The Missing Object in Drug Development
July 23, 2026
3 minute read
Why drug development needs a new kind of software - and what Cheiron is building.
Preface
A drug can take up to fifteen years and anywhere from two to ten billion dollars to reach a patient. A rational expectation is that most of this time, effort, and money is spent on science. It would, in a sense, be tragic if any substantial portion of these resources was spent on activities like file tracking.
Our internal and external narrative is frequently that these timelines are due to the complexity and the stubbornness of biology — and biology is stubborn. But sit inside a program for a quarter and you'll notice something stranger than nature. The molecule and the biology are rarely the core bottleneck. In a complex, regulated environment, the most expensive bottleneck is a company that has lost the thread of its own work — the lineage between the data it generated, the promises and decisions it made, and the conclusions it's standing on today.
A purification step changes in the lab. Six months later, that change must echo through eight documents, two regulatory commitments, and three active protocols. You are incredibly lucky if today the person who could still answer this one question "What else does this touch?" - is still with the company. The unglamorous and painful cross-referencing and reconciliation happen manually, late, under pressure, and - frequently - incompletely. This is the industry's technical debt; except the interest here, in this business, is paid in patient time. When the FDA puts the trial on hold, the clock everyone pretends is about science becomes about filing cabinets and frequently preventable mistakes.
Three-quarters of clinical trials require substantial amendments. Nearly half of those amendments are avoidable. The average amendment cycle now runs 260 days. Biology did not write those amendments. People did — to fix what other people didn't write down, notice, or check. Why? One reason - the people doing this work have never had software built for the work they actually do.
The thesis of this memo is therefore simple: the missing software object in drug development is the drug program itself.
A village of maps
Walk through any drug program and you'll find the same archaeology. Protocols. INDs. Tox reports. CMC packages. SAPs. Meeting minutes. Decks. Emails. Spreadsheets that someone's lead statistician has been quietly maintaining since 2019.
Each of these precious, expensive, and consequential artifacts is necessary. But none of them is the program.
The program is the thing between them — the living web of claims, evidence, commitments, and consequences that decides whether a molecule moves forward or quietly dies. It exists. Everyone working on it can feel it and that is what the CEO of your favorite biotech company dreams about. It is the main topic of conversation in board rooms and across company meetings. Everyone has a role in it. But it has no home in software.
So we stuff it into documents. I invite you to think of a document as a photograph. It is a snapshot that tells you what was true when the shutter clicked. It cannot tell you what has changed since, what depends on it, what now contradicts it, and what consequences a flaw you notice in a photograph taken a year ago has for you now.
A drug program is not a photograph. A drug program is an organism.
This is why traditional biopharma operates in the realm of very expensive rituals - the meetings are an awkward implementation of organizational memory; and many calls effectively execute synchronization and alignment. That is ok. The most senior people on the program become the village elders — scheduled in fifteen-minute increments to confirm the map is there and consulted before any decision.
This is usually mistaken for a people problem. It really is not. This company and its talented employees are carrying an object model that should have been built in software — a living, cross-functional, evidence-bound system running on what they remember, what they found, filed, and who happened to be in the room.
AI makes it worse first
The obvious first wave of AI in pharma is faster writing. Faster protocols, faster CSRs, faster regulatory responses. Appealing as a useful starting point for the industry, velocity of generation alone was never going to be enough. The bottleneck was never how fast people could write. It was that nothing could help hold the writing together once it was written - and scrutinized.
Because once writing is cheap, the bottleneck moves. The scarce thing stops being text. The scarce thing becomes trusted coherence.
Anyone — anyone, including our favorite grandma your CFO at midnight armed with a laptop and a vague prompt — is or soon will be able to generate a plausible protocol section using modern tooling. The hard question is whether that section is true for this program, at this moment, given this evidence, these prior commitments, this safety profile, this manufacturing reality.
Plausible is now free. Coherent almost never is.
More drafts, more summaries, more near-final artifacts means more surface area to validate, not less. AI doesn't remove the need for judgment. It raises the premium on systems that can point judgment where it is actually needed.
A Drug Program as a software object
A software object is something a computer holds the meaning of, not the bytes. A bank account is an object. The pixels are not the account; the account is a position the system defends against every transaction that touches it. You cannot overdraw it by editing a PDF. It knows what it is.
We believe a drug program should be that kind of thing.
Today it is the opposite: a position no system holds, scattered across heads, files, and the social memory of whoever was in the room. Ask the program a question and the program cannot answer. You assemble the answer by interviewing the villages. It is exhausting to the villagers and its leadership, and it's dangerous, because the complexity of this industry has long overcome the capabilities of the most capable of teams.
We believe we can help.
In our view, a prized and trained clinician should not spend hours checking whether the exclusion criteria still match the safety narrative. A regulatory lead should not manually reconcile every prior agency comment against every new draft. A CMC expert should not have to wonder whether a process change has silently invalidated a clinical assumption downstream.
Experts should make judgments. Software should hold the state of the world those judgments depend on. It should enable drug program state management as a collective, shared process and goal that is user-friendly, workflow-oriented, compliant from a regulatory perspective, and speaks the language of verified ground truth.
Drug development has systems of record, systems of workflow, systems of document generation. It has never really had a system of truth for the program itself.
That is the object Cheiron is building.
