecules, Victor shares his expertise in leveraging Monte Carlo simulations for protein discovery and optimization. Victor explains the value of simulations in molecular science, detailing how they generate data to predict molecular behavior and improve drug discovery, enzyme engineering, and material science. He contrasts Monte Carlo and molecular dynamics methods, emphasizing their respective strengths and his advancements in creating more efficient simulation tools. Victor also discusses the synergy between simulations and AI, highlighting how combining virtual data with machine learning accelerates innovation and improves accuracy. Drawing from his dual roles in academia and industry, he reflects on the disconnect between academic research and industry needs, advocating for practical applications that make scientific work more impactful. The conversation concludes with insights into the benefits of multidisciplinarity, as Victor shares how diverse interests and experiences have shaped his creativity and career. 00:00 - 01:13 Introduction to Victor Guallar 01:13 - 05:57 Molecular Simulations and Their Applications 05:57 - 10:30 Monte Carlo vs. Molecular Dynamics 10:30 - 13:32 How Simulations Generate Data and Integrate with AI 13:32 - 16:56 Sampling vs. Optimization 16:56 - 20:35 The Role of AI in Molecular Modeling 20:35 - 25:30 Applications of Virtual Data in Drug Discovery & Protein Design 25:30 - 31:00 Victor’s 3rd M Word Category: Knowledge