Key Takeaways

  • Annual drug approvals have stayed roughly flat for decades, even as scientific publications, budgets, and therapeutic modalities grew exponentially.
  • Max Jaderberg, President of Isomorphic Labs, attributes this slowdown to biopharma running each drug discovery effort as an isolated, bespoke science project.
  • The industry has exhausted the low-hanging fruit accessible through pure trial-and-error chemistry, forcing teams toward computational design.
  • A shared foundation model enables learnings from one therapeutic campaign to immediately improve predictions for the next disease target.

Pharma Treats Every Drug as a One-Off Experiment

Biotech spends more money every year to produce roughly the same output. Scientific papers multiply, venture funding surges, and new modalities like mRNA and cell therapy enter the clinic. Yet the FDA approval line remains stubbornly horizontal. Elad Gil points out the core contradiction: “If you look at the number of new drug approvals per year, it's roughly flat. But that's despite an explosion in knowledge.” (play from 5:39)

Jaderberg explains that the traditional playbook hit a physical ceiling. For a century, pharmaceutical research relied on empirical screening: synthesize thousands of chemical compounds, test them in wet labs or animal models, and hope one hits the right target without poisoning the patient. That method worked when the easiest biological mechanisms remained undiscovered. As Jaderberg notes, “the lowhanging fruit of medicines that can be found with our empirical process of designing medicines are being found.” (play from 8:32)

When a traditional team finishes a drug program, win or lose, they pack up the data into a siloed filing cabinet. The medicinal chemists move on, the assays sit in static spreadsheets, and the next team begins from scratch. Jaderberg describes how biopharma operates: “a lot of the techniques and the way we go after medicines or the industry goes after medicines is really treating each one as like a segmented, you know, it is like a bespoke frontier science project.” (play from 10:29)

Because each project is an island, the marginal cost and time required to discover drug number 100 is almost identical to drug number one. No cumulative software advantage emerges.

Foundation Models Turn Biology Into Compounding Software

Isomorphic Labs approaches the problem by treating biology as an information processing challenge rather than an endless series of wet-lab bets. By training deep learning models on structural biology, molecular dynamics, and synthetic human tissue data, the system shifts discovery into predictive code.

The strategic unlock is not just designing a single molecule faster. The unlock is feedback loops. In traditional labs, failed assays are pure waste. In a platform model, every physical measurement feeds directly back into the neural network, refining its internal map of protein folding and binding affinities.

Jaderberg defines the central question for modern drug platforms: “What do you learn from that that helps you work on the next medicine much quicker and then the next medicine even quicker after that?” (play from 10:12) When the platform sits beneath every campaign, he argues, “if you have this platform underneath going after these medicines that allow you to really suck up and understand the learnings and help impact your core foundational understanding of disease biology and how to design good molecules and exquisitly engineer these medicines, then you can hope to actually get compounding improvements the more medicines you go after.” (play from 10:59)

This shift turns drug development from an artisanal craft into an engineering discipline governed by data scaling laws. Instead of relying on crude animal studies that fail to replicate human physiology, teams can run millions of in silico simulations, testing toxicity and specificity before ordering physical synthesis.

What to Do With This

Audit your product and operational architecture this week to spot where your company behaves like an old-school biopharma shop. If your team treats client deliveries, sales deals, or engineering sprints as isolated projects without automated data capture feeding back into your central systems, you are throwing away your compounding advantage. Instrument a shared data store or evaluation set so every project automatically improves the tools used on the next one.