The AI revolution in drug discovery is not just about speed

Let me preface this by saying that in more than fifteen years as a working scientist in drug discovery, including years spent very close to the actual drug design process over and over again, I have never seen anything like this. I don’t think I’ve seen anything that evokes quite the same sense of wonder, the same sense of suddenly gaining superpowers, almost a sense of magic. I hesitate to use that word, but I genuinely struggle to find a better one. And I am someone who has long been an AI skeptic.

There have certainly been moments when individual tools or scientific advances were deeply impressive. Docking, when it emerged, was incredibly powerful. Free energy calculations represented another major step forward. Large-scale 3D shape and electrostatic searching was groundbreaking. Each of these, in its own way, felt like a leap. But all of them shared a common limitation: they were constrained by how much time and effort it took to actually use them, by how effectively you could deploy them, and by how long you were willing to spend debugging them, integrating them, and stitching them together into something usable.

Anyone who has worked in computational chemistry, medicinal chemistry, computational biology bioinformatics or related disciplines knows exactly what that experience feels like. And describing it as “tedious and painstaking” doesn’t really do it justice, because that makes it sound merely slow. It was slow, yes, but more importantly it was deflating, even demotivating. If you knew it was going to take three weeks to build a pipeline, very often you simply wouldn’t do it, and that is the critical point that tends to get missed.

When people talk about accelerating workflows, they usually frame it in purely quantitative terms: five times faster, ten times more productive. That framing is incomplete, because the real cost of slowness wasn’t just time - it was opportunity. The slowness actively prevented scientists from pursuing ideas, discouraged exploration, and shut down approaches that might have been valuable simply because the activation energy was too high.

That is why what we are seeing now feels fundamentally different, groundbreaking - possibly even revolutionary, though time will ultimately decide that. This isn’t just about doing the same things faster. Yes, it allows a drug discovery scientist to construct complex, multi-step pipelines - docking, molecular dynamics, machine learning–based ADME modeling, patent searching, shape-based similarity; almost every conceivable application involved in the process from a simple prompt in minutes. But the deeper shift is what happens once you no longer have to spend one or two weeks building that pipeline, because what you gain is not just speed, but time, and with that time comes the ability to explore.

Under the old model, a computational chemist typically worked on two projects, maybe three at most, and that wasn’t arbitrary since it reflected the sheer amount of effort required to get pipelines working for each target. In drug discovery everything is organized around targets - a cancer target, an Alzheimer’s target, and so on - and each one required its own setup, its own customization, and its own troubleshooting. Building those pipelines was an ordeal: installing software, gathering and standardizing files, cleaning and processing data, ensuring everything was in the right format, and then dealing with inevitable issues like input errors, parameterization problems, and tools that simply did not communicate with each other. These systems were never designed to integrate seamlessly, so passing output from one into another was often a project in itself, and the entire process was deeply manual, deeply frustrating, and quietly corrosive to any enthusiasm for exploring new ideas. Junior scientists especially find the activation barrier so high that they would often retreat to doing what was easy rather than what was important.

Now imagine that all of that friction is abstracted away, so that instead of wrestling with tools and infrastructure, the scientist simply specifies the intent and the desired outcome, and everything else is handled by the system. And that means every single thing. The system identifies or assembles the appropriate agents capable of achieving that intent, those agents select the right tools, databases, and models needed to deliver the outcome, and if the necessary agents do not already exist, the system recognizes this and automatically constructs them, often without being explicitly asked. And it manages error handling seamlessly, trying out various approaches to fix various errors until it finds the right one. It’s astonishing when you think about it, and I am seeing it unfold before my eyes every day. And all of this happens autonomously, with minimal to no human intervention, while still remaining highly customizable: if you want to impose constraints, select specific tools or databases, or guide the process with your own expertise, you can. In fact, the more expertise you bring, the more powerful the system becomes, but even without that expertise you are able to obtain results that are lightning fast and that would have been inconceivable just two or three years ago.

The implications of this are immediate and profound. A scientist who could realistically handle two projects can now handle ten, explore an order of magnitude more targets, move across multiple disease areas, and pursue lines of inquiry that would previously have been abandoned before they even began. And this is where the change becomes qualitative rather than merely quantitative, because increasing the number of targets, compounds, and disease areas is not just about efficiency, it is about dramatically increasing the number of meaningful attempts, the number of real opportunities to discover something that matters. It is, in the most literal sense of a phrase too frequently overused, more shots on goal.

But even that framing still undersells what is happening, because terms like workflow automation or productivity gains miss what is probably the most important ingredient in science, which is creativity. To see this clearly, consider an analogy: imagine you are a master chef trying to create extraordinary dishes, but you are constrained by everything that comes before cooking - procuring ingredients, preprocessing them, finding the right utensils, sourcing materials, and preparing everything correctly - and imagine that all of this takes one to two weeks. In that world, even the most talented chef would realistically produce only two dishes, not because of a lack of imagination, but because the overhead is so overwhelming.

Now imagine that entire process is accelerated tenfold, so that all of those preparatory steps are streamlined or automated, and suddenly instead of two dishes you can prepare ten. The difference between two and ten is not just quantity or the chef breathing easier, it is expanded creativity; because with ten dishes you can experiment, explore variations, take risks, and try ideas that would never have been worth the effort before. You are more willing to fail because the barrier to trying out things are easier, and we all know how important failure is to scientific success. That is exactly the shift we are seeing here. When a scientist can interrogate a hundred times more compounds, explore ten times more targets, and move across six disease areas instead of three, the result is not just increased throughput but a fundamental expansion of the space in which they can think and create.

And in drug discovery, that expanded creative space translates directly into outcomes: better ideas, more novel approaches, and ultimately more transformative drugs across cancer, Alzheimer’s, inflammation, diabetes, mental health, and beyond. This is why terms like workflow acceleration and productivity gains feel so inadequate, because they sound incremental, almost mundane, when in reality what is happening is that we are removing one of the biggest historical constraints on scientific creativity.

That is the core message I think we need to appreciate: this is not really about speed. Speed is the enabler, but creativity is the outcome. And creativity is the core of all groundbreaking science.