Life sciences organizations have been using AI for years. Machine learning has guided drug discovery, and algorithms have long optimized manufacturing yields. But something is different about this current moment: the speed, scale, and depth of what’s now possible.
In this video Explainer, McKinsey’s Alex Devereson, Delphine Zurkiya, and Lieven Van der Veken draw on decades of experience advising pharma, biotech, and medtech companies to explore what AI actually means for the industry. They examine where real value is being captured today, why implementation so often falls short, what agentic AI makes possible that wasn’t conceivable before, and what a genuinely AI-transformed life sciences industry could look like within a decade. Along the way, they challenge some of the most common assumptions about what it takes to make AI work in one of the world’s most complex and consequential industries.
This interview has been edited for length and clarity.
Why is AI different from previous tech waves in life sciences?
Lieven Van der Veken: This technology cycle is different in three ways: It is broader, faster, and more holistic.
Past cycles might have affected a specific activity—how we manufacture an antibody or how we run a single function. AI is everything, all at once.
It’s also faster. It’s nearly impossible to keep up with the speed at which new variants of this technology are arriving. The adoption rate is faster than anything we’ve ever seen.
And it’s more holistic. It calls into question what a life sciences company of the future even looks like. Activities once outsourced can be brought back in-house. Activities once done internally may be better served by entirely new players.
Alex Devereson: What’s different is this: Until now, progress in life sciences has been limited by—and dependent on—the brilliance of individuals: scientists, designers, and decision-makers.
AI changes that. For the first time, you can tap into the entirety of human knowledge on a subject—a disease, a molecule—instantly.
You can identify connections across data sets that no human has ever been able to make. You can run a hundred or a thousand times more experiments than anyone could previously conceive. And you can break down the organizational silos that have long separated research, clinical development, and patient care.
Delphine Zurkiya: What’s new is that AI now has the potential to transform not just how drugs and devices are brought to market, but how the entire enterprise operates.
In the past, AI in life sciences lived mostly inside the R&D department, used by data scientists and specialists. Now, with generative AI, people who aren’t AI experts can use these tools. And its reach goes far beyond R&D.
Where is AI creating value in life sciences today?
Lieven Van der Veken: I think about AI in life sciences in two categories: scientific AI and operational AI.
On the scientific side, we’re already seeing the first real impacts: modeling clinical trials with greater precision, predicting outcomes, and understanding diseases at a level of granularity—subdiseases and subpatient populations—that simply wasn’t possible before.
On the operational side, life sciences companies are large, complex machines with enormous volumes of documents and processes. Many of these can now be condensed, accelerated, and improved through AI. That’s happening today.
Alex Devereson: There are several areas where we have real, demonstrable, tangible impact from AI right now.
In manufacturing and supply chains, companies are optimizing yields and running processes more efficiently. In clinical operations, trials are being run faster and more effectively. In research labs, AI is accelerating routine processes that previously required significant human time.
And we’re starting to see the next wave: agents that can identify new therapeutic targets, design molecules, and simulate entire clinical trials before a single patient is enrolled.
Delphine Zurkiya: What surprises many of our clients is just how broad the opportunity is. When you really map out all the workflows in a life sciences company and ask which ones could benefit from AI support, the answer is: nearly all of them.
About 80 percent of workflows are what we’d call “agentifiable.” And the impact is surprisingly consistent: We see roughly a 5 to 10 percent improvement in growth and a 3 to 5 percent improvement in margin when AI is deployed at scale.
What is agentic AI and why does it matter for life sciences?
Delphine Zurkiya: An agent is a computer system that can take a number of autonomous steps on behalf of a human.
We’ve had automation before, but it required scripted rules. The moment a rule broke, the software stopped. What’s new with agents is that they can reason and interpret as they go. If something isn’t working, they can try a different approach.
I like to think of it as hiring an intern. The intern has the capability to accomplish a task, but you decide when they need to check in with you and when you trust them to keep going.
Alex Devereson: Agentic AI is the shift from passive tools—ask a question, get an answer—to active operators: systems that can plan, execute, iterate, and run end-to-end workflows independently.
Consider what this means in a lab. You could have agents that read papers, design experiments, order reagents, run experiments on automated machinery, review the outputs, and propose the next round—all in a continuous loop. The human becomes the orchestrator, not the operator.
Lieven Van der Veken: The big impact of agentic AI is that individual tasks—once requiring constant human intervention—can now be done autonomously, and those tasks can grow larger and larger over time.
Take clinical site engagement. Historically, it was expensive, prone to information loss, and largely outsourced. Today, some life sciences companies are using agent systems to manage communications with clinical sites, capturing high-quality patient information at a fraction of the cost.
Play that forward, and you start to see how agentic AI could compress timelines across the entire drug development process.
How will AI change day-to-day work inside life sciences companies?
Lieven Van der Veken: I see change happening at three levels.
First, AI helps people do their current jobs better. If you’re on a team writing a regulatory submission, AI can make you faster and better.
Second, AI links jobs across the value chain. Think about the insights gathered from physicians in the field that actually go to the investigator at the beginning of a clinical trial. How can you use and reuse those insights down the line? AI can help think about processes that may be siloed within a company.
Finally—and this takes time—AI allows us to reimagine processes entirely. R&D today involves roughly 30 sequential steps, which is part of why it takes a decade to bring a drug to patients. But what if, within those 30 steps, there were actually only five decisions: the right target, molecule, patient, evidence, and manufacturability? AI can model all five in parallel from day one.
Alex Devereson: Consider clinical trial design. Today, a team selects a small number of trial designs, works through them manually, and picks one to run.
With AI, you can simulate hundreds of different designs—varying parameters like trial size, patient criteria, end point selection, and cost—and arrive at an optimal design with full information, not best guesses.
Or take drug discovery: Scientists today explore only a small portion of the possible molecular space, limited by what they personally know and can process. AI models can generate and simulate millions of potential experiments, opening up territory no individual researcher could ever reach.
Delphine Zurkiya: What’s also exciting is that about 40 percent of what agents can do is actually work that isn’t being done today. It’s the long tail—the tasks that don’t get prioritized because they don’t justify the human time. Reaching smaller customer segments. Checking compliance on every contract. Synthesizing every relevant paper on a disease. Agents make that work economically viable for the first time.
Why do some life sciences AI efforts scale, and others stall?
Delphine Zurkiya: This is my favorite question—because the answer is almost always the same: It’s about humans, not technology. The technology works. What needs to evolve is the human willingness to change how they work.
That means role modeling from senior leaders. It means training people where they are. It means constant reinforcement. And it means rewiring incentives. Because ultimately, people do the work they’re rewarded for.
Lieven Van der Veken: Let’s start by acknowledging how hard this is. For real impact, many things need to be true simultaneously: the right use cases, the right people, the right technology, the right data, organizational buy-in, and disciplined implementation.
If any one of those breaks down—even if everything else is in place—you don’t get impact. You can have perfect data infrastructure but fail to bring along the medical writers or the clinical scientists. And without adoption, there is no transformation.
What I see in companies that succeed is a perfect storm, or a perfect marriage. It needs to be a marriage of technology and business, working together with a very clear goal. Not a thousand flowers blooming—many pilots, many teams experimenting, everyone thin on resources. Instead, a smaller set of larger domains, with all the energy and focus directed there. Learn as you go. Because every company has its own DNA, and an AI transformation that works in one organization will fail in another if transplanted wholesale.
How can life sciences companies move fast without losing trust?
Alex Devereson: Without compliance and trust, you don’t just lose your competitive edge—you lose medicine. The speed imperative isn’t primarily about beating your competitors. It’s about beating the diseases you’re trying to treat.
There are two dimensions to building that trust. Inside the organization, make clear that AI augments the mission people already have. It doesn’t replace it. It helps them be faster, more rigorous, more effective in the work they do.
Externally—with patients, regulators, healthcare providers—everything AI produces must be explainable and interpretable. Glass box, not black box. You never just “take what the model says.” There must be validation, auditability, and clear documentation of how AI-informed decisions were made.
Lieven Van der Veken: Here’s the encouraging thing: Life sciences companies are actually better positioned for responsible AI than many other industries.
They already have a deep culture around data reliability, bias detection, and protocol compliance. These are exactly the instincts you need when deploying AI. The questions life sciences companies ask around their AI tools—What are the limitations? What are the failure modes? What guardrails are in place?—are the right questions.
We have to remember the limitations. Hallucination risk hasn’t disappeared—it’s just gotten quieter in our conversations. The bar for rigorous testing and validation must stay high. And the guardrails need to remain in place.
What could AI make possible for life sciences in the next decade?
Alex Devereson: There’s unfortunately no good answer to this question. This is an absolute poisoned chalice, because people always overestimate the near-term impact of AI and underestimate the long-term impact. In a two-to-three-year horizon, I think we’ll see the operational dimension of AI reach real maturity—AI embedded across supply chains, clinical operations, medical writing, and lab processes.
In ten years, we could see paradigm shifts: labs and factories designed from the ground up to be run entirely by agents, with human orchestrators working remotely. All decisions made with hundreds of simulated outcomes tested in advance. And stepping into entirely new disease areas that were previously out of reach.
I believe that within a ten-year time frame, we will see diseases that today have no treatment begin receiving treatments that save and change lives. And perhaps in 20 years, we could talk about certain diseases entirely in the past tense, the way we talk about polio today.
Lieven Van der Veken: My dream has three parts.
First, we fully harness AI to identify and develop genuinely novel therapies—not incrementally better treatments, but cures for diseases we haven’t been able to crack. I’m a doctor by training, and I care deeply about motor neuron diseases. We’ve made some progress, but many of these conditions remain devastating. The combination of AI-discovered drug targets, compressed development timelines, and precision patient matching should make it possible to crack some of these within the next five to ten years.
Second, we shift from treating illness to preventing it. The data, the behavioral science, the clinical understanding—these are problems AI is exceptionally well suited to help solve.
Third, the industry itself is restructured. New players emerge—AI-native companies built from the ground up—and the best life sciences organizations are those that had the courage to reimagine themselves.
Delphine Zurkiya: In five years, we’ll start to see leading companies look genuinely different—and we’ll see AI-native companies, built from scratch, pushing the frontier.
In 15 years, if you walk through a pharma or medtech company, you may not recognize what you see. Humans managing teams of agents. No one asking, “Why use AI?” —because it will simply be how work is done.
I personally hope that AI can help us—as an entire ecosystem, from pharma and medtech to payers and providers—make sure that no one has a disease that goes untreated. From diagnosis to finding care, to finding the right care, to staying on treatment—there are so many inefficiencies in the system today. AI should help support humans throughout that entire journey. We have all had relatives for whom the system could have worked better. I really hope that’s what AI will bring us.
