Building zOS — my own AI operating system, in daily use
The situation
Running a portfolio of ventures and advisory work alongside life, at high tempo — more context than one person can hold, more decisions than one person can gate by hand. A single AI chatbot doesn’t solve it; an undisciplined fleet of them is worse.
What I did
- Built zOS: a fleet of context-walled AI roles, each in its own room, each shipping real work while I stay the human who gates it.
- Designed it verify-not-promise: roles prove state before claiming it — no fabricated success, real failures surfaced plainly.
- Stood up a values layer, a priority stack, and a review cadence so the system stays honest as it scales.
The outcome
- [TEO: confirm specific outcome line — e.g. roles/agents in daily use across N contexts]
- [TEO: confirm specific outcome line — e.g. hours/week reclaimed or throughput multiple]
- [TEO: confirm specific outcome line — e.g. error/escape rate, or what the gate has caught]
I dogfood exactly what I sell — the operating model I bring to clients is the one I run on myself every day.
Form Health
SVP Product & Technology (CPTO) · 2024–2026 · obesity and cardiometabolic care
The situation
A B2C obesity-care company that needed to scale, lift margin, and put AI to work in a clinical setting — without breaking the care model.
What I did
- Led the business-model pivot from B2C to employer B2B2C, navigating payer and employer contracting and compliance.
- Reorganized Product / Design / Eng / Data 3× (to ~25+) into four autonomous units.
- Put AI into the clinical workflow: record review, scribe, summaries, inbound-call handling.
The outcome
- Scaled ARR and patients 15× in two years while doubling gross margin past 50%.
- 50% reduction in clinical visit time; patient-access lead time cut 10× (months to days).
- ~50% of inbound calls handled by AI; 30% less clinician time on messaging.
Strategy first, then AI — the model pivot is what made the AI gains bankable.
UnitedHealth / Rally Health · Optum Digital
Senior Director of Product · 2020–2022 · Fortune-10 health plan
The situation
A national health plan needed a coherent, personalized digital member experience and a virtual-care strategy at scale.
What I did
- Re-envisioned the personalized member experience.
- Defined virtual-care strategy and shipped an end-to-end Virtual Care Hub.
- Launched UHC’s Virtual-First Health Plan (NavigateNOW) and a new Virtual Primary Care offering.
The outcome
- Member experience reaching ~4M monthly active users.
- Virtual Care Hub shipped to ~40M members across Optum Care and partners (Teladoc, Amwell, Doctor On Demand).
- Launches covered by Businesswire, CNBC, and others.
At payer scale, the win is making one coherent experience out of many vendors.
Sana Benefits
VP Product & Engineering · 2022–2024 · self-funded health insurance for SMBs
The situation
An SMB self-funded insurer needing growth, margin, and a new in-house care offering.
What I did
- Grew the SMB & broker solution; rearchitected care navigation and provider search.
- Launched Sana Care, a new Virtual Primary Care practice.
- Built product-led-growth infrastructure with journey tracking.
The outcome
- Grew the solution 50%+ (Q3’22–Q3’23), gross margin +20 points (to 60%+), R&D spend −40%+.
- Managed benefits/enrollment for 2,000+ businesses, 30,000+ members.
- Sana Care: 1,000 patients and break-even within four months; NPS doubled to 50+.
A new care line can pay for itself fast if the unit economics are designed in from day one.
Collective Health
Director of Product · 2014–2020 · health benefits administration platform
The situation
An early-stage benefits-administration platform scaling from zero.
What I did
- Early product hire; helped scale 0→1.
- Shipped provider search, HIPAA-compliant messaging, claims transparency, a personalized member hub, and a mobile app.
The outcome
- 30→500+ employees, 0→60+ employer clients, 0→250k members.
- 0→USD 40M+ ARR on a platform supporting USD 1B+/yr in healthcare spend.
- A study linked the model to 45% lower employer total cost (JAMA Network Open).
The 0→1 product decisions compound — get the data model and the member hub right early.