AI-assisted design has moved from the lab’s edge to the center of how new biologic drugs get built. According to MIT Tech Review, companies like AstraZeneca are staffing up engineering teams specifically to push this further, treating computation as a core part of every step. “Everything we do, whether it’s design, make, test, or analyze, is now computationally enhanced,” says Puja Sapra, senior vice president and head of R&D biologics engineering and oncology targeted discovery at AstraZeneca. “The cycle times are getting shorter while productivity and innovation increase.”
What stands out here is the shift from AI as a helper tool to AI as the engine that decides where scientists spend their scarce lab hours.
The build-measure-learn loop
Sapra describes AstraZeneca’s approach as a tight feedback cycle. It works like this:
- AI generates or ranks candidate molecules computationally, predicting which designs are most likely to succeed.
- Scientists then commit lab resources only to the top-ranked candidates.
- Each result, win or lose, feeds back into the models.
The payoff is fewer dead ends and faster iteration. The number of possible molecular combinations is far larger than any human team could test by hand, so using AI to narrow the field has become central to biologics design. That matters now because it changes the economics of research. You stop burning money on molecules that were never going to work.
Going after harder targets
Speed is only half the story. The bigger prize is a new class of medicines. Traditional biologics usually hit one disease pathway. The next generation can hit two or three targets at once, or deliver a therapeutic payload straight to specific cells.
That kind of design means balancing many variables together: potency, stability, manufacturability, and safety. Sapra says AI models could help pick which targets to prioritize based on the biology, then optimize across all those parameters at once. “Drugging the undruggable is becoming a reality,” she says. “These technologies will eventually enable us to develop medicines against targets once thought impossible to reach.”
Why data is the real moat
McKinsey estimates that generative AI, paired with other computational tools, could cut drug discovery timelines by as much as 50%. But a model is only as good as what it learns from, and in drug discovery that means large volumes of high-quality biological data.
This is the part practitioners should pay attention to. Sapra calls data “our differentiator,” pointing to proprietary, multimodal datasets that include molecular structures, binding measurements, safety profiles, and manufacturing outcomes. AstraZeneca has also invested in deep screening technology to generate more of that data at volume. The lesson generalizes well beyond pharma: in specialized AI, the defensible advantage isn’t the model everyone can license. It’s the proprietary data nobody else has.
The lab of the future
To pull it all together, AstraZeneca is building a “lab of the future” in Kendall Square, Cambridge, where AI and robotics form a continuous, closed-loop discovery system. Sapra compares it to a self-driving car: AI makes predictions, robots run the experiments, instruments generate the data, and that data flows straight back into the models.
Humans stay in the loop by design. Scientists provide “the oversight, judgement, and strategic direction” that keep outputs explainable and pointed at patient benefit, Sapra says. Eventually these systems could make and evaluate thousands of molecular interactions every week, producing AI-ready data at a scale manual workflows can’t touch.
What to watch
For businesses and AI practitioners, three takeaways stand out:
- Own your data pipeline. The competitive edge is shifting from model access to proprietary, multimodal datasets you generate yourself.
- Design for closed loops. The real gains come when prediction, execution, and data collection feed each other automatically.
- Keep humans on judgment. The winning setups use AI for scale and people for direction, not one replacing the other.
Over the next one to three years, expect more of pharma’s R&D spend to move toward these automated discovery engines, and expect “undruggable” to slowly stop being a permanent label. The companies that get there first will be the ones sitting on the richest, cleanest data. More detail is available in the original MIT Tech Review report.