Key Takeaways

  • Building software wrappers or AI content tools creates low returns because identical models yield zero pricing power.
  • Developing a single pharmaceutical drug requires 10 years, costs $2 billion, and suffers an 80% to 90% failure rate.
  • DeepMind founder Demis Hassabis shifted focus to Alphabet's Isomorphic Labs to tackle computational biology and cancer cures.
  • Incumbents like Eli Lilly are putting massive capital into laboratory infrastructure rather than pure digital software.
  • Founders are using AI tumor sequencing to deliver personalized therapies directly to cancer patients today.

Software Wrappers Are a Race to the Bottom

Shaan Puri sees countless pitches for AI wrappers, automated copywriting tools, and digital workflow software. His verdict on those companies is blunt: “Using AI in the world of bits, which is basically like digital stuff, I think has very low returns. I think using AI, anybody who's like, I'm using AI to make content, it's a race to the bottom there.” (play from 2:58)

When every founder has access to identical language models through identical APIs, margins collapse toward zero. A software tool that writes marketing copy or summarizes customer emails can be copied over a weekend. Defensibility disappears because the underlying intelligence belongs to the model providers, while the wrapper produces little durable value. If your startup lives entirely on a screen, you are fighting millions of developers for thin slices of enterprise software budgets.

The Economics of Drug Discovery Are Broken

The opportunity sits in what Puri calls “AI in the world of things.” (play from 2:57) When you move from digital text to physical biology, the economic math changes.

Traditional pharmaceuticals operate on a brutal model. As Puri explains, “A drug typically took 10 years and $2 billion to make and had like a 80 or 90% failure rate.” (play from 4:17) When a company spends two billion dollars and a full decade on a compound only to watch nine out of ten attempts fail in clinical trials, the barrier to entry is high, but the waste is staggering.

Applying computational models to biological structures changes those unit economics. Alphabet saw this dynamic and directed its top research talent at the problem. Demis Hassabis, the co-founder of DeepMind who led Google's central AI efforts, turned his personal attention toward Isomorphic Labs. In Puri's words, Hassabis is “now just kind of like all in on Isomorphic, which is basically great, how do we cure cancer?” (play from 4:58)

Meanwhile, pharmaceutical giants like Eli Lilly are funding infrastructure to integrate computational models directly into laboratory pipelines. Instead of guessing how compounds bind through years of trial and error, computational biology models simulate interactions before technicians synthesize a single chemical in the physical world.

Physical Biology Offers Moats That Software Cannot Match

Software tools produce outputs that any user can copy, discard, or replace with another prompt. Biology produces physical outcomes that change human lives.

Sam Parr pointed out that this shift is already delivering results. Founders are building tools around AI tumor sequencing to design custom therapeutic regimens for sick patients. “He has a service that you could sign up for,” (play from 4:00) Parr noted. “My friend with cancer signed up for it and he's doing great.” (play from 4:01)

When a company sequences a patient's specific tumor mutations and matches them to targeted therapies, it creates value that software dashboards cannot replicate. The defensibility comes from real clinical data, physical laboratory partnerships, and measurable patient outcomes. Puri urged young technical builders to reconsider where they spend their careers: “If you're a smart person, join one of these companies. Go build one of these companies. Like, there's never been a better time to go do something that has high impact.” (play from 6:14)

What to Do With This

Audit your project list this week and drop any product whose only defensibility is an API call to a text generator. If you are picking a new market, look for physical bottlenecks where high failure rates create room for algorithmic efficiency. Reach out to three computational biology or tumor sequencing startups hiring operators today, and direct your skills toward atoms instead of bits.