The Centers for Medicare & Medicaid Services launches ACCESS on July 5: a 10-year outcome-based payment program covering 150 participating organizations testing AI-driven care for diabetes, hypertension, chronic kidney disease, obesity, depression, and anxiety. The mechanism distinction is structural rather than incremental. Traditional Medicare reimburses by clinician time โ every billing code ties payment to a credentialed human hour. ACCESS pays for measurable patient outcomes (lower blood pressure, reduced pain) regardless of who or what delivers the care. For the first time at federal scale, Medicare has a payment mechanism for an AI agent that monitors patients between visits, calls to check in, coordinates a housing or transportation referral, or makes sure someone picks up their medication. Pair Team's voice AI "Flora" โ a 24/7 patient-facing interface deployed 9 months ago โ is one of the 150 selected participants.
ACCESS โ Advancing Chronic Care with Effective, Scalable Solutions โ was designed by Abe Sutton (Director of the CMS Innovation Center, formerly a healthcare VC at Rubicon Founders) and Jacob Shiff (Chief AI and Technology Officer of the CMS Innovation Center, formerly a healthcare founder). Both joined CMS under the Trump administration, and the program's startup-fluent design is visible in three places: outcome-based payments rather than activity-based, direct-to-consumer enrollment paths rather than referral-only, and deliberate competition between participants rather than the traditional regional-monopoly structure. Pair Team's specifics: 850 clinical professionals, the largest community health workforce in California per the company, nine-figure revenue, $30 million raised from Kleiner Perkins, Kraft Ventures, and Next Ventures. CEO Neil Batlivala's example of Flora in action: a 67-year-old woman managing PTSD and congestive heart failure, living out of her car, had an hour-long call with Flora โ "probably the only person she'd talked to in weeks about her situation." Reimbursement rates under ACCESS are explicitly low. Batlivala calls that "a feature, not a bug": "The economics only work if you're running a lean, AI-first operation." The first cohort spans AI doctor startups, virtual nutrition therapy providers, connected device companies, and wearables (Whoop named as a participant). Pair Team's current addressable population: about 500,000 patients; target 1 million within three years.
The structural shift matters more than the dollar amount. Bundled-payment ACO models from the 2010s tried similar outcome-based shifts but kept the human-clinician billing requirement intact. ACCESS removes it. That creates the first real economic surface for AI healthcare agents at federal scale โ billing codes that don't require a credentialed-human-hour as the unit of payment. For the broader "can an AI legally provide professional advice" question running across legal (Anthropic Claude for Legal #829), journalism (NYT freelancer ban #830), and medical (OpenAI wrongful-death suit #827), Medicare ACCESS is the most aggressive answer yet in any sector: yes, and we'll pay outcomes-based for it. The "unauthorized practice of medicine" legal theory becomes harder to litigate when Medicare itself is paying outcomes-based for AI-delivered patient management at federal scale. The risks are real and named in the source reporting: CMS Innovation Center's 2023 Congressional Budget Office analysis found $5.4 billion in added federal spending across the Center's first decade rather than the projected savings; low reimbursement structure means most legacy healthcare providers cannot make the math work without aggressive AI automation, which is the policy's selection pressure operating as designed; and sensitive patient data โ intimate conversations about housing, mental illness, chronic disease โ flowing into federal infrastructure with a documented history of breaches, including exposed Social Security numbers, is not an impractical concern for the vulnerable populations ACCESS is designed to serve.
Goes live July 5, runs 10 years. The 150 participant slots are already filled, but the structural precedent matters more than this cohort. Once Medicare establishes outcome-based AI-friendly billing at federal scale, private insurers historically follow within two to three procurement cycles. For healthcare AI builders not in the first cohort: the 24-month outcome data from these 150 participants is the gating event for whether the program gets renewed and whether private payers replicate the model. For patients in vulnerable populations: voice AI agents replacing some clinician time has real downsides (privacy of intimate health conversations in federal data infrastructure with breach history, reduced human contact for people who already have too little) and real upsides (24/7 availability, attention for the lonely-and-sick population traditional care doesn't reach reliably). For builders watching where the policy weight lands: this is the most concrete US federal AI-procurement signal in healthcare to date, more directly consequential than FDA AI/ML device pathway updates because it touches reimbursement instead of just regulatory approval. The piece TechCrunch makes โ that the tech world hasn't noticed โ is correct, and worth fixing.