AI Can't Deliver in Oncology Without Governed Data Infrastructure

I’ve spent nearly a decade watching life sciences organizations accumulate data they cannot fully unlock. Siloed by therapeutic areas, fragmented across incompatible tools, and inaccessible to the teams who need it most. Such a pattern was consistent in biopharma then, and sadly I still hear it a lot today. At AACR last week, I heard them again, in sharper detail, now that AI has raised the stakes for everyone.

Here are the three things worth carrying out of San Diego.

1. Life Sciences Organizations Still Struggle with Data Management. And Have for Years

Here's what made the conversations at AACR land differently for me: I've been watching this exact problem for almost a decade.

Large organizations, fragmented therapeutic areas, each accumulating rich data assets and each struggling with the same three things. Data siloed within business units, with no shared visibility across the organization. Different tools and methodologies making data harmonization a slow, manual process. And organizational friction where different groups inside the same company were trying to download and work on the same underlying data without any coordinated infrastructure to support it.

Over the past few years, I've watched the same pattern spread far beyond biopharma to health centers trying to improve patient outcomes, to organizations sitting on valuable data assets they want to commercialize but can't scalably link and surface for partners to consume. The problem that used to live primarily inside large pharma organizations exists everywhere in life sciences.

The gap these institutions are missing isn't access to better models. You can deploy the most powerful AI to accelerate research or commercialize data, but if the underlying data is fragmented across systems and can't be accessed in a coherent, governed way, the fundamental problems remain. Conversely, you can invest years in data infrastructure, de-silo everything, and still end up with a data lake no one actually uses because the analytical tools aren't there to meet researchers in their natural scientific language.

Both things have to move together. A de-siloing strategy that doesn't bring AI-native analysis tools into the workflow doesn't deliver on the promise. And an AI strategy that doesn't first solve data readiness is just an expensive chat interface. The organizations that break through will be the ones who stop treating these as separate workstreams and find a platform that advances both hand-in-hand.

2. Cancer Centers Need a Complete Solution, Not Just AI

The most clarifying conversations of the week were with the cancer centers I met in person. While I went in expecting to hear about AI ambitions, I heard a more honest account of where things actually stand.

These organizations have been pitched a parade of AI solutions. The pitches are polished, and the demos are compelling. But none of them address what actually needs to be solved first: centralizing and linking multimodal data so that AI has something real to work with. The integration layer underneath connecting clinical, genomic, imaging, and real-world data is still the unsolved problem. Most solutions pitch the intelligence layer without building the foundation it requires.

What struck me is that the cancer research centers already understand this. They just haven't found anyone willing to do the foundational work. And they've rightly concluded that they can't, and shouldn't, try to build the agent platform themselves. Their real advantage lies in their translational science and domain expertise. The platform is someone else's job to build well. That's ours.

Several of the groups I spoke with were already familiar with the work we're doing at peer institutions like University of Virginia Comprehensive Cancer Center and Indiana University Melvin and Bren Simon Comprehensive Cancer Center. This recognition has meaningfully shaped how Manifold is perceived, not as just another AI solution, but as a trusted partner that does the hard, foundational work others won't. The proof points from our existing cancer center partnerships are opening doors and setting us apart in the space.

3. The Agents Are Doing Real Work

Last week, I had the opportunity to engage in deep conversations with two of our customers and the feedback from both reinforced exactly what we're building at Manifold.

I heard firsthand how our AI agents are performing on real data and real workflows, helping teams unlock insights that have long been buried in their data. The signal from both was consistent: the agents are doing meaningful work. They're compressing months of manual effort spent stitching together siloed, multimodal data into days. They're closing the translation gap between biology and data by enabling scientists to query in natural language and get answers in real time. And importantly, they're freeing up time and resources so teams can focus on what they do best—advancing science—while Manifold continues to provide the connective data and AI infrastructure that helps them continuously innovate.

This kind of feedback, concrete and unsolicited, from people with every reason to be candid is the signal that matters most.

Where That Leaves Me

I left San Diego energized. AI is rapidly transforming how we approach this challenge that has persisted in life sciences for decades. We've entered an era where AI capabilities are increasingly accessible. However, the organizations advancing are those that recognize the real advantage isn't the AI itself. It's the infrastructure beneath it: a connective layer, a harness, that integrates siloed data sources, enforces security and governance, and puts powerful tools directly in the hands of scientists.

But I'm also clear-eyed, in a way that comes from having watched this same problem go unsolved across different corners of the industry for years. There's still a lot of hard, necessary work between where most organizations are today and where they need to be. Building the infrastructure that closes the gap between expert intent and technical execution, between data assets and data utility, is exactly where Manifold plays. It's what drew me here. And it's what we've been building toward for the past four years.

The urgency I felt coming out of those conversations makes one thing obvious: the time to close that gap is now.

Reach out directly or book time with our team.