Isomorphic Labs, DeepMind's drug-discovery spin-out, raised a further $2.1 billion in 2026, taking the AI-native drug discovery bet to a scale that would have been unthinkable five years ago.[1]

What's more revealing than the number is who runs the place day to day. Demis Hassabis remains founder and CEO, but sitting either side of him, until recently, were two peer executives: Miles Congreve, a 30+ year medicinal chemistry veteran who co-invented Kisqali and has helped shepherd 29 drugs into clinical evaluation, as Chief Scientific Officer, and Sergei Yakneen, an engineer and machine-learning specialist, as Chief Technology Officer, currently serving as an advisor.[2]

Isomorphic Labs could have built its leadership team around the algorithm and treated chemistry as a service function underneath it. It didn't. It built a structure where a career medicinal chemist and a career computer scientist sit at the same level, reporting to the same CEO, neither formally subordinate to the other. That is a deliberate answer to a question most of the rest of the industry hasn't actually asked yet: when the machine and the scientist disagree about which molecule is worth advancing, whose judgement wins?

The Org Chart Hasn't Caught Up With the Hiring

Eleven of the industry's largest drugmakers have made an AI leadership hire of some kind, according to a 2026 census of pharma AI hiring.[3] Novartis brought in Sander Timmer as Global Head of Data Science and AI for Global Drug Development in Basel and is separately recruiting an Executive Director of Applied AI as part of a long-term push to become an "AI-first organization" with sponsorship at the highest levels of the company. Roche, through its Genentech unit, launched a Computational Sciences Center of Excellence (CSCoE) in 2025 that pools computational scientists from both organizations, backed by a combined on-premise and cloud compute footprint exceeding 3,500 graphics processing units (GPUs), which Genentech describes as the largest announced GPU footprint at any pharmaceutical company.[3]

Pfizer has built out a multi-tier AI executive structure under Jeremy Forman, who was promoted to Chief AI, Data & Analytics Officer, with two new vice-president-level roles, Head of AI Center of Excellence and Chief AI Architect, reporting to him. Pfizer is simultaneously hiring a Head of AI Clinical Excellence to lead AI application across the clinical development lifecycle. Eli Lilly and Nvidia announced a co-innovation AI lab that will co-locate Lilly's biology, science, and medicine experts with Nvidia's AI model builders and engineers in the San Francisco Bay Area, backed by up to $1 billion in combined talent, infrastructure, and compute investment over five years.[3]

AstraZeneca has created a dedicated Enterprise AI Unit as a global center of excellence spanning R&D, commercial, operations, and corporate functions, led by Chief Digital, Data and AI Officer Anne-Claire Gerbaldi, paired with a company-wide "100% workforce AI proficiency commitment" that frames AI competence as core to every role, not just specialist teams.[3]

What's notable is what these moves have in common structurally: nearly all of them sit outside, or alongside, the traditional R&D hierarchy, rather than inside it. The same 2026 analysis that catalogued these hires made a point of noting that "general restructuring and targeted AI hiring are distinct and largely non-overlapping trends" – companies are cutting R&D and operations headcount by the thousand (Novo Nordisk, 9,000 roles; Bayer, 13,500) at the same time as they build out parallel AI functions.[3]

That is not necessarily wrong. But it does mean that in most large organisations today, the group that decides whether an AI-generated hit is worth £2 million of medicinal chemistry time is not the same group, does not report through the same line, and was not hired by the same process, as the group that used to make that call alone.

A Task, Not a Department

I spend much of my working life placing the people who will sit on one side or the other of that line – Heads of Discovery Research, Heads of Computational Biology, Chief Scientific Officers, and the emerging cohort of "translational AI" leaders who don't fit neatly under either heading.

The candidates who interview well for these roles right now are, almost without exception, the ones who can articulate a specific, defensible answer to a specific, uncomfortable question: when your model's top-ranked candidate and your most experienced chemist's gut disagree, what actually happens next? Who has the standing to overrule whom, and on what evidence?

Too many organisations answer that question with an org chart rather than a process. They point to a title – Chief AI Officer, Head of Computational Biology – and assume the reporting line settles the argument. It doesn't. The Isomorphic Labs structure is instructive precisely because it refuses that shortcut: by making the CSO and CTO formal peers, it forces every genuine disagreement between the model and the chemist to be resolved on the evidence, in a room, rather than by whoever happens to sit higher on the chart.

The uncomfortable truth for a lot of boards is that this isn't really an AI hiring problem. It's a decision-rights problem that AI has simply made impossible to ignore. Discovery has always involved judgement calls under uncertainty – which lead to progress, which mechanism to trust, when to kill a programme.

What's changed is the volume and speed at which those calls now arrive, and the fact that one of the two parties in the room is a model rather than a person, which makes it strangely easier to defer to it by default, precisely when deferring is least appropriate.

What Boards Should Actually Be Asking

Before the next senior R&D hire, or the next AI leadership appointment, it is worth asking the question plainly rather than assuming the title answers it: who, specifically, has the authority to override a model's output, on what basis, and how is that decision logged and reviewed afterwards?

If nobody in the room can answer that without hesitating, the organisation doesn't yet have a discovery leadership structure fit for how discovery actually happens now – it has an AI hire and a discovery hire, coexisting, and hoping the question never comes up in a way that matters.

It will. The molecule that gets killed because the model was wrong, or advanced because the chemist was overruled and shouldn't have been, is not a hypothetical. The organisations that have quietly worked out who owns that call – not by title, but by design – are the ones whose discovery pipelines I'd want to be betting on.

Bottom Line

AI makes early drug discovery faster and cheaper, but as of July 2026 no AI-designed drug has received full FDA approval. AI supplements the wet lab rather than replacing it: every prediction still has to be tested experimentally.

Dr Andy Theodorou is a Partner at The RSA Group, specialising in R&D, Bioinformatics and Data-Science leadership search.

References

1. BioPharma Dive, "Cancer, immune drugmakers dominate VC funding so far in 2026."

2. Isomorphic Labs, "Isomorphic Labs announces first phase of management team."

3. IntuitionLabs, "Pharma AI Hiring Trends 2026: Top Drugmakers Census."