Pre-institutional infrastructure for aging

The infrastructure aging forgot.

Aging care runs on a layer the health system does not see, does not measure, and does not pay. One in three family caregivers is at extreme intensity right now. Seventy percent of healthcare staff are themselves unpaid caregivers for someone at home. Forty percent of patient outcomes are shaped by people with no badge, no shift, and no record in the EHR.

That is the vital sign nobody is measuring. Vimty is the argument — and the build — for the infrastructure that measures it, owns it, and compounds it.

1 in 3
family caregivers experiencing extreme intensity
70%
of healthcare staff are also unpaid caregivers at home
40%
of patient outcomes shaped by unpaid caregivers

Figures drawn from ARCHANGELS caregiver-intensity research, founded by Alexandra Drane. Her frame — the caregiver as vital sign most organizations are not measuring — is the argument under this page.

Seven layers · what you actually buy

A coherent organization of layers to humanity.

Vimty is your read access into a network of physician-governed agents running across seven layers of every member's life. You don't pay for a dashboard. You pay for the attested observation feed produced by people actually being cared for.

Layer 1 · Arrival
Members enter the network
Risk-stratified intake from 12 condition funnels + cooperative onboarding. Cohort assembly happens here.
Layer 2 · Belonging
One identity across 41 surfaces
Apple Wallet ID + LCA member share + cooperative governance vote. Same identity for every observation in the feed — no duplicate-record problem.
Layer 3 · Daily flow
Sage runs in the background
Persistent context, cross-LLM portable. Every interaction adds to the longitudinal record without form-fatigue.
Layer 4 · Meaning
Family timeline + values clarification
Advance directives, care goals, multi-generational story. The narrative scaffolding that makes Layer 6 humane instead of transactional.
Layer 5 · Body your feed
Physician-attested SDOH observations
This is what you actually buy. Every observation hashcare-signed, ICD-coded, time-stamped, member-linked. Risk-adjustment-ready, audit-defensible. PMPM + per-care-gap + uplift share.
Layer 6 · Care production engine
Worker-owned cooperative caregivers
The same caregiver shift that delivers the care also produces the observation. The same shift pays twice — patient pays $19/mo membership, you pay PMPM for the resulting SDOH feed. Closed loop, low CAC.
Layer 7 · Mind
Compounding personal context
Member's own AI-grounded context, owned by them, signed by us. Members stay because their context compounds. Retention becomes mechanical.

Brands are doors; layers are the rooms. You buy Layer 5 output, produced by Layer 6 operations, sustained by Layers 1–4 + 7 infrastructure. Every layer is agentic. Every observation is physician-attested. Every member is sovereign-owned. The same shift pays twice.

The unmeasured layer

Aging is held together by people the system does not count.

The family caregiver is the single most important variable in whether a discharge holds, a medication is taken, a fall is prevented, a hospital readmission is avoided. None of that caregiving shows up on a claim, a chart, or a board-level dashboard.

It does show up — in strain, in absenteeism, in cost shifted to emergency departments, and in outcomes the health system takes credit or blame for without knowing the real cause. ARCHANGELS calls the caregiver the vital sign most organizations are not measuring. That framing is correct. The implication is that the industry is running blind on the variable that matters most.

When we talk about aging infrastructure, we do not mean hospitals, senior-living towers, or another app. We mean the human layer that is already doing the work. It needs a record, a structure, and a way to compound.

Why current fixes fail

The two dominant answers — the gig economy, and the chatbot — are each structurally broken.

1. The gig model cannot retain the caregiver.

U.S. home care runs on ~85% annual caregiver turnover (industry baseline reported by Sequoia’s 2026 services analysis). Matching a family to a worker who will churn inside the year is not infrastructure. It is a fee on catastrophe.

Honor (Sequoia-backed) has applied AI to scheduling preferences and brought churn down to the mid-30s. That is meaningful. It is also still gig — the caregiver is a contractor in somebody else’s platform, and the improvement is priced in to the platform, not the worker.

2. The unsupervised chatbot cannot be trusted with medicine.

A 2026 wave of peer-reviewed studies is converging on the same conclusion, in four journals in the space of months: chatbots cannot be the patient-facing clinical interface without a physician in the loop.

BMJ (2026)
Generative AI chatbots and medical misinformation — accuracy, referencing, readability audit
Five popular chatbots tested on 10 medical questions. Nearly half of responses were highly problematic. Outputs consistently expressed with confidence and certainty, filled with hallucinations and fabricated citations.
JAMA Network Open (2026)
Large Language Model Performance on Clinical Reasoning Tasks
21 frontier models across 29 clinical questions. Conclusion: current LLMs remain limited in early diagnostic reasoning and cannot yet be relied on for unsupervised patient-facing clinical decision-making.
Nature Medicine (2026)
Reliability of LLMs as medical assistants for the general public
Randomized, preregistered study. LLMs in lay-public hands identified the relevant condition in fewer than 34.5% of cases — no better than a control group. Patients could not guide the model to the right questions.
Nature Medicine (2026)
ChatGPT Health performance in a structured test of triage recommendations
Among gold-standard emergencies, the system undertriaged 52% of cases, with inconsistent activation of crisis safeguards. Safety concerns warrant prospective validation before consumer-scale deployment.
I warned three years ago that these systems were “purveyors of authoritative bullshit” that should not be trusted. That is still true — and it very much applies in medicine. — Gary Marcus, April 2026, revisiting a 2023 warning in light of four new peer-reviewed studies

The gig economy churns the worker. The unsupervised chatbot hallucinates the plan. Neither has the structural shape the aging population needs.

The shift under all of this

Services are the new software.

Julien Bek of Sequoia framed the shift cleanly in March 2026. The next wave of AI winners will sell outcomes, not tools. Pure software gets replicated by incumbents, underpriced, or built in-house. The durable companies are full-stack — they own the care model, the patient relationship, the clinical data, and the operations. AI embedded across that stack reinforces the service, instead of being the service.

A copilot sells the tool. An autopilot sells the work. — Julien Bek, Sequoia Capital, “Services: The New Software” (March 5, 2026)

In healthcare the shift is already moving. Honor, Ro, Akido, Everlywell, Aledade, Virta, Origin, Cityblock — eight services companies cataloged in Sequoia’s follow-up analysis, all with AI embedded across a care-delivery stack they own end-to-end. Each one is proof that the margin profile of a services business changes when AI takes real cost out of care delivery.

Vimty’s argument is one step past Sequoia’s: if services are the new software, then the ownership structure of the service is the moat. Capital-owned services compound for the fund. Worker-owned services compound for the worker, the family, and the community around them.

Community leverage = the moat

Make the worker an owner. Make the community the customer of last resort. Then embed the AI.

The home-care-specific proof already exists. Cooperative Home Care Associates (CHCA) in the Bronx has been running the model Vimty is arguing for since 1985 — 2,500+ caregiver worker-owners, ~20% annual turnover in an industry that runs at 50-85%, wages ~20% above regional benchmarks, and ~80% of profit distributed back to the worker-owners who generated it. Forty years. One borough. Still the largest worker cooperative in the United States.

The Mondragon Corporation is the broader proof the worker-owned model scales. Founded in the Basque Country in 1956, now a federation of cooperatives across 35 countries, generating €11.05B in 2023 with 70,500 worker-owners. Independent data (Co-ops UK) show cooperatives survive at roughly 80% over five years vs. 41% for conventional small businesses. Worker ownership is not idealism. It is a survival characteristic.

2,500+
CHCA caregiver worker-owners in the Bronx. Forty years of proof in home care specifically.
~20%
CHCA annual caregiver turnover. Industry baseline for non-cooperative home care is 50-85%.
80% / 41%
Co-op 5-year survival vs. conventional small-business survival (Co-ops UK, independent data).
§1042
U.S. tax-code mechanism letting a founder sell to a worker-owned cooperative and defer capital gains. A quiet, legal exit path.

What this looks like in aging care.

The gig stack
Caregiver is a contractor. Data is the platform’s. Equity is the fund’s.
  • 85% annual caregiver turnover (industry baseline)
  • Family rebuilds the relationship every year
  • Care data sits in a vendor portal and evaporates on churn
  • Upside accrues to the platform and capital stack
  • AI is added to keep contractors from leaving
The cooperative stack
Caregiver is an owner. Data is community-held. Upside is shared.
  • W-2 employment + equity in the cooperative
  • The relationship compounds — same neighbor, year after year
  • Clinical-grade record owned by the family, portable across any clinician
  • Profit reinvests into care capacity and caregiver compensation
  • AI is embedded to make the owner more valuable, not the contractor cheaper

Honor uses AI to hold the gig worker. A cooperative makes them the owner. Same retention curve, structurally more durable, and the moat is not the model — it’s the legal form.

The build — three layers, shipping now

One operator. One attestation layer. One user-owned memory. All three live.

Vimty is not a product. It is the argument for a three-layer stack, each layer a separately shipping company, each answering a different structural failure. Together they remove the architectural harms (gig labor, unattested AI output, vendor-held records) before attempting to add capability on top.

Layer 1 — The operator
co-op.care
co-op.care
A worker-owned, physician-supervised cooperative for aging care. Caregivers are W-2 with equity. Families get one sign-up and one physician relationship. Technology runs on CareOS — 9,000+ lines of Claude-powered care-operations software, built to run our own cooperative first, licensable to operators second.
Visit co-op.care
Layer 2 — The attestation
HarnessHealth
harnesshealth.ai
Every AI output in the ecosystem passes through a physician-governed attestation layer before it reaches a patient. Hard intercept on clinical content. Authority-consumption tracking for the reviewer. The infrastructure answer to the 2026 chatbot studies: AI generates, physicians attest, patients get reviewed content.
Visit harnesshealth.ai
Layer 3 — The memory
chanio
chanio.com
The user-owned knowledge graph every AI in the stack reads from, and none of them keep. A local-first health and identity substrate — CGM, labs, wearables, clinical notes, supplements, conversations — portable across Claude, ChatGPT, Gemini, or a local model. Any AI can read it. No AI can hold it. The disk is the durable layer.
Visit chanio.com

Operator + attestation + memory. The three layers are separately fundable, separately shippable, and structurally interdependent. No one of them works without the other two.

The pre-institutional moment

Aging care today is where public health was in 1890.

Enough scientific foundation to be serious. Not enough institutional infrastructure to be systematic. A credibility spectrum that runs from rigorous research to influencer advice. No agreed scoreboard for what aging well even looks like.

Public health’s early wins came from removing harms first — clean water, vaccines, sanitation — before it tried to add capability. The architectural harms in aging today are fragmented data, gig labor, and unattested AI clinical output. Vimty’s thesis is that those harms have to be removed first. The capability layer comes next, and safely.

MIT AgeLab launched a Longevity Preparedness Index in October 2025 to try to measure what preparedness for a longer life looks like at the individual level. Joseph Coughlin’s broader argument in The Longevity Economy is that an aging society is the most misunderstood market of our era. Vimty’s contention is that individual preparedness is not enough. The unit of preparedness is the community, and the infrastructure is cooperative.

Community leverage is how removal happens. Capital alone cannot retain the caregiver, audit the chatbot, or hold the record. A cooperative can.

Join the first community cohort.

Founding members, founding caregivers, founding physicians, and the first families. We are building in Colorado, shipping in public, and replicating with whatever community adopts us next.