Rigged, not broken

Most of the world's persistent problems aren't waiting on an invention. Water filtration, vaccines and insulin have existed for a century or close to it, and people still go without. About 2.1 billion people, one in four, lack safely managed drinking water.² Around 95,000 people died of measles in 2024, most of them unvaccinated children under five.³ Human insulin costs an estimated three to five dollars a vial to make.¹ The bottleneck is price, patents and permission. Not technology.

"Healthcare is broken" makes the system sound like a natural disaster, indifferent and nobody's fault. "Rigged" is more accurate: the friction pays someone. So when a startup sells "literacy" or "navigation," I read it as a tell. The industry underneath profits from being hard to navigate, and the startup has found a way to charge rent on that difficulty too.

I had a roughly three minute telehealth call to get a routine blood draw approved. It was billed at $330. Someone I know had a fifteen minute one billed at $500. Nobody invented anything during either call.

The bottleneck test

A company makes this series when two or more of these are true:

1. The solution already exists. There's a known, working fix (a drug, a test, a procedure, a public service) that people can't reach.
2. The barrier is price, patent or permission. Access is blocked by pricing, IP, a gatekeeping step (an order, a referral, a prior authorization) or deliberate complexity. Not missing technology.
3. It sells navigation of a maze it didn't build. Watch for "literacy," "navigation," "advocacy," "concierge," "appeals," "benefits clarity."
4. It gets paid by the friction. Revenue scales with how bad the system is. Fix the system and the company dies.
5. It sells to both sides. The same vendor serves the gatekeeper and the people trying to get past the gate.
6. "AI" is applied to the symptom. The pitch says AI will fix something whose real constraint is distribution or price.

Every autopsy in the series adds one question after the tell: what actually stands between people and the thing they need, and would this company exist if that were removed?

Three columns

Insulin 2

Tech fixes for problems that are already solved. What is this inventing that already exists, and what's stopping the existing version from reaching people?

The Literacy Tax

Companies that charge for helping people navigate complexity someone else profits from. Bill negotiation, insurance appeals, benefits navigation, financial "literacy."

The Arms Race

AI on both sides of the same gate. Insurers using algorithms to deny, startups using algorithms to appeal, vendors automating the phone calls in between. A machine eating its own tail with patients in the middle. You can see why I like this one.

Exhibit A: "AI will cure cancer"

The line goes: "Maybe AI should be used to cure cancer instead of making art." It's offered as the good version of AI, the use nobody could object to. In 2025 the head of Google DeepMind said AI could help end disease within about ten years.⁴

So say it works. Follow a hypothetical AI cancer drug through each gate and let the numbers answer.

Gate 1: Does it exist yet?

No. No AI-discovered drug of any kind had FDA approval as of mid-2026. AI-discovered molecules were about 1% of the clinical pipeline (67 of 6,147) in a 2024 count.⁵ The furthest along, Insilico's rentosertib, entered Phase III in July 2026, and it treats lung fibrosis, not cancer.^5,6 Isomorphic Labs, the company behind the "end all disease" pitch, raised $2.1B in May 2026, is focused on oncology, and as of September had given no timeline for its first human trials.⁷

Gate 2: Will it work?

Cancer is where drugs go to die. Of oncology drugs entering Phase I between 2000 and 2015, about 3.4% reached approval.⁸ AI has improved one part of this. AI-discovered molecules pass Phase I, the safety stage, 80 to 90% of the time against a historical 40 to 65%. But in Phase II, where a drug first has to show it works, they succeed about 40% of the time. Same as everything else.^5,9 So far AI is good at designing molecules that behave like drugs. It hasn't gotten better at designing ones that beat the disease.

Gate 3: Who will own it?

Isomorphic's partners are Eli Lilly, Novartis and Johnson & Johnson.⁷ If an AI-designed cancer drug gets through, it will be patented, licensed and priced the way those companies already price drugs. How a molecule was discovered doesn't change who holds the patent.

Gate 4: What will it cost?

The median annual list price of a new US drug was $300,000 in 2023, up 35% from $222,000 the year before, per a Reuters analysis of 47 approvals. One drug-pricing researcher put it plainly: many cancer and rare disease drugs "are not being priced relative to their benefit."¹⁰ Cheaper discovery doesn't mean cheaper drugs, because the price was never set by what the drug cost to make. Insulin is the proof. A few dollars a vial to produce, and still rationed in this country.¹

Gate 5: The control group

Look at the cancers we already know how to stop. Cervical cancer kills about 280,000 people a year, mostly in low- and middle-income countries. The WHO calls it "largely preventable through HPV vaccination and regular screening," and the vaccines cover the HPV types behind about 76% of cases.¹¹ Children with cancer survive at better than 80% in high-income countries and below 30% in many low- and middle-income ones, a gap the WHO puts largely down to "variable access to health products."¹² Same diseases, same treatments, opposite outcomes, decided by geography and money.

If more technology were the bottleneck, these numbers would look the same everywhere. They don't.

Honest counterweight

AI is a real tool in biology. AlphaFold won a Nobel, the Phase I gains are real, and an AI-discovered drug may well be approved in the next few years. I'm not saying AI is useless to science. I'm saying "use AI to cure cancer" assumes the thing standing between patients and survival is a missing molecule, and for most people who die of cancer today, it isn't.

The structural fixes exist too, and they're not glamorous. Civica Rx, a nonprofit drugmaker, launched insulin glargine with Biocon in January 2026 at no more than $55 out of pocket for five pens.¹³ The Open Insulin Project has been working on an open production protocol since 2015, though it has no approved product and tells people not to try making insulin at home.¹⁴

So

Whenever I hear "AI should be curing cancer, not making art," I want to ask a follow-up question. Say it works. Say an AI finds the molecule. Who do you think owns it? The median new drug in this country launched at $300,000 a year. We already have a vaccine that prevents most cervical cancer, and it still kills about 280,000 people a year in places the vaccine and screening don't reach. Kids with cancer survive at 80% here and under 30% in a lot of the world, with the same treatments. We don't have a discovery problem. We have a who-gets-it problem, and no model is being trained on that.

Next in the series: The Arms Race. An insurer's algorithm, a startup that appeals its denials for a fee, and a nonprofit that does the same thing for free.