Agentis: The Strategy Layer
Which clinical questions deserve an agent, and why
The top of the stack. Agentis takes a strategic R&D question, decomposes it into prioritized clinical AI use cases with a defined MVP and value, and hands the build to Navari with AIthena's data. By late 2025 it had 14 use cases running in parallel. It builds nothing itself; it decides what gets built.
The problem
The Prioritization Problem
An agent platform attracts demand faster than anyone can prioritize it. Without a strategy layer every function builds its own agents, the data team curates for whoever asks loudest, and the stack turns into a pile of point solutions. The clinical teams with the most valuable questions often had the least AI expertise to frame them.
- No shared answer to which use cases were worth building
- Data, models and agents planned by three different teams with three different roadmaps
- Clinical questions arriving without an MVP definition or a value case
- Prototypes that never became sustained capabilities
What I built
Decide What Gets Built
Agentis answers one question: what agents should we build and why? It prioritizes research questions from clinical development, defines the MVP and the value framework, and hands execution to the layers below. Agentis chooses the destination. AIthena supplies the data. Navari builds the agent.
- Decomposes a strategic question into candidate AI use cases
- Scores them on a value framework and keeps one prioritized backlog
- Defines the MVP and success criteria, then hands the build to Navari and the data request to AIthena
- Runs each use case through experiment, learn and adapt cycles until it is a sustained capability
How it fits together
Architecture
Agentis sits above the data and execution layers. A use case flows from a clinical question through prioritization, then down into AIthena for data and Navari for the agent, and back up as a measured capability.
Timeline
How it went
- 2025Framework defined: one strategy for data, models and agents
- 2025Prototype stage: three clinical studies delivered in six weeks
- Nov 2025Wave 1 live: 14 use cases in parallel
- Apr 2026Wave 1 in execution; ownership split across clinical vertical leads
- Jun 2026Handed off
What it taught me
Lessons
- Strategy without systems is a slide deck. Systems without strategy is a pile of point solutions.
- The hardest part was never building a capability. It was deciding which one, and saying no to the rest.
- One backlog, one value framework. Three teams with three roadmaps will not converge on their own.
- The stack only stays coherent while someone understands all three layers at once.
Stack
Connected work