Daphne Koller, founder and CEO of Insitro, has secured major partnerships with pharmaceutical giants like Eli Lilly and Bristol Myers Squibb to advance A.I.-powered drug discovery. Courtesy of Insitro
Daphne Koller, founder and CEO of Insitro, has established herself as one of the most influential leaders in A.I.-powered drug discovery, earning recognition on this year’s A.I. Power Index. After co-founding and leading Coursera to democratize education globally, Koller pivoted to tackle one of science’s most complex challenges: transforming pharmaceutical research through artificial intelligence. Under her leadership, Insitro has secured major strategic partnerships with industry giants, including Eli Lilly for metabolic disease therapies, Bristol Myers Squibb in a $25 million collaboration for ALS genetic target research, and Moorfields Eye Hospital for A.I. foundation models targeting neurodegenerative eye diseases. These partnerships validate her company’s ambitious approach to reimagining drug discovery from the ground up. Rather than simply applying A.I. to existing pharmaceutical processes, Koller advocates for fundamentally re-architecting how biological data is collected and analyzed. Her perspective, shaped by experience in both fast-feedback educational technology and biotech’s slower, higher-stakes world, offers unique insights into A.I.’s potential and limitations.
What’s one assumption about A.I. that you think is dead wrong?
Many people believe that A.I. is a magic wand that can allow you to move super fast. That assumption can be true in the virtual world, where everything moves at the speed of computation. But it’s not true in the physical world, where bits meet atoms. Here, things are slower, more complex, and can be much higher stakes. A related assumption is that the data we have already collected—text and images from the web—contain all the answers that we need. But the data that we need to disentangle biology and derive truly novel insights mostly does not exist yet. We need to generate the right data—data that is fit-for-purpose for machine learning.
If you had to pick one moment in the last year when you thought, “This changes everything,” about A.I., what was it?
From a global level, one such moment was when tool-using A.I. agents transitioned from experimental demos into reliable, everyday products: when people could hand an agent a complicated, long-running real-world task and it completed the job end-to-end with high accuracy and little human oversight; or when an agent showed the ability to observe humans performing complex tasks in action and learn to perform those tasks. From the perspective of our work at Insitromy most impactful moment was to see the strong impact of multiple A.I.-discovered targets in both ALS and MASH on functional endpoints in truly disease-relevant model systems. These results provided compelling evidence that A.I. is capable of actually changing the most important metric in our industry: the probability of success.
What’s something about A.I. development that keeps you up at night that most people aren’t talking about?
My concern about A.I. in science isn’t the distant risk of superintelligence, but the erosion of rigor from the seductive plausibility of generative A.I. In a scientific setting, an A.I. hallucination isn’t just an error; it’s a convincing falsehood that can launch a multimillion-dollar research program down the wrong path. These models are optimized for fluency, not factual accuracy, creating a powerful “illusion of causality.”
My core fear is that the ease of generating plausible answers will tempt organizations to bypass the hard-won ground truth of prospective, experimental validation. The danger isn’t that A.I. becomes too intelligent, but that we become complacent, trusting articulate outputs over real data. That would silently erode the scientific method and waste years chasing beautifully worded mistakes.
If you were tasked with using A.I. to protect students in schools, what would be your approach, and what key challenges would need to be solved?
My approach would focus on personalized education, not surveillance. The goal is to use A.I. as a tool that empowers both students and teachers. The future of most human endeavors is a partnership between a human and a machine, and we need to teach our students towards that future. At the same time, we can leverage learning-platform data to detect subtle signs a student is struggling or disengaged, and provide those to a teacher to help re-engage that student. The machine’s insight will inform the teacher’s uniquely human ability to connect. The key challenge is to build these systems with privacy, equity and empathy at their core.
You’ve made strategic partnerships with major pharma companies while also making significant staff cuts in 2025. How do you balance the pressure to demonstrate quick wins to partners with the inherently long timelines of drug discovery, and what lessons have you learned about managing investor and partner expectations in A.I. biotech?
Our focus at Insitro is on surfacing new biological insights that unlock potential therapeutic pathways for the patients who are waiting; this is a long journey, but it’s also the most important need for transforming drug discovery. And we have been proud to find milestones along the way that allow us to demonstrate that we are on the right track. For example, in our collaboration with BMS they recently nominated our novel target against ALS—an area where no effective treatment exists. We were able to demonstrate that this target reverts multiple functional consequences of ALS, which provided considerable confidence in that target. In metabolic disease, we are advancing several targets into drug discovery that we are now working on with Lilly to design the molecules to turn these targets into medicine. Here, we have incredibly strong evidence from human genetics, cell-based systems and translatable animal models. While speed is important, failing faster is not helpful; our north star is using A.I. to win more. We believe that this is the path for meaningfully transforming this industry.
The pharmaceutical industry has historically been skeptical of A.I.-driven drug discovery promises. What specific technical milestones or data points have you found most convincing to partners like Eli Lilly and Bristol Myers Squibb, and how do you differentiate Insitro’s approach from the numerous other A.I. drug discovery companies competing for pharma attention?
Our goal is to use machine learning to disentangle biology and see what others cannot. The most convincing proof is providing multiple orthogonal evidence about a target or a molecule that demonstrates meaningful impact on disease-relevant biologies. This is what caused BMS to nominate our ALS target.
While there are many other companies in this space, most focus on a part of the process—typically that of designing a molecule against a target. In many cases, that target was previously derisked by others, which means that there are other molecules already in development. Making a better molecule can certainly be useful, but doesn’t address the fundamental need of finding entirely new treatments that can help patients that currently have no other options. We have built an A.I.-first engine that transforms drug discovery and development end-to-end—from target discovery, through drug discovery, and into the clinic. Our approach turns the artisanal journey into a repeatable process that can continue to deliver multiple therapeutics, getting faster and better over time.
You’ve transitioned from democratizing education at Coursera to applying A.I. in drug discovery at Insitro. How has your perspective on A.I.’s practical limitations and capabilities evolved through these different applications, and what misconceptions about A.I. in biotech do you most frequently encounter?
Education taught me that A.I. shines when feedback loops are fast and data are abundant. In biology, you’re in the physical world where bits meet atoms. Data are scarce, biased and causality matters. You have to build the dataset and couple learning tightly to experimentation. The biggest misconception is that traditional drug discovery is ready for A.I. You can’t just drop A.I. onto hundreds of incoherent spreadsheets and expect breakthroughs. We need to re-architect the systems and data collection around A.I. Done right, A.I. is an amplifier of rigorous biology—not a substitute for it.
