We promised at the outset to keep the main text product-agnostic, and to say plainly, at the end, what we are building and why. This is that page.
River Records is the company we founded on the ideas in this book. Stream is the product: an AI documentation platform for clinicians, organized — as you might by now expect — by medical problem rather than by encounter date.
The origin was not a technology looking for a use. It was the research you read in Part II. When our duplication study showed half the record to be copies, the conclusion we drew was not that clinicians needed scolding but that they needed persistence — and that nobody was going to give it to them inside the note paradigm. So we built outside it.
Concretely, Stream is our version of the bridge described in Chapter 14, and we try to hold it to this book’s own standards:
- Problem threads. Every piece of documentation attaches to the problem it concerns, automatically, so each condition’s story — assessments, plans, medication changes, relevant results — reads as one thread across visits rather than a pile of snapshots.
- Structure at creation. The ambient scribe produces structured documentation — subjective and objective context, then assessment and plan split out per problem — because every downstream capability depends on that foundation.
- Micro-updates and visible deltas. Update what changed; see what changed. No re-telling the whole story each visit, and no guessing what’s new in a colleague’s documentation.
- What carries forward. Visit-specific context stays with the visit; problem-level documentation carries. The distinction from Chapter 11, implemented.
- Tasks born from plans. Follow-ups written in the assessment become tracked commitments without a second entry, and resurface when due.
- Summarization as a scoped tool. We use AI summarization where it is the right tool — the story of one problem over years, on request — and refuse it as a default, because compression loses exactly the details clinical work runs on.
- Pre-visit orientation. A huddle view that answers the real pre-visit questions — what’s active, what did I say last time, what’s falling off, what’s still open — from structure, not from paraphrase.
Stream works alongside the EHR rather than replacing it — which means, per Chapter 14, it is a bridge, with a bridge’s honest costs. The destination remains a system of record in which documentation, structure, and the chart are the same thing. We intend to keep walking toward it, and we built the bridge to be walked on by others in the meantime.
Two commitments, in closing.
First: the ideas in this book do not depend on our success. If a better team builds the workspace first — or an incumbent opens the integration that makes bridges unnecessary — medicine wins, and this book will have done its job. The argument was always product-agnostic because it had to be: paradigms should be judged on evidence, not on who profits from them.
Second: we will keep publishing. The research program that produced the duplication study and the noteless-EMR feasibility work continues, and the honest accounting — including the costs and failure modes of our own approach — belongs in the open literature, where colleagues can check it.
If you’ve read this far, you are probably the kind of clinician or builder this whole effort depends on. Come argue with us. The record won’t fix itself.
— Jake & the River Records team
riverrecords.ai