Earley AI Podcast - Episode 96: AI in Clinical Trials, the Vibe Coding Fallacy, and Bending Eroom's Law with Patrick Leung

Why Applying AI to Drug Development Is One of the Most Technically Demanding Problems in the Industry - and What Is Finally Making It Solvable

Guest: Patrick Leung, Chief Technology Officer at Faro Health

Host: Seth Earley, CEO at Earley Information Science

Published on: August 4, 2026

In this episode, Seth Earley speaks with Patrick Leung, Chief Technology Officer at Faro Health, who spent over a decade at Google including working on Google Duplex before bringing that technical depth to one of the most regulated and high-stakes domains in medicine. They explore why generative AI is in the trough of disillusionment in pharma, what the vibe coding fallacy costs organizations that believe they can build clinical software by prompting, how classical machine learning models and modern LLMs are working together to forecast trial outcomes, and why every day of clinical trial delay can cost up to half a million dollars in lost revenue. Patrick shares candid and specific insights on prompt injection as the new SQL injection, why human experts cannot be removed from clinical AI workflows, and what bending Eroom's Law would mean for patients worldwide.

Key Takeaways:

• Generative AI is in the trough of disillusionment in pharma - the initial hype that AI could automate entire clinical processes has collided with the real complexity of the domain and the limits of the technology.
• Vibe coding hits an event horizon of complexity - demo apps are achievable by prompting, but real enterprise software requires proper engineering, security design, testing discipline, and architectural decision-making that AI cannot replace.
• Prompt injection is the new SQL injection - any tool that uses AI to process user input is now vulnerable to a class of attacks that did not exist before, and most organizations are not yet protecting against them.
• Classical machine learning models and modern LLMs are more powerful together than either is alone - survivor curve models from insurance analytics proved directly transferable to clinical trial forecasting with strong results.
• Every day of clinical trial delay can cost up to half a million dollars in lost revenue - and a typical amendment forcing a trial redesign and resubmission runs three to six months.
• Generic general-purpose models cannot replace domain-specific knowledge engineering in clinical contexts - the claim that they can is an easy sales pitch that does not survive contact with the actual complexity of the problem.
• The goal is not to automate clinical professionals out of existence but to remove the rote and repetitive work so they can focus on the judgment calls that only they can make.

Insightful Quotes:

"There's no escaping the fact that you need to test software. There's no escaping the fact that you need to have specs that are really well thought out. As you add more features to a codebase, it gets more complex and unwieldy and difficult to maintain. You can't vibe code your way out of those key design decisions." - Patrick Leung

"I found myself applying models I'd learned about in a completely different domain. Survivor curve models we used for predicting insurance policy claims worked pretty well when applied to clinical trials. Transferability is really a thing." - Patrick Leung

"Eroom's Law is not sustainable. Any exponential increase in cost is not sustainable by definition. So we want to bend Eroom's Law - and hopefully reverse it. Why not?" - Patrick Leung

Tune in to discover why AI in clinical drug development is one of the hardest and most consequential problems in the field - and what is finally making it tractable.

Links

LinkedIn: https://www.linkedin.com/in/puiwah/

Website: https://www.farohealth.com

Thanks to our sponsors:

• VKTR
• Earley Information Science
• AI Powered Enterprise Book