Care Pathways Agent - Design Time Experience
Care pathways encode decades of clinical expertise. Designing the experience that lets informaticists turn that expertise into a trustworthy, executable AI agent — without writing a line of code.
Context: Care pathways are evidence-based clinical protocols — step-by-step decision guides that tell care teams what to do when a patient presents with a specific condition, like pneumonia or sepsis. This institutional knowledge is painstakingly developed, highly specific, and almost always lives in a PDF. SAB is the no-code platform that turns those PDFs into executable AI agents.
Estimated Impact At a Glance
< 60 minutes to author a pathway agent with provenance
Estimated Time to Deploy
< 1 day from artifact to testable agent
The Problem
Getting a care pathway into a clinical AI agent requires significant engineering involvement today — translating branching clinical logic into structured, executable workflows that an agent can reason over. SAB removes that barrier. The design challenge: make the process trustworthy enough for a clinical informaticist (a clinical IT professional who bridges clinical practice and technology) to own themselves.
My Role
Product thinking lead: No dedicated PM
Cross-functional: PM, engineering, data science
The insight that shaped the interaction model
Watching a senior nurse teach a junior nurse to chart — narrating observations naturally, the way one colleague talks to another. The agent needed to fit that mental model.
The Solution
An informaticist uploads a care pathway PDF. SAB processes it and presents a structured summary — purpose, entry criteria, core steps, current defaults — in the conversation panel. From there, the informaticist can review the generated flowchart on the Canvas, view recommended order sets, or return to the source document. The PDF stays accessible throughout.
Key Design Decisions
Making AI-generated clinical logic reviewable
When a PDF becomes a flowchart, the informaticist faces a question they've never had before: did the AI get my institution's protocol right? We designed for trust at each step — structured summary before the flowchart, source document always accessible, and a design concept for future phases where tapping any node traces it back to the exact source content that generated it. Every design decision in this flow answers the same question: how does the informaticist know the AI got it right?
The recommendation model: turning text into executable clinical actions
Phase 1 flowcharts reference order sets as text labels — the agent can read them but can't act on them. Phase 2 makes them executable: the agent can initiate an order set when a patient reaches the relevant node in the pathway. Rather than asking informaticists to manually locate and link every order set, the AI recommends them in the conversation panel, explains its node-placement reasoning, and auto-links on acceptance. The same Canvas+Conversation principle that works at the bedside on mobile works in a clinical authoring environment on desktop.
What's Next
Phase 2 completes the executable pathway. Phase 3 brings the test suite, performance metrics, and reviewer workflow. A platform-level test framework — covering structural, logical, and patient data validation across all SAB agent types — is a dedicated portfolio piece in progress.