Where Chief AI stands

An honest read, as of Oct 3, 2026. The experience has come together. The engine exists separately. The two halves are not connected yet.

Has it come together?

The stories have. Rosa and Leon show one connected journey feeding every touchpoint, persona by persona, with safety first and a "why" behind every card. It isn't working intelligence yet.

Experience layer · built chief-ai.js · Visitly + Programly

Toggle on/off inside the real apps
Physician · Nurse · Care-manager workflows
Why? panel: V1 (how Chief decided) / V2 (Intelligence Report)
Reads the host app's real safety alerts
All content is scripted: 2 hero patients written in depth, the rest generic

Engine · vertical slice cs-cocarely/chief-intelligence (Worker)

Event ingestion, state, open loops: tested
Evidence grounding + sub-agent dispatch: tested
create_task / update_goal with verification: tested in memory
Anthropic wiring: contract-tested, never called live
D1 storage untested, not deployed

The missing bridge: the overlay's cards are hand-written; the engine's outputs aren't shown anywhere. The next real step is making one overlay card (e.g. Rosa's pre-visit brief) come from the engine's context packet + a live Claude call, then measuring it against the scripted version.

Platform building blocks

Shown in UI (simulated)Engine code, testedNot started

All use cases

Every Chief touchpoint across the family, in one format: who uses it, the moment, what Chief needs, what it produces, and who decides. Use this as the backlog. Built = clickable in the overlay today. Engine = exists in the CoCarely Worker. Planned = not built.

Same capability, many moments

The PRD's core idea: build each capability once in Chief, then reuse it in every product. Each row is one capability; each column shows where it appears.

● Built (overlay)◆ Engine○ Planned

Chief AI reference · synthetic examples · the overlay itself lives in app/chief-ai/chief-ai.js (Visitly + Programly)