Artificial intelligence may help turn biology from a process of fortunate discovery into something more deliberately engineered, but its progress depends on data that cannot simply be collected from the public internet.
That tension is central to investor and scientist Vijay Pande’s view of AI-driven biotechnology. In a TechCrunch interview, the former Stanford chemistry professor and a16z investor described opportunities in drug targets, clinical trials and precision medicine while warning that biological datasets remain scarce and fragmented.
Pande now runs VZVC with investor Zach Werner. The small firm plans about five concentrated investments a year, uses AI agents for daily operations and focuses part of its work on AI for healthcare delivery and clinical trials.
AI could change several stages of drug development
Pande argues that AI and machine learning can help computers reason about complicated biological systems. He sees potential for identifying the targets a drug should affect, designing the drug and assisting with clinical trials, which he described as the most expensive part of development.
The economics remain difficult. Pande said a trial can cost hundreds of millions of dollars and that only 20% of drugs entering the first trial successfully complete the third. He attributed many failures to the limited ability of animal models to predict human outcomes.
His expectation is that an AI model will not be perfect but can eventually become more predictive than an animal model. That is an investor-scientist’s forecast, not evidence presented in the interview that a specific model has already crossed the threshold across drug programs.
Pande also challenged the idea that synthetic data has already made clinical trials substantially cheaper. He called that outcome an aspiration. He said the cost and time required to reach trials have fallen, particularly with AI, while the trials themselves can still demand enormous spending.
The distinction places AI’s nearer-term value in improving decisions before and during development rather than assuming that expensive human trials can simply be removed.
Precision medicine needs a richer view of patients
Another opportunity is selecting the right treatment for an individual. Pande said doctors often have limited information, prescribe one drug and switch to another if it fails. Precision medicine aims to compare a patient’s measurements with what is normal for that individual instead of relying only on population averages.
Genomics has long been central to this effort, but Pande compared a genome with a home’s original blueprint: informative, yet unable to show everything that has changed since construction. He pointed to proteomics and other measurements as potentially more relevant to a person’s current disease state.
He also connected AI with increased automation in robotic measurement. Together, those technologies can generate and analyze more biological observations. Pande said AI for biology and AI for chemistry have both advanced significantly over the past decade, with one side asking how a disease can be treated and the other how a drug can target a particular protein.
The broader vision is an AI that can combine knowledge across medical specialties. Pande said such a system could, in principle, see relationships beyond the view of any single specialist. The phrase “in principle” matters because the necessary information is not yet uniformly available or shared.
Scarce and siloed data set the boundary
Unlike text, biological data cannot be gathered at internet scale by scraping public pages. Pande said companies therefore tend to build separate, protected datasets, and the information cannot readily be distilled from one model into another.
That creates a structural limit. Even a capable algorithm cannot find patterns in measurements it does not possess. Medical specialties and companies may each hold a portion of the relevant evidence while incentives to protect discoveries keep those portions apart.
Pande expects biological information atlases, often implemented as foundation models, to become more common. He believes open-source biological models could eventually have broad impact, much as open models have in language AI. The interview does not establish how those atlases would obtain access to private data or resolve competition among their owners.
His warning is direct: claims that AI will cure everything overlook the quality and quantity of available data. VZVC’s concentrated investment model reflects that caution. The firm is placing fewer bets and emphasizing long relationships with founders rather than treating technical novelty alone as sufficient.
AI may improve biological engineering, clinical development and personalized treatment. Pande’s account also makes the constraint clear: advances will be bounded by whether researchers can produce, connect and responsibly use the biological evidence that models require.
