The Pande Pivot: Engineering Breakthrough or Just New Packaging?

For over a decade, Vijay Pande was the architect of Andreessen Horowitz's foray into the life sciences. As TechCrunch first reported, Pande—a former Stanford chemistry professor—scaled a16z's healthcare bet into a practice managing nearly $4 billion. Now, Pande has walked away from that behemoth to co-found VZVC with longtime investor Zach Werner.

On the surface, VZVC is a radical departure. While a16z operated on a scale of dozens of bets, TechCrunch reports that VZVC focuses on a handful of concentrated investments per year, employs no associates, and leans heavily on AI for its daily operations. Pande describes this as a shift from biology as a "science of discovery" to something that can be engineered.

In my view, the central question is whether this "AI-native" approach actually solves the systemic fragility of biotech, or if it is simply a rebranding of the same high-risk discovery bets that have historically failed to deliver clinical results.

Pande argues that AI and machine learning can now wrap a level of understanding around complex biological systems to identify drug targets and optimize clinical trials. He posits that AI models will eventually be significantly more predictive than the animal models—such as mice—that currently lead to high failure rates.

However, the math of clinical success remains stubbornly grim. As Pande noted to TechCrunch, the probability of a drug successfully navigating from the first trial to the end of the third is only 20%. When 80% of candidates fail, and trials cost hundreds of millions of dollars, the amortized cost of success remains astronomical. Pande's pivot toward "engineering" suggests a belief that AI can bridge this gap, but the reality is that AI is only as good as the data it consumes.

This is where the "AI-native" promise hits a wall. Pande admits to TechCrunch that biological data cannot be scraped from the internet like text; instead, companies are forced to build walled-off, proprietary datasets. This creates a fragmented landscape where data cannot be easily distilled from one model to another. While Pande suggests AI could eventually act as a "specialist in everything," synthesizing knowledge across oncology and endocrinology, the current reality is one of silos.

If VZVC is merely betting on the hope that proprietary data will yield a better predictive model than the previous generation of biotech, then Pande hasn't changed the game—he's just changed the size of his portfolio. The shift to a leaner, concentrated model is a prudent move in a volatile market, but it doesn't inherently solve the "animal model" problem that plagues the industry.

Ultimately, for VZVC to be a genuine pivot toward scalable engineering, it must prove that AI can move the needle on that 20% success rate. Until then, the distinction between "discovery" and "engineering" is largely semantic. We are still betting on the same biological lottery; we're just using faster computers to pick the tickets.