Clinical development is full of ambiguity, trade-offs, political dynamics, and knowledge that never gets written down.
So Cori is AI-guided. It authors the documents and surfaces the decisions, and the team decides. The system provides the speed; the people keep the judgement.
Where the chaos actually comes from
The chaos in clinical development comes from fragmented information. The people are not the problem. Documents live in different places. Prior decisions get buried. Two people reference different versions of the same file without knowing it. Small misalignments compound into costly delays.
That is the coordination Cori is built for: helping a team work from one base of structured, verified information. Cori flags contradictions and surfaces missing context, and it brings past decisions back into view when they matter again.
It does not make the calls. When the trade-off is participant burden against data quality, or speed against regulatory risk, that takes human judgement. Strategy and precedent do not follow clean rules, and neither does politics.
And much of what matters is not written down anywhere. A method you remember working once before, a site you have heard struggles with participant engagement, a reviewer who prefers data laid out a particular way. Software cannot intuit any of that. What it can do is keep the written record in one place and flag where it has drifted, so your judgement stands on solid ground.
What AI brings to the surface
What AI can do is bring the key decisions to the surface, so clinicians, regulators and operators decide with the context in front of them instead of hunting for it. The biggest risks in clinical development come from blind spots: trade-offs made without context, inconsistencies nobody noticed, assumptions nobody stated.
Take the protocol. Authoring one involves hundreds of interdependent decisions about eligibility criteria, endpoints, assessments and regulatory strategy. AI can surface relevant past trials, suggest accepted standards, and flag gaps and inconsistencies. It cannot decide which trade-offs suit a specific therapeutic area, or how far a sponsor's risk appetite stretches. It can help you decide. It cannot decide for you.
From tool to teammate
This split has already been tested in diagnosis. In a randomised trial published in npj Digital Medicine, clinicians and a customised GPT each made an independent diagnostic assessment, and the AI then combined the two, marking where they agreed, where they differed, and commenting on each. Clinicians working this way reached 85% diagnostic accuracy when the AI gave the first opinion and 82% when it gave the second, against 75% for clinicians using conventional resources.
The fine print matters. The AI on its own scored highest of all, at 87%. But the cases were clinical vignettes, short structured reports of a patient's history and findings, the kind a student doctor meets in an exam. Real patients and their problems are messier.
That is where human expertise still holds the edge. In radiology, the pattern reported is that AI performs well on large-scale image analysis while humans remain better on the edge cases involving ambiguity, context or prior experience. The reliable gains come from working with AI as a collaborator rather than a replacement.
How Cori works with you
The workflow we are building follows the same shape. Cori does not start generating a document the moment you ask for one. It works out what the document needs, then checks that against what you actually have.
Where something required is missing, Cori flags the gap and prompts you for the additional information before drafting. Once the gaps are closed, Cori authors the document and your team reviews it and decides.
That back and forth is the point. The system moves fast precisely because it stops to ask.
The judgement stays with your team
Humans stay in the loop because there is a great deal they do better than machines. The useful question is how AI and people work together to improve decisions, and through them, outcomes for participants.
Let AI handle the organising and the repetitive work, so people spend their time where their experience counts: the strategic and political calls, and the messy real-world edge cases.
AI should fit into the way your team already works.