A working prototype. Claude turns each client's measurement based care history into a brief you can read between back to back sessions. Pick a client.
Safety, rules onlyNeeds attentionGoing wellEngagement
Written by Claude from PHQ-8, GAD-7, alliance and quality of life.
Every brief was checked three ways. Did it catch each signal planted in the data? Did it use a number or date that isn't in the record? Did it cross a line a brief should never cross: diagnosis, medication or risk? These are real Claude runs, captured once and replayed here.
A perfect score only means something if the checker can fail a brief. So I broke six on purpose.
The model never touches risk.
PHQ-9 item 9 never reaches Claude. Rules catch it and pin it to the top of the brief for the clinician to act on. Look at Grace: her brief is all good news, and the pin sits above it anyway.
Two Chairs already does the hard part. Measures come in at nearly every session, and the outcomes are why health plans sign. The open question is what the scores change between sessions. A therapist with a full day of back to back sessions has about a minute to look. So the first bet is small and specific: turn data you already collect into the one thing worth raising today.
Keeping risk with rules costs a flashier demo. It buys clinician trust, and clinician trust is the whole adoption curve. A brief that's wrong about a plateau is a bad minute. A brief that's wrong about risk ends the product.
Opened before the session most of the time, with a target the team sets in shadow mode. Rated useful by therapists most weeks. More plateaus and alliance dips raised in the session they appear. On the other side of the ledger: zero risk language from the model, zero invented numbers, and a clinician override rate we read every week.
An outside in roadmap, built from public information only. The first 90 days is where I'd find out what's wrong with it.
AI makes building cheap, so the scarce skill is judgment. I grow senior PMs toward principal by handing them a domain and a number to own across clinical, client and commercial, then getting out of their way.
One on ones with every PM and with my Engineering, Design and Data peers. Sit in on clinical ops and intake. Learn how a session actually runs before saying anything about it.
Write down where product is strong and where it needs attention. Agree on one planning rhythm with Engineering that both sides trust.
A strategy the team co-wrote, a roadmap story leadership can repeat, and one bet in flight with its numbers attached.
I'm Phil Morehead. I led a 60+ person product org at Launch by NTT Data and shipped a patient intake app to 36 hospitals on Epic and Meditech data. Today I'm the AI native builder at Sparq, where I took the firm's Claude certifications from 36 to 100 in under four weeks. I built this to show how I think, not just what I've done.