I. What is a Health Companion?

A Health Companion is a persistent, AI-native fiduciary agent that serves as a patient’s primary interface to the healthcare system. Unlike patient portals, which are passive, provider-owned Systems of Record, the Companion is an active, patient-owned System of Agency. It is the only entity in the ecosystem functioning solely to optimize the patient’s best interest.

The Companion functions as a clinical chief of staff, lifting the administrative and emotional burden of health management. The Companion accompanies the patient, takes ambient notes, and manages follow-up logistics like scheduling and paperwork. No one should have to go to an appointment alone. 

Think of it as the digital equivalent of a parent managing a child’s care. A parent doesn’t just access data, they advocate and empathize. They fight for the earlier appointment slot, remember the allergy the specialist missed, and demand answers when the bill looks wrong. Crucially, they listen to the patient’s feelings, not just the logistics – often translating clinical jargon into comfort and listening to worries that provider systems often ignore. The Health Companion brings this same level of protective, executive agency to every user, offering four core capabilities:

1. Universal Memory: Aggregating a longitudinal health record across all providers, wearables, and payers into a single, user-owned graph
2. Executive Function: Possessing ‘Write Access’ to the system to autonomously execute tasks, including booking appointments, refilling prescriptions, and initiating transfers
3. Financial Defense: Auditing bills, navigating benefits, and optimizing network selection
4. Empathetic Bonding: Providing consistent, non-judgmental support; validating fears, translating jargon, and ensuring no patient ever feels like they are facing the system alone

This is incredibly hard. If it were easy, it would have been solved by now. Building a true System of Agency requires untangling the most complex, adversarial, and fragmented industry in history. Today, dozens of companies are building pieces of the puzzle: AI scribes, billing advocates, and record aggregators. No one has yet unified them into a cohesive Consumer OS. This will not happen overnight – it may take 5+ years for this shift to occur and will likely produce multiple winners rather than a single monopoly. But the shift is inevitable.

We are witnessing the ‘Browser Moment’ for the $4.9+ trillion U.S. healthcare economy. Just as web browsers decoupled the user interface from the underlying server, Health Companions will decouple the patient experience from provider silos. By establishing the first true System of Agency, this platform captures the interface layer for 18% of U.S. GDP. The winner becomes the primary gateway for care, eventually orchestrating appointments, prescriptions, and referrals. This marks a structural shift in power: from those who supply care to the patient who owns the demand for it.

II. The Patient as CEO

The U.S. healthcare system is architected for provider convenience, payer risk mitigation, and intermediary profit. The patient is left as the uncompensated project manager, tasked with bridging interoperability gaps and deciphering billing codes. Patients and caregivers spend an average of 8 hours a month on administrative labor, a load so heavy that 61% have skipped care to avoid the hassle.

We are witnessing an inversion of this hierarchy. Three forces are converging to center the patient as CEO: maturing agentic AI, national data interoperability (TEFCA), and the acute crisis of the ‘Sandwich Generation’. A unique window exists for a Health Companion: a trusted, fiduciary agent that abstracts away administrative complexity. This is not just a digital front door; it is an Iron Man suit, wrapping the consumer in a layer of intelligence and agency.

The last era of digital health digitized the chart via Systems of Record (EHRs like Epic). We now need a System of Agency for the process – an active, patient-owned layer that empowers the consumer to fight back. In an era of algorithmic denials, unassisted patients are not just inconvenienced, they are outmatched.

Health Companions represent a generational opportunity to reshape the patient experience by establishing a persistent, trusted, and agentic interface to the U.S. healthcare system. Companions can serve as advocate and administrator, capturing subscription, transaction, and patient-permissioned data revenue while building long-term patient relationships. Winners will combine AI-native consumer UX with healthcare integration depth to create a new front door to care that puts the patient in control.

III. Administrative Determinants of Health

The U.S. spends nearly $1 trillion annually on administrative costs, with 40% of hospital spend going purely to administrative activities. The patient effectively pays for the bureaucracy that blocks them. While the importance of Social Determinants of Health are well understood, these Administrative Determinants remain in the shadows.

This friction is by design. It reduces payouts and prevents shopping behaviors that drive competition. Post-care navigation is equally hostile: 60-80% of medical bills contain errors. Correcting them requires auditing CPT codes and navigating appeals, tasks no consumer should face alone.

IV. Patient-First Architecture

Amid the ongoing consumerization of healthcare, patients expect consumer-grade UX, transparency, and direct data ownership (not just better portals). Consumers have first-choice platforms for other categories of spend, but no ‘home screen’ for health. A Health Companion fills this void by architecting a privacy-first OS around the user.

To function as a true fiduciary, the Companion must own the central consumer health data wallet. This reverses the traditional model where all data lives in provider systems, and the patient requests access. Going forward, patient data should live with the patient, with the hospital requesting access for data originated outside their systems. 

Health Companions require privacy-first architecture and business models. The user is the customer, not the product. The Companion can help patients understand, control, and monetize their information. No patient data (anonymized or otherwise) should be sold without the patient’s consent. The patient must be a financial beneficiary of any monetized data.

The Companion must have an ambient ingestion engine to minimize reliance on manual data entry. By ‘sitting in’ on visits, the Companion generates the clinical record in real-time. It captures the ‘exhaust data’ that EHRs miss: the doctor’s tone, the patient’s worries about side-effects, etc.

With patient permission, the system should act as the General Contractor, using Read/Write APIs to orchestrate specialized agents. It supplements provider data and dispatches agents trained for specific tasks (e.g., billing agent for claims), often interacting with provider and payer agents as needed. The health Companion unlocks do-it-for-me healthcare with the patient in control.

This architecture is critical for caregivers managing the health of children, aging parents, or other loved ones. Unlike current portals that force caregivers to juggle multiple logins, the Companion utilizes a multi-tenant permission layer, allowing a single user to manage multiple health profiles from one dashboard. It serves as the family’s institutional memory: when a home health aide quits or a specialist retires, the Companion retains the context (medication schedules, behavioral nuances), ensuring continuity of care.

V. The Trust Ladder

Trust is earned in drops and lost in buckets. The Companion cannot demand full autonomy on Day 1; it must climb the trust ladder. The rollout mimics the clinical training hierarchy: start with documentation before supporting intervention.

Previous attempts at a Health OS have failed in part because they required patients to upload/input data to build a health record from scratch. An AI-native platform will likely use ambient scribing and financial defense as product wedges to bypass this friction and drive immediate consumer ROI. An ambient scribe and record keeper can record doctor visits, note follow-ups, and organize scattered bills. Patients can take pictures of bills to scan for errors. Both products allow a memory layer and clinical history to populate without a tedious onboarding process. Further, these actions are low-liability but high-value – no more misunderstanding around What did the doctor say?

Once the Companion ‘knows’ the patient, it earns the right to perform analytical functions: comparing costs, flagging billing errors, and suggesting in-network or low-cost, cash-pay providers. Finally, it graduates to executive functions: booking appointments, transferring prescriptions, and executing payments.

VI. Components of Competition

Many components of the Health Companion already exist but are scattered across point solutions.

• Generalist Giants: Just yesterday, OpenAI launched ChatGPT Health, allowing users to link their medical records (via b.well) and wearable data. However, they remain a ‘Read-Only’ advisor. They can explain a diagnosis, but they cannot call an insurer to fight a denial. The opportunity remains for the Executive Agent that does the work. Crucially, without the power to act, skeptics argue these tools risk becoming ‘Anxiety Engines,’ surfacing problems that they are powerless to fix, while struggling to overcome deep consumer mistrust regarding inadequate guardrails.
• Social Companions: Platforms like Replika and Tolan are great at empathetic bonding, proving that humans will form deep relationships with AI. (Tolan’s shift to GPT-5.1 voice features yesterday proves just how ‘real’ these interactions feel). However, they are clinically illiterate and unprepared to interact with the health system. They can comfort you about a breakup, but they cannot evaluate your symptoms or navigate a prior authorization. They are friends, not fiduciaries.
• Care Navigation: Human-led care navigation and advocacy services (e.g., Solace Health, Accolade, Included Health) are the best early examples of this process. However, historically they have relied heavily on human navigators, limiting the speed and scope of what they can cover. Included Health recently launched an AI assistant ‘Dot’, but these tools are still tied to employer contracts. If you lose your job, you lose your companion. AI-first care navigation services are starting to emerge (e.g., Mira Mace), but most are focused on Medicare given reimbursement dynamics. Currently no reimbursement codes exist for AI-only care navigation, but that should change over time. Major health systems and insurers are adding AI assistants but face mistrust given misaligned incentives.
• Financial Defense: Companies like Sheer Health, Reclaim, and Goodbill have validated the consumer demand for billing advocacy. However, financial defense is ideally a feature of a broader OS. Once the bill is fixed, the user needs help avoiding the next error, which requires upstream coordination (scheduling/network selection).
• Unified Records: Companies like Picnic Health, Citizen Health, and Novellia provide data aggregation and standardization to provide a complete medical history. However, they remain systems of record – passive repositories that can answer questions but cannot perform tasks. They solve the interoperability problem but do not yet solve the agency problem.
• Condition-Specific Care: So far, DTC players have focused on specific condition areas: Trellis (maternal health), Omada (weight management), Hinge (musculoskeletal pain). A winning Health Companion must transcend individual conditions and work alongside vertical providers.
• Caregiver Platforms: Existing tools for caregivers have largely failed to reach venture scale because they function as digital binders. They ask the already-overwhelmed caregiver to manually enter medications, scan documents, and update calendars. They focus on organizing the chaos rather than automating it. They lack the ‘write’ access to fix problems.

Established giants are structurally handicapped. Apple holds the data but refuses to touch the transaction. They want to be the vault, not the lawyer. Amazon is deploying agentic AI via One Medical to crush administrative overhead, but its walled garden model breaks down the moment a patient needs care outside its owned network. Pilots from Epic (‘Emmie’) and United Healthcare (‘Ava’) cannot scale into Companions; they are bound by the scope of their own data silos and limited by a lack of neutrality. White space exists for a consumer-owned agent with long-term memory and identity across providers and payers. From there this platform can build a health data wallet designed for agentic use. Companions can be somewhat verticalized for specific life stages / personas (e.g., young families).

Short term differentiation is defined by data and experience: integrating fragmented sources into a consumer-grade UX backed by a robust permissions graph. As the underlying tech commoditizes, long-term defensibility comes from trust and network intelligence.

At the individual user level, trust relies on accumulated context and indemnification. Even with full medical data interoperability, the platform’s ‘contextual memory’ on the unwritten preferences and history that drive the last (and most valuable) 5% of personalization, remains proprietary. Patients only share private context (e.g., “I didn’t take my meds because they make me dizzy”) when they feel safe. Empathy is the key to unlocking that proprietary data. Clinical adherence is often an emotional problem, not a logical one. Like a long-tenured executive assistant, this context creates high switching costs; once a user trusts that a Companion ‘knows’ them, displacing that trusted slot becomes nearly impossible. 

Crucially, true agency requires accountability. A winning platform must offer a consumer guarantee. If the system breaks it, the system pays to fix it. By using a strong balance sheet to indemnify the patient against administrative errors (e.g., a misrouted payment), the platform can turn safety into a barrier to entry (much like Airbnb’s insurance coverage for hosts).

At a network level, defensibility comes from system visibility. Value lies in the orchestration intelligence to route, automate, or escalate tasks efficiently to optimize both clinical outcomes and consumer costs. This creates a ‘Waze for Healthcare’ effect. By observing millions of interactions, the platform generates a live map of the system’s friction – for example, knowing which payers auto-deny specific codes or which front desks answer at 4 pm. This creates an unfair advantage regarding the system’s behavior that new entrants cannot replicate.

The ultimate winner is the platform trusted with ‘write’ access, autonomously managing the Iron Man suit to defend the patient against a hostile system.

VII. Why Now

The vision of a Health Companion has existed for decades, but the ecosystem has historically lacked the infrastructure and incentives to support it. Regulatory mandates, platform capabilities, and consumer behavior have converged to create a unique window of opportunity. 

• Regulatory Rails: The full operationalization of TEFCA and the 21st Century Cures Act has commoditized the data for the memory layer. By prohibiting ‘information blocking’ (Section 4004) and mandating standardized APIs, the Cures Act transformed patient data from a proprietary asset into a public utility. Simultaneously, the designation of Qualified Health Information Networks (QHINs) has turned universal data access from a business impossibility into a solvable engineering challenge, allowing apps to query records nationwide without individual hospital contracts. However, these mandates largely solve for access (reading data), not action (writing data). The winner must still build the ‘last mile’ infrastructure to normalize messy feeds and bridge the gap between passive observation and active execution.
• Agentic Capabilities: Agentic AI has matured to increasingly handle the harder problem of taking action. We have moved beyond Chatbot 1.0 to a Mixture of Experts model, where specialized agents manage distinct tasks. These agents utilize a multimodal approach, bridging the gap between open APIs and legacy systems by using RPA or voice agents to navigate portals and front desks. While this ‘adversarial’ interoperability serves as the initial bridge, the long-term winner will transition to direct partnerships and APIs, as health systems eventually realize that blocking these agents means blocking their own patients.
• Consumer Readiness: As consumers shoulder more out-of-pocket costs amid the rise of high deductible plans, they are adopting a retail mindset (prioritizing convenience, cost, and quality). Simultaneously, the psychological barrier to non-human support is dissolving. The rise of AI friends and general consumer education around LLMs signal a new comfort level with non-human interaction. With 1 in 6 adults already using health chatbots monthly, the leap from trusting AI with travel to trusting it with health administration is a natural evolution. Increasingly, patients seek not only data access but also data ownership. Additionally, workforce shortages force incumbents to accept non-human support.

VIII: Rise of the Health OS

The killer app for AI in health is coordination, not diagnosis. By automating the $1T+ bureaucratic friction of care, the Health Companion positions the consumer as the center of the ‘Primary Health OS’ – the only integration point that sees the whole picture. Only the patient holds the complete view of their medical history, making them the most effective integration point for a fragmented system.

Success requires moving beyond passive information to active, reliable agency. The goal is not to replace the doctor, but to automate the bureaucracy. The era of the passive patient is ending. Trust serves as the primary defensive moat here, earned through the consistent execution of high-stakes tasks. This dynamic creates a relationship closer to the intimacy of AI Social Companions than to traditional software.

However, the path forward is gated by significant execution risks. Building a trusted consumer relationship, navigating regulatory gray zones (medical advice vs. general info), ensuring healthcare-grade privacy, and getting the business model right remain non-trivial hurdles. Yet, the opportunity is massive for those who can balance safety with utility. The most valuable healthcare tool is simply a partner that shows up, remembers the details, and gets the job done.

If you or someone you know is working in this space, please reach out to kathryn [at] firstmark [dot] com