The operating backbone
for early-stage pharma.
Not a chatbot. Not a single-purpose app. An opinionated substrate that holds a programme’s state, knows the domains, applies the right mode, produces the deliverables, and leaves a defensible trail behind every move.
It does not replace Veeva or Benchling. It orchestrates them.
Speed is bought in operations, design and CMC in parallel.
- The consultant decksPaid for twice, and they leave when the engagement does.
- The ad-hoc Google DocsTrue when they were written. Nobody knows if they still are.
- The Slack decisionsMade, agreed, and unfindable by Thursday.
- The tribal knowledgeWalks out the door when someone quits.
The difference between what we know and what we need to know.
The difference between what we want people to do and what they understand we want.
The difference between what we expect our actions to achieve and what actually happens.
Discovery is a data problem. AI is all over it.
Development is a different animal. You cannot get a drug approved with data alone, because the cost, the risk and the timeline are set by forces that do not yield to algorithms:
- Cross-functional specialists who speak different professional languages
- Work that lives in documents, models and decks, never built to be read by a machine
- Exceptions that exist only in the heads of experienced practitioners
- Coordination overhead that degrades everything downstream of it
- Regulatory positions that are argued, not computed
Point solutions attack the first gap. Programmes die on the other two.
The rules are in the guidance. The exceptions are in people’s heads.
The negotiated precedents, the unwritten regulatory norms, the study design choices that look counterintuitive and work - none of it is in any training set.
As agents absorb the act of producing development artifacts, the valuable skill becomes judgment: whether the output is scientifically sound, regulatorily defensible and strategically aligned. That judgment cannot be generated by the AI. It has to be encoded into the system that governs it.
Every schedule you run models delay as proportional. Drug development mostly waits - on a suite that runs every six weeks, on a queue with a median, on a deadline that does not move. So a slip is free until it isn’t, and then it costs a cycle.
Most people want to select a mode, not adjust twenty dials. Leadership sets one operating posture for the programme - speed, capital efficiency, regulatory de-risking, partnership readiness - and every function executes against it.
A mode change is a state change. The same fact comes out material under one and noise under another, computed rather than asserted.