The word buried in the ARPA-H announcement is "artisanal."

It appears in the description of what GEMMABio and Profluent Bio have been funded to replace. The agency wants AI-designed, highly modular gene editors that promise, in its words, ultimate scalability, "as opposed to current bespoke and artisanal editors." That is a federal agency describing the state of the art in personalized gene editing as craft work.

It is also, read closely, a description of what happened to KJ Muldoon.

What it actually took

KJ was born in August 2024 with severe CPS1 deficiency, a urea cycle disorder that leaves the body unable to clear ammonia. Roughly half of babies with the neonatal-onset form do not survive infancy. He was too young for a liver transplant, which was the only real option on the table.

So a team at the Children's Hospital of Philadelphia and Penn built him a drug. They sequenced his genome, found two truncating variants in CPS1, designed a base editor and a guide RNA to correct one of them, tested it in animals, manufactured it, and got it into him. The whole thing, from diagnosis to first infusion, took about six months. The case was published in the New England Journal of Medicine in May 2025, and it is a genuinely extraordinary piece of medicine. KJ went home in the summer of 2025, eating more protein and needing less medication to control his ammonia.

Now the parts that did not make the headlines.

It took more than 45 scientists and doctors, plus pro bono work from several biotech companies, with Danaher-owned Aldevron and Integrated DNA Technologies manufacturing the therapy. Kiran Musunuru, who led the design work, said he could not estimate what it cost in time and effort. Rebecca Ahrens-Nicklas, who led the clinical effort, said the drug will probably never be used again. It was built for one boy's specific mutation and is useless to anyone else.

Musunuru's rough guess was that custom editing treatments might eventually land somewhere near the cost of a liver transplant, around $800,000, not counting lifelong care.

That is the thing ARPA-H is actually reacting to. Not the triumph. The unrepeatability.

The bottleneck was never the editing

Here is the detail that reframes the whole program. The scientific part of KJ's treatment was fast. Musunuru's team worked out how to correct one of his variants within a few weeks.

The other five months went to everything that is not science. Animal toxicology. Clinical-grade manufacturing. The regulatory submission. Assembling a coalition of institutions willing to eat the cost. The FDA, to its credit, moved in about a week once asked, because a baby was dying and the agency treated it accordingly.

So the rate-limiting step in personalized gene editing is not CRISPR. It is the scaffolding around CRISPR: tox packages, GMP supply, INDs, and the fact that every single one of these is currently rebuilt from scratch for every patient. Base editing worked in 2016. The problem is that the delivery system for it, in the institutional sense rather than the molecular one, is a hand-built one-off every time.

This is what makes the THRIVE design more interesting than the dollar figure.

The umbrella is the invention

ARPA-H's announcement sets three gates. In year one, teams must show gene editing platforms that can generate multiple drug products sharing common biodistribution and toxicology profiles. By year three, they must have started first-in-human trials in which a single trial accommodates several different products and several different disease phenotypes. By year five, they must expand that umbrella investigational new drug application to more products and more diseases.

Read the year-one gate again, because it is the whole ballgame. Common tox profiles across multiple products means you stop doing a fresh toxicology package for every child. Read the year-three gate, and you stop writing a fresh IND for every child.

Program manager Daria Fedyukina described it as a modular building-block approach that swaps one key piece instead of rebuilding the whole set for every disease. ARPA-H director Alicia Jackson framed it around testing multiple treatments for multiple diseases in a single clinical trial.

Neither of them is describing a scientific advance. They are describing a manufacturing and regulatory architecture. THRIVE is a bet that the marginal cost of the fiftieth bespoke editor can be driven toward zero if the fixed costs are paid once and shared, and that the thing standing between rare disease patients and treatment is paperwork and pipework rather than biology.

That is, I think, correct. It is also why the program is exposed in a way that a purely scientific bet would not be.

The dependency ARPA-H cannot fund

An umbrella IND requires the FDA to accept an umbrella IND.

ARPA-H says clinical and regulatory innovations are mandatory in the program, which is the agency acknowledging the problem without being able to solve it. It can fund seven teams. It cannot fund a regulatory pathway into existence, and it does not control the agency that would have to build one.

The KJ precedent is thinner here than it looks. FDA moved in a week for a dying infant under a one-off compassionate use request. That is a well-worn mode for the agency and it involves accepting enormous uncertainty because the alternative is a dead baby. A platform IND covering multiple products for multiple non-emergent phenotypes is a structurally different ask. It requires the FDA to agree, in advance and in general, that a class of drugs sharing a delivery vehicle and an editing chassis can be regulated as variations on a theme rather than as new molecular entities.

There are reasons to think the agency is open to it. There are also reasons it might not be, and none of them are unreasonable. Every guide RNA has its own off-target profile. St. Jude's award explicitly includes developing the most sensitive off-target detection methods available, which tells you the field knows this is the crux. Modularity is a promise about the parts that stay the same. Safety questions live in the part that changes.

About the money

$160 million, up to, over five years, across seven teams, milestone-contingent. The Broad Institute's slice is reportedly up to $34.5 million.

Some researchers on the receiving end have said quietly that this is less than they expected. They are not wrong to notice. A single conventional gene therapy program routinely burns through more than any of these teams will receive, and each of these teams is expected to build a platform, run tox, manufacture clinical material, and open a first-in-human trial within three years.

But the size of the check may be the wrong thing to argue about. ARPA-H is not trying to fund seven cures. It is trying to buy a template and then, per the announcement, have the teams teach everyone else how to copy it. The line about frequent publications and demonstrations, so that rare disease colleagues across the country can replicate the path, is the actual theory of change. If it works, the value is not in the seven programs. It is in the fiftieth team that does this without ARPA-H money because the trail has been cut.

If it fails, $160 million buys seven excellent papers and a small number of treated children, which is not nothing, and is also not what was promised.

The scale problem does not go away

About 95% of rare diseases have no approved therapy. Tens of millions of Americans live with one. THRIVE covers seven teams working on metabolic and blood disorders, inborn errors of immunity, bone marrow failure, pediatric epilepsies, genetic heart disease, rare vascular diseases, and epidermolysis bullosa.

Even total success across all seven touches a rounding error of the problem. That is not a criticism. It is the nature of a platform bet. You do not fix 7,000 diseases by treating them. You fix them by making treatment cheap enough that someone else can afford to bother.

Which brings it back to the uncomfortable thing sitting under the celebration of KJ Muldoon. Forty-five people and several donated corporate supply chains rallied to save one child, and it was beautiful, and it is not a healthcare system. It is a charity sprint that happened to have the right people in the room and the right hospital and enough press attention to make the pro bono work worth it.

The next KJ will not necessarily have any of that.

ARPA-H's wager is that the answer to a miracle is not more miracles. It is a factory.

Further reading