From Consumer Tech to Curing Disease: Why TPMs Belong in Science
After years shipping consumer products at Google, I moved to the Chan Zuckerberg Biohub to lead technical program management for AI research. Friends asked the same question: what does a product-launch person do in a research organization? More than I expected — and less, in ways that matter.
Science and consumer tech run on different physics. In product, you can define done, decompose it, and march. In research, the destination is discovered along the way — a negative result can be a triumph, and a pivot can be the entire point. Importing product-style execution wholesale into science isn't just ineffective; it can actively damage the thing that makes research work.
And yet. Modern science increasingly runs on engineered systems — frontier AI models, petabyte-scale data pipelines, distributed GPU training clusters, open tools used by thousands of labs. Those systems have roadmaps, dependencies, infrastructure costs, and users. They need to ship. In May 2026, our team at the Biohub launched a world model of protein biology — ESMC, ESMFold2, and the ESM Atlas with 6.8 billion proteins and 1.1 billion predicted structures — openly, to the global scientific community. Getting there took every ounce of launch discipline I learned in consumer tech. This is the seam where technical program management belongs.
What transfers
Making dependencies visible. A frontier model launch has supply chains as real as any phone launch: dataset procurement feeding model training, GPU cluster capacity gating experiments, evaluation gating release. Scientists shouldn't have to hold that graph in their heads. That's our job — ours ran on real-time dashboards tracking model milestones, GPU utilization, and pipeline health across distributed teams.
Protecting focus. The scarcest resource in any research organization is uninterrupted scientific attention. A great TPM is a heat shield — absorbing coordination overhead, sequencing asks, and making sure a principal investigator's week is spent on science rather than logistics.
Turning ambition into milestones without lying. "Cure, prevent, or manage all disease by the end of the century" is a mission, not a plan. The craft is decomposing audacious goals into fundable, staffable programs while being honest about uncertainty — building plans that hold direction firmly and details loosely.
What doesn't transfer
Velocity worship, for one. In consumer tech, faster is almost always better. In science, speed that compromises reproducibility is negative progress. I've had to retrain my instincts: the question isn't "how do we go faster?" but "where is speed actually the constraint, and where is rigor?"
In product, the plan is a commitment. In science, the plan is a hypothesis. Program management has to hold it accordingly.
Status culture also translates poorly. Red/yellow/green implies a known destination. For open-ended research, I've found it more honest to track learning velocity: what did we find out this quarter, and what decisions did it unlock?
Why this matters beyond the Biohub
The biggest scientific opportunities of the next decades — AI-driven biology, large shared datasets, open tooling — are engineering-intensive by nature. The organizations that figure out how to pair scientific freedom with operational excellence will simply learn faster than those that don't. That pairing is a program management problem, and it's one of the most meaningful ones our discipline has ever been offered.
If you're a TPM in consumer tech wondering whether your skills matter beyond engagement metrics: they do. Science needs people who know how to ship. It just also needs the humility to learn why, sometimes, it shouldn't ship yet.