← BACK TO BLOG
BIOTECH

The Super Patient and Biotech's Structural Problem

Why miraculous one-off cures don't scale, and what that says about how biotech is built.

MARCH 02, 20267 MIN READ
BIOTECHITERATIONINSTITUTE DESIGN

Every few years, a cancer patient goes public with a story that is, frankly, miraculous.1 They threw everything at their disease — combination therapeutics, neoantigen vaccines, heavy surveillance, forward-deployed company staff running pathology the hospital couldn't handle — and they survived. You're glad for them and their family. The fact that we can do this at all — stack therapies, deploy custom vaccines, surveil a tumor in real time — is extraordinary.

But the super patient also reveals something uncomfortable about how biotech actually works. Or doesn't.

What the super patient teaches (and doesn't). The super patient generates data, sure. But it's uncontrolled, the interventions are stacked, and you still need trials to learn anything generalizable. A lot of the tactics are worth studying — forward-deploying specialists to hospitals, bespoke diagnostics, layered therapies — but they depend on a concentration of resources that just doesn't scale.

The deeper thing is about how learning works. Most biotech programs treat the clinic as a final, waterfalled, confirmatory step. If the data doesn't generate clear signals, the program dies. The super patient lives in a world of constant adjustment: try, measure, adjust, repeat. That iterative feedback loop is the opposite of how most companies operate — and the opposite of how they're funded.2

This plays out brutally at the platform level. Clinical trials are expensive enough that one asset failure gets over-extrapolated to the whole platform, even when the problem was target biology or trial design, not the underlying technology. CytomX is a good example: underwhelming results on their anti-PD-L1 probody CX-072 cast a shadow over all protease-masked drugs, regardless of whether the masking was actually the problem.3 One bad read on one program can set back an entire modality. This is actually pretty understandable given the costs involved — we are so constrained on real-world human data, especially publicly available clinical data, that everyone tends to overfit on whatever is there to see.

The rationing problem. We ration care extensively in the United States — in ways that are overt and in ways that are harder to see. Combination therapeutics, customized vaccines, heavy diagnostic surveillance — all of these butt heads with the market sizes, timelines, and reimbursement structures that biotech is built around. The system selects for single-asset plays with clean regulatory paths, not for the kind of multi-modal, deeply personalized approaches that actually seem to move the needle for individual patients.

And when the drug fails, the company fails. There's no marginal buyer stepping in to figure out why it failed. The data from dead companies is basically wasted.4 Someone should fix this — cleaner markets for data from failed programs, better incentives for rigorous data collection even in failure. I keep waiting for this to happen and it doesn't.

The voice of capital. People who haven't been through a biotech fundraise don't appreciate how directly investor quality shapes company strategy. Unlike tech, biotech has no revenue for a long time. When revenue does arrive, it's usually through pharma partnerships with carefully prescribed milestones — and big pharma is very experienced at pricing these deals so the small company gets almost no margin to work with. More importantly, the budget is usually precisely labeled during the fundraise — you raise $X for this indication, $Y for manufacturing, $Z for headcount — and you really have to stick to that. There's very little room to reallocate once the money is in.5

This makes investor quality existentially important.6 Tying in capital from inception — with real conviction and flexibility — lets founders make good allocation decisions without contorting themselves to look like a "Series A company" in two years. I don't think people appreciate how much mediocre venture capital pushes otherwise smart founders into (a) asset plays that will be made obsolete by international competition, (b) preclinical studies that don't actually derisk the asset, or (c) a flashy indication where the drug won't really shine.

The super patient model, whatever its scientific limits, at least demonstrates what happens when capital shuts up and lets the team allocate freely. That part is worth paying attention to.

Can anyone change this? Changing bad market structures is really hard — probably harder than curing any individual patient. Almost no companies in biotech history have actually redefined what the industry is supposed to be. Maybe Genentech.7 The CRISPR companies could still be the ones to do it — the technology is transformative enough to force the industry to reorganize around it, and the story is still early.

Regulatory shakeup is the most powerful lever, because it kicks the whole industry at once. But absent that, I think the best any individual biotech can do is: work on important areas, stay legible enough to raise at good valuations, and actually bother to understand which rules are real constraints versus which patterns people just naively slip into. The hard part is that a biotech's real strategy often gets locked in incredibly early in its life. Most of the consequential decisions happen before you have any data. But that's the game.

Footnotes

  1. The most recent and best-documented example is a tech founder who stepped down from his CEO role to go "founder mode" on his osteosarcoma — maximal diagnostics, single-cell sequencing, organoid models, 25+ terabytes of data, and public release of everything at osteosarc.com. Elliot Hershberg wrote a good piece on it: Going Founder Mode On Cancer, Century of Bio (2025). It's inspiring and I'm rooting for him. But it's also a case study in what's possible when one patient has essentially unlimited resources and total freedom to allocate them. ↩

  2. Ruxandra Teslo makes this point well — clinical trials should be a feedback loop, not a pass/fail gate, but the economics of drug development make it almost impossible to act on what you learn from a failed trial. See: Teslo, R. Clinic-in-the-Loop, Asimov Press (2026). I wrote about this more in MagBodies: Combining the Best of Surgery and Drugs. ↩

  3. CytomX's anti-PD-L1 probody CX-072 (pacmilimab) showed a 15% partial response rate as monotherapy in the PROCLAIM-CX-072 trial — not meaningfully different from standard anti-PD-(L)1 antibodies. The stock dropped ~16% on the data. The broader problem is that the probody platform got blamed for a result that could just as easily be explained by the target. See: The CytomX paradox, Evaluate Vantage; Deals mask CytomX's fundamental problem, ApexOnco. ↩

  4. Sun et al. found that ~90% of clinical drug candidates fail, with lack of efficacy and unmanageable toxicity accounting for the vast majority. The informational value of those failures is enormous — and almost entirely wasted. See: Sun et al., Why 90% of clinical drug development fails and how to improve it?, Acta Pharmaceutica Sinica B (2022). ↩

  5. This line-item budgeting has had some funny downstream effects. Biotech software companies have historically struggled to sell into biotech because companies literally didn't have a "software" budget — the money was labeled for reagents, synthesis, headcount, CRO fees. Synthego figured this out and successfully sold computational design services by bundling them into DNA synthesis orders, because companies did have a synthesis budget. The margin on the software was hidden inside the margin on the oligos. It's possible this is finally changing now that AI is hot enough that having "token spend" as a big line item in your financial statements might actually increase your valuation. If your budget looks like it's heavy on compute, investors pattern-match you to a tech company. Funny how that works. ↩

  6. Steve Holtzman wrote the definitive piece on this — how platform companies should think about capital structure, the role of non-dilutive partnerships, and why the quality of your early investors shapes your entire strategic trajectory. Millennium raised only $8.35M in VC and was ultimately acquired by Takeda for $9B. That only works when your investors let you play the long game. See: Holtzman, S.H. Early-Stage Biotech Value Creation: The Roles of Equity and Partnerships (2018). Hershberg and Malone wrote a good follow-up: On biotech platform strategy, Century of Bio (2023). ↩

  7. For the Genentech story and how they managed to redefine the industry, see: Hall, S.S. Invisible Frontiers: The Race to Synthesize a Human Gene, Oxford University Press (1987). ↩