Genomics

I came to this through a podcast, not a laboratory

The five theses before this one describe technology reaching outward, and each of them eventually runs into a physical place that’s hard to get to. This one turns inward, where the hard-to-reach place is a cell.

I should be honest about where this one started, because it’s the thesis where I know least.

I became interested in genomics listening to Cathie Wood talk about it. What I took away was simple enough to state in a sentence: instead of testing medications until one works, you could look at a person’s DNA and pick the right treatment first. And further out, you might correct some diseases at the level of the biology causing them rather than managing the symptoms forever.

That was the whole of my understanding. No genetics background, no medical training, and a small position in ARKG rather than any view on individual companies.

Researching this page changed that starting belief in two directions. Part of it turned out to be more real than I expected, with treatments already approved and working. Part of it turned out to be constrained by something nobody mentions in podcasts. This page is mostly about the second part, because it’s where the investment case actually lives.

The part I had roughly right

Medicine really is becoming less trial-and-error, and cancer is where you can see it most clearly.

The old sequence was diagnosis, then standard treatment, then wait and see whether it works. Increasingly the sequence starts by measuring the biology: sequencing the tumour, identifying which mutation is driving it, and choosing a drug aimed at that specific mechanism. A companion diagnostic tells you whether a patient is likely to respond before they’re given the drug. Patients whose tumours lack the relevant target don’t get put through a treatment that was never going to help them.

That’s a genuine change in how medicine is practised, and it’s the thing I was gesturing at. Reading biology has also become extraordinarily cheap. Sequencing a human genome went from a multi-year international project costing billions to something a hospital can order as a test.

But the cost of generating the data fell far faster than the cost of interpreting it. A sequence is only valuable if it changes a decision, and for many common diseases a genome still doesn’t produce a simple treatment answer. Single-gene disorders can be genetically tractable, because one variant does most of the causal work. Alzheimer’s, cardiovascular disease, diabetes and psychiatric illness are also genetic, but differently: many variants each contributing a little, interacting with environment, age, lifestyle and biological processes that change over the course of the disease. Genetics matters enormously there without pointing at a switch to flip.

So I’d revise my original belief rather than abandon it. Medicine is becoming more precise, and the precision comes from measuring disease biology, not from reading DNA and receiving an answer.

What “cure disease” actually looks like

When I was asked what attracted me to genomics, my answer was two words: cure disease. I want to replace that with something accurate, because the accurate version is still remarkable.

Two therapies for sickle cell disease were approved in late 2023. Casgevy, developed by Vertex and CRISPR Therapeutics, is the first approved treatment using CRISPR gene editing. Neither it nor its competitor Lyfgenia repairs the mutation that causes sickle cell. Casgevy edits a patient’s own blood stem cells to change how fetal haemoglobin is regulated, so the body produces a form of haemoglobin that doesn’t sickle. That detail matters more than it sounds: even the first successful CRISPR therapy works through a biological workaround rather than a simple correction of the broken gene.

In their pivotal trials, the large majority of evaluable patients were free of severe vaso-occlusive crises over the assessed follow-up period, with reported figures in the high eighties to low nineties as a percentage. Those are the episodes of blocked blood vessels that cause severe pain and cumulative organ damage.

For a disease that previously offered symptom management and a bone marrow transplant if you were lucky enough to have a matched donor, that is close to transformational. But the language matters. These are described as potentially curative rather than curative, because the long-term follow-up doesn’t exist yet and regulators require years of continued monitoring. This isn’t one dose and done.

Then in April 2026 came the result I found most significant. Intellia reported Phase 3 results for lonvo-z in hereditary angioedema, which the company described as the first Phase 3 success for in vivo gene editing, meaning the editing happens inside the patient rather than to cells removed and returned. A single infusion reduced mean monthly attacks by 87% against placebo. More striking to me, 62% of patients who received it were entirely free of both attacks and ongoing therapy across the six-month evaluation period, against 11% on placebo. The most common side effects were infusion-related reactions, headache and fatigue, all reported as mild or moderate. The results were published in the New England Journal of Medicine in June 2026 and the company has begun a rolling submission to the FDA.

That’s the direction I was interested in, demonstrated in a controlled trial. Two things keep it in proportion. The efficacy period assessed was six months, which is short for a permanent genomic change, so durability is still an open question. And it’s worth noting what it treats and how: a rare condition affecting a small population, by switching off a gene in the liver.

The constraint nobody mentioned to me

Here’s what changed my understanding most, and it isn’t in any of the promotional material.

Look at what’s approved or most clinically advanced in gene editing, and a pattern appears immediately. The programmes have clustered heavily in blood and liver. Casgevy edits blood stem cells outside the body. Intellia’s therapy targets the liver. Most of what sits closest to approval falls into one of those two categories.

That isn’t a statement about which diseases matter most. It’s a map of where delivery is comparatively tractable.

Blood cells can be removed, edited in a laboratory and returned, which sidesteps the problem of delivering the editor to the right tissue inside the body. The liver is reachable because the lipid nanoparticles carrying the editing machinery naturally accumulate there. Other tissues, the brain, heart, muscle and lungs, are much harder, because getting the tools into the right cells in sufficient quantity without provoking an immune response is the constraint. A 2026 review from Jennifer Doudna’s group made the point directly: the difficulty of tissue-specific delivery currently limits the applications of editing technology.

Viral vectors, the alternative route, have their own limits. An AAV vector carries roughly 4.7 kilobases of cargo, which complicates delivery of the larger and more precise editing systems and often requires workarounds such as splitting the payload or using more compact editors. Many people also carry pre-existing immunity to the viral shells, which constrains both who can be treated and whether a dose can be repeated.

I want to be careful not to overstate this. Delivery is not unsolved. It works reliably in the contexts above, and vector engineering keeps improving. What’s accurate is that it remains one of the field’s largest constraints, particularly outside currently accessible tissues.

So the honest framing of this thesis is that reading biology got easy and reaching it didn’t. We can sequence almost anything and design an edit with real precision. What we largely cannot yet do is deliver that edit into most of the tissues where disease happens.

That’s not a reason to dismiss the field. It’s a reason to know what you’re betting on. If delivery broadens, an enormous set of diseases becomes addressable. If it doesn’t, this stays a powerful technology for blood disorders, liver conditions and a set of rare genetic diseases, which is medically important and commercially much smaller.

The number that reframed everything for me

Casgevy works. The trial results are not in dispute. It was approved in late 2023.

Through the first quarter of 2026, more than 500 patients had begun the treatment process, out of roughly 60,000 estimated eligible worldwide.

That distinction matters. Initiating treatment means starting a months-long sequence of consultations, cell collection, manufacturing and conditioning. It is not the same as completing it. Vertex reported 64 patients receiving infusions across the whole of 2025, and reporting in early 2026 put the total number of people treated anywhere in the world since approval at roughly sixty.

Sixty people, in two and a half years, for a therapy that works.

The reasons are mundane and instructive. The US list price is around $2.2 million, though what payers actually pay under negotiated arrangements is lower and not public. Treatment requires months of coordinated care at a small number of specialist centres. And before receiving the edited cells, patients undergo an intensive chemotherapy conditioning regimen to clear their existing bone marrow, which carries real risks of its own. For someone weighing that against a disease they have managed their whole life, it is not an obvious decision.

Access has been widening rather than narrowing, which makes the gap more striking rather than less. Around 90% of eligible US patients now have reimbursed coverage, the therapy is approved in a number of countries outside the US, and in July 2026 the FDA expanded the label to children as young as two, having previously restricted it to twelve and over. So the eligible population is growing while the number of people actually treated stays in the dozens.

I keep returning to this because it corrects the instinct that a working therapy becomes a large business. The science was the part that succeeded. Manufacturing, specialist capacity, reimbursement, treatment burden and patient willingness are what determine whether it reaches anyone.

Curing something once is commercially awkward

This is the counterargument I take most seriously, and it has nothing to do with science.

A one-time potentially curative therapy compresses a lifetime of clinical value into a single reimbursement event. That has to cover the research, the manufacturing, the failures along the way and a return, all at once, which is why these prices reach into the millions. Insurers and health systems then absorb an enormous immediate cost against benefits that accrue over decades, possibly after the patient has changed insurer entirely.

I don’t want to overstate it in the other direction. Chronic medications aren’t automatically better businesses. They face patent expiry, generic competition, patients who stop taking them and shifting standards of care. But their value arrives gradually, which the payment system was built for, and a single multi-million-dollar payment is not.

The system is adapting. In the US, Medicaid runs an outcomes-based arrangement for the sickle cell therapies in which reimbursement is linked to whether agreed clinical results are achieved, with a large majority of states participating. That’s a sensible response and it’s early.

Meanwhile bluebird bio, which developed Lyfgenia and spent years bringing genuinely groundbreaking science to approval, was acquired by Carlyle and SK Capital in mid-2025 and taken private. Shareholders could elect $5.00 per share in cash, or $3.00 plus a contingent right to a further payment if the product portfolio reaches a sales milestone by the end of 2027. The board’s own filings stated that without the deal the company was at significant risk of defaulting on its loans, and that shareholders would likely receive nothing in a bankruptcy.

A company can deliver a medical breakthrough and still produce a disastrous investment outcome. That isn’t hypothetical here. It happened to one of the two companies that achieved what this entire field was aiming at.

My own position is the strongest argument against my own thesis

I own ARKG, and it isn’t a large position. I want to be straightforward about how it has behaved, because it’s the most honest evidence on this page.

ARKG launched in late 2014. Across the decade since, a period in which the science advanced enormously, the first CRISPR therapy was approved, sequencing costs collapsed and in vivo editing reached Phase 3, the fund’s long-term return has substantially lagged the broad US equity market. It has also been extraordinarily volatile: up around 180% in 2020, down more than 50% in 2022, and up sharply again in the year to mid-2026. Its assets fell from roughly $9 billion at the 2021 peak to around $1 billion, reflecting both weak performance and investors withdrawing money rather than performance alone.

I’d rather describe that pattern than quote a precise since-inception figure, because the number moves a great deal depending on the end date, and a fund that can rise 90% in a year can’t be summarised honestly with one statistic. The shape is what matters. This has been a rollercoaster rather than a compounder, during exactly the decade when the underlying science delivered.

That is the “great technology, poor investment” problem sitting in my own portfolio. I’ve written some version of that distinction into all five previous pages. This is the one where I can point at my own holding and show it happening.

It also raises a fair question about the basket approach. Owning thirty-five companies in a field where most fail means owning the failures alongside the winners, and if the winners are rare enough, the arithmetic doesn’t rescue you. The counterargument is that I have no ability to identify which company survives, and concentrating without that ability is worse. I hold the basket because of my own ignorance, not because I think it’s optimal.

There’s a related question I can’t yet answer, and it may be the most important one for how I’d eventually size this. Biotech companies present themselves as platforms, meaning a technology that can generate many therapies rather than one. If that’s true, a company that succeeds once becomes more likely to succeed again, and the economics start to resemble a technology business. If it isn’t, these companies remain collections of individual drug programmes, each carrying substantial independent clinical risk, and one success tells you less about the next than the word “platform” implies.

Casgevy is the test case and the verdict isn’t in. CRISPR Therapeutics proved its editing approach works in humans, which is not nothing. Whether that translates into a second and third approved therapy, in different tissues, is exactly what the next few years will show. Platform is an investment claim that has to be earned through repeatability, and until it is, I’d treat it as a hypothesis rather than a reason to concentrate.

Where AI fits, briefly

T·01 covers intelligence. I won’t repeat it, and I’m sceptical of the way “AI plus biotech” gets sold.

Better tools for predicting protein structure, identifying targets and interpreting genetic variants are genuinely useful, and interpretation is exactly where the bottleneck sat in my earlier section. But the industry’s problem has never mainly been generating hypotheses. It’s that most drug candidates fail in human trials for reasons that weren’t visible beforehand.

AI may compress parts of discovery and development, including trial design and patient selection. It does not remove the need for clinical validation, and it does not solve physical delivery.

I’d treat it as an accelerator on parts of a long process, not as the thing that changes the economics.

What would make me wrong, and what would make me more confident

The thesis fails if delivery stays where it is. If gene editing remains effectively limited to blood and liver for another decade, this becomes a valuable set of treatments for a narrow group of diseases rather than a change in how medicine works.

It also fails, in investment terms, if the pattern I described above keeps repeating: therapies that work, reach very few patients slowly, and produce poor returns for the companies that developed them. Great medicine and poor economics can coexist indefinitely.

The other risks are the ordinary ones for this industry and they’re severe. Most drug candidates fail. Safety problems in gene editing would set the field back years. Competition can compress pricing on the therapies that work. And large pharmaceutical companies may acquire successful technologies at prices that capture most of the value before public investors see it.

Survival probability matters more here than in any thesis I’ve written. A company with no revenue, burning cash and repeatedly issuing shares to fund the next three years may never reach the point where its science gets tested properly. Scientific promise doesn’t help a shareholder whose stake has been diluted to a fraction of what it was, or whose company was sold at $5 a share because the alternative was bankruptcy.

What would raise my conviction has two halves, and I need both.

The scientific half: delivery to tissues beyond blood and liver, demonstrated in patients rather than animals; durable efficacy and safety over real follow-up periods; and evidence that a platform genuinely produces repeatable therapies rather than one asset that happened to work.

The commercial half: treatment protocols that don’t require months of specialist care and conditioning chemotherapy; reimbursement that functions at scale; approved therapies reaching a meaningful share of eligible patients; and companies earning attractive economics without diluting shareholders to survive the wait.

The science scaling is not sufficient. This whole page is an argument that it isn’t. If a second or third gene therapy reaches thousands of patients rather than hundreds while the company selling it earns a decent return, the machinery is working. If the therapies keep working and the businesses keep struggling, I’ll have been right about the medicine and wrong about the investment.

The hardest place to reach

What I found, turning inward, is that the inward problem is harder. Not because we can’t understand biology. We’re getting remarkably good at reading it. Because a cell is a far more difficult place to reach than a data centre, and the body defends itself against exactly the sort of intervention we’re trying to make.

I hold a small position because the direction seems right and my ability to pick the winners is close to zero. If delivery broadens and the therapies that follow can reach patients at scale with workable economics, I’d expect to want more exposure than I currently have, and I’d expect to have far better information about where to put it. Until then, small is the honest size.