On March 29, Eli Lilly — one of the world’s largest pharmaceutical companies — agreed to pay up to $2.75 billion for a pipeline of drugs designed entirely by artificial intelligence. The deal with Insilico Medicine, a Hong Kong-based AI drug discovery company, includes $115 million upfront and milestone payments and royalties that could reach the full $2.75 billion if the drugs progress through clinical trials and reach the market.

This is not a research collaboration. Lilly is licensing actual drug candidates that Insilico’s generative AI designed from scratch, with the intent of taking them through human trials and commercialisation.

What Insilico actually built

Insilico Medicine uses its Pharma.AI platform to do something that previously required entire departments of PhD-level scientists: design novel drug molecules. Not screen existing compounds against known targets. Design new molecules from the ground up, starting from a biological target and generating molecular structures the system predicts will bind effectively, be safe, and be manufacturable.

The results are real. Insilico has produced 28 AI-designed drug candidates. Nearly half are already in clinical trials. Lilly’s deal gives it exclusive worldwide rights to develop, manufacture, and sell the drugs from Insilico’s existing preclinical pipeline, plus the option to collaborate on new programs.

One of those candidates is a GLP-1 oral drug — a category of enormous commercial interest given the success of drugs like Ozempic and Mounjaro. Lilly, the maker of Mounjaro, clearly sees strategic value in an AI-generated oral alternative.

This is the moment the argument changes

For years, the question in AI adoption has been: does it just automate simple tasks, or can it do genuinely expert work?

Drug discovery is about as expert as work gets. It requires deep knowledge of biology, chemistry, protein structure, clinical pharmacology, and regulatory science. It takes hundreds of researchers, years of work, and hundreds of millions of dollars to get a single drug candidate to trials under traditional methods. Failure rates are high.

Insilico cut through that with AI-generated candidates that a company like Lilly considers worth $2.75 billion of commercial risk.

The implication is not that AI will replace pharmaceutical researchers overnight. It is that the category of “work that requires deep human expertise” is now much smaller than it was. AI is not just augmenting expert work. In specific domains, it is producing the output.

What This Means for Business

You probably do not run a pharmaceutical company. But the question this deal raises applies to your business directly.

Where in your company does expert knowledge create the most value? Where do you have bottlenecks because you need a specialist to produce something, and specialists are expensive and hard to find?

Legal research. Financial analysis. Engineering design. Market research. Technical writing. Risk assessment. Every knowledge-intensive function in your business has a version of the same question: how much of what our experts produce is replicable by an AI trained on domain-specific data?

Lilly’s bet says that in at least one highly regulated, highly technical field, the answer is: enough to pay $2.75 billion for it.

For most businesses, the starting point is simpler than drug discovery. AI agents can handle research tasks, draft complex documents, synthesise data across systems, and produce specialist-quality work product in a fraction of the time. The Insilico deal is proof that the ceiling is much higher than most business owners have assumed.

Three things worth doing this week:

1. Identify the top three knowledge bottlenecks in your business — places where specialist expertise is the constraint on speed or output.
2. Ask whether any of those bottlenecks involve tasks that are pattern-based at their core, even if they require expertise to do well. Pattern-based expert work is the first to become automatable.
3. Look at what AI tools already exist for your industry’s domain. What was science fiction three years ago is production software today.

The businesses that start thinking about AI at this level now will have a meaningful head start on the ones that are still treating it as a productivity shortcut.

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.