AI for the Aging Population: The Shortage You Can't Hire Your Way Out Of — why the most obvious market in the batch is the least forgiving to build for

This is the eighth article in a fourteen-part series reading Y Combinator's Fall 2026 Requests for Startups one request at a time. Most requests in the list have to argue that a market exists. This one does not. Max Kolysh's "AI for the Aging Population" opens with a sentence that needs no venture framing at all: "By 2030, one in five Americans will be over 65, and there's nowhere near enough people to take care of everyone."

What the request actually says

The request is short and unusually concrete. Kolysh writes that the US "is projected to have millions of unfilled caregiving jobs within the decade, and 53 million family members are already doing this work unpaid," and that "almost no technology is actually built for older people. Even Alexa and Google Home are frustrating for most seniors to use." He then names four product categories AI newly makes possible: "voice interfaces that can hold real conversations, monitoring that helps older adults stay safe and independent, robotics that can assist with physical tasks around the home, and software that helps family caregivers coordinate care, appointments, and emergencies." He closes by calling this "one of the largest, most underserved markets in the world."

That is the whole request: a demographic wave, a labor gap, four wedges. What makes it worth a careful read is that every one of those wedges has been tried before AI, and most attempts died on the same three rocks — trust, distribution, and who pays.

The numbers hold up, and they are worse than the request says

The demographic claim checks out against Census projections, but the labor numbers deserve their own paragraph because they are the actual thesis. PHI, the research organization that tracks the direct care workforce, counts about 5.4 million direct care workers in the US today, including nearly 3.2 million home care workers — a workforce that roughly doubled over the past decade and is still nowhere close to demand. Between 2024 and 2034, PHI projects 9.7 million total direct care job openings to fill, the largest of any job sector in the country, at a median wage of $17.36 an hour. Those jobs will not all be filled at that wage. That is what "millions of unfilled caregiving jobs" means operationally: the work lands on families instead.

And the family side has already outrun the request's own statistic. Kolysh cites 53 million unpaid family caregivers, the widely used 2020 figure. The 2025 AARP and National Alliance for Caregiving study counts 63 million — nearly one in four American adults, up roughly 20 million since 2015, with over half performing medical tasks like medication management and wound care that only a fifth have any training for. The market grew by an AARP report's worth of people while the request was being drafted. "Growing every single day" is not rhetoric here; it is the one RFS claim that understates its case.

What is actually hard

Start with the split that defines every product in this category: the user, the buyer, and the payer are three different people. The 82-year-old uses the voice companion; her daughter buys it; a Medicare Advantage plan, a Medicaid waiver program, or a long-term-care insurer decides whether it scales past the early-adopter families willing to pay out of pocket. Consumer eldercare companies have repeatedly discovered that direct-to-family revenue plateaus, and that the payer path means enterprise sales cycles, evidence requirements, and reimbursement codes. A founder here is signing up for a three-sided sale from day one.

Second, reliability is the product, not a feature of it. A chatbot that hallucinates a restaurant recommendation is an annoyance. A monitoring system that misses a fall, or a voice agent that confidently misstates a medication schedule, is a harm — possibly a fatal one, certainly a litigable one. The inverse failure is quieter but just as lethal to the business: false alarms train families to ignore alerts, and over-surveillance makes elders rip the sensors out. Add cognitive decline and the consent question — who agreed to this monitoring, and does the person being monitored still understand it? — and "move fast" stops being an available strategy.

Third, the regulatory surface arrives earlier than founders expect. The moment a product touches a provider, a payer, or health information, HIPAA applies. The moment it makes claims about detecting or preventing a medical condition, FDA territory begins. Home care agencies themselves are state-licensed and state-regulated. None of this is prohibitive — companies clear these bars constantly — but each bar reprices the build in time and specialized headcount.

Fourth, adoption by people the technology industry has spent thirty years not designing for. Kolysh's Alexa jab is fair: mainstream voice assistants assume crisp speech, tolerance for failure, and a smartphone nearby. The counter-evidence that it can be done comes from Intuition Robotics' ElliQ, deployed through the New York State Office for the Aging to hundreds of older adults, with the company reporting high daily engagement — and, notably, distribution through a state agency rather than a retail shelf. Intuition Robotics has reportedly raised $25 million led by Toyota's growth fund to scale it. On the harder end, robotics startups like Andromeda Robotics are reportedly raising to put companion robots into senior care, and PitchBook has argued elder care may be where home robots first find real demand — but physical assistance in cluttered, unstructured homes remains the highest-liability, longest-timeline wedge in the request.

Which points at the pragmatic entry ramp: of Kolysh's four categories, caregiver-coordination software is the one you can ship without a device, a clinical claim, or an FDA conversation — scheduling, medication reminders, sibling communication, emergency escalation for those 63 million families. It is also the least defensible, which is why the durable version probably starts there and earns its way into monitoring and voice over years of trust, not months.

The honest synthesis: the demand is the easiest to verify of any request in this RFS, and the tolerances are the tightest. This market does not reward the fastest demo. It rewards whoever survives the trust-building years.

Where Gwen stands

The lane statement comes first, because this is a request where overclaiming would be ugly. Eldercare hardware, safety monitoring, and clinical workflows are not what Gwen does. Gwen is not a clinical or monitoring product, it does not handle regulated health data, and nothing in this article should be read as Gwen positioning for that work.

Where Gwen honestly fits is one layer out: the businesses doing the care. The home care agencies competing for those 9.7 million openings, the franchises, the family-caregiver services — these are small operators who need a credible web presence, working intake forms, email and CRM follow-up, content, and operational research. That is Gwen's actual job: it builds and hosts websites and small web apps from a plain-language description on live shareable links, the customer owns the code and can export it to GitHub, and the work runs as long-lived missions in a workspace with durable transcripts and human approvals gating anything outward. Customers buy a Work Budget with enforced spend ceilings; underneath, a routing rail spreads tasks across many AI models by measured quality and cost.

One caveat stated rather than buried: a care company's marketing site and operations tooling must be built so that client health information never flows into them, because Gwen does not and will not process it. For a care founder, that boundary is a design requirement, not a footnote. If your business is caring for aging people, Gwen can build the part of your company that wins you the family's trust online — and it will tell you clearly where its part of the work has to stop.