Hi Y Combinator Community! We’re Yash, Danny, and Arthur, the founders of Legion Health.

At Legion, we’re building autonomous mental healthcare.

We previously launched Legion as an AI-native psychiatry clinic (see here). We’re relaunching now because we recently crossed an important threshold: Legion has received scoped regulatory authorization for fully AI-powered renewals of psychiatric maintenance medications. To our knowledge, this makes Legion the first mental health company ever to have approval to let AI provide actual medical care.

This is a meaningful product update for us because it is the first time AI in Legion can directly execute a real clinical workflow in production. Until now, most of the value we created came from making the clinic itself dramatically more efficient by automating 95%+ of the admin work. This is different. This is AI-native care delivery moving from support work into clinical action inside a live, regulated system.

TL;DR

Legion Health is building autonomous mental healthcare. We started by building an AI-native psychiatry clinic, and we’re now launching the next step: in Utah, Legion has received regulatory authorization for AI-led renewals of eligible psychiatric maintenance medications.

We think this is an important proof point that mental healthcare can move from clinician-led workflows, to AI-led workflows with supervision, and eventually to fully autonomous care in well-scoped pathways.

If you or a loved one wants early access to the easiest way ever to get incredible mental health care, join and/or share our waitlist here: https://ai.legionhealth.com/

The problem

Mental healthcare is one of the clearest categories where a new site of care is needed.

The issue is not that people do not want care. It is that the system cannot deliver enough of it. There are too few psychiatric providers, too much administrative overhead, and too much friction between a patient needing help and a patient actually getting actionable clinical resolution.

Even in the simplest cases, the system breaks down. A patient who is stable on a medication that is already working can still be forced to wait days or weeks, book a visit, incur costs, chase the pharmacy, and spend time and energy navigating the process instead of getting better. That is not a small inconvenience. It is a direct consequence of a care model built around scarce clinician minutes and fragmented operations.

But continuity is the unit, not visits. Mental healthcare works best when care is longitudinal, not episodic; when the system can follow the patient over time, resolve routine needs quickly, and escalate humans when something truly requires human judgment.

That is why we believe mental health needs a new digital front door to care: one that is always available, clinically actionable, and connected to licensed clinicians whenever escalation is needed.

Our progression: AI-native clinic -> semi-autonomous care -> fully autonomous care

We have always thought the path to autonomy matters just as much as the destination.

Phase 1: AI-native psychiatry clinic

Phase 1 was the AI-native clinic. We built a real psychiatry clinic where human psychiatrists and AI agents work together to provide high-quality care directly to patients. That forced us to solve the hard, messy parts of the real world first: intake, scheduling, medical records, pharmacy coordination, prior authorizations, reimbursement, follow-up, and clinical escalation. In the process, we built a clinical + operational + reimbursement + regulatory stack in one system.

Demo: Garry Tan recently featured Legion on Y Combinator’s YouTube, and here we show our AI operating system that allows us to take care of a huge clinic with an extremely small admin team, using automations and agents.

https://www.youtube.com/watch?v=rWUWfj_PqmM&t=273s

The Legion segment starts around 4:33.

Phase 2: Semi-autonomous care with clinician supervision (today)

Phase 2 is what we are launching now: semi-autonomous care with clinician supervision. In this model, AI handles the default workflow in narrow, safety-scoped pathways, and clinicians supervise exceptions, ambiguity, and higher-complexity cases. The Utah renewals workflow is our first real example of this. It is the bridge from AI assisting clinicians to AI executing narrowly scoped clinical work, with clinician escalation when needed.

Our social media and press launch for our first-ever pilot in mental health:

LinkedIn: https://www.linkedin.com/posts/yashmpatel1_today-we-took-another-huge-step-toward-truly-activity-7443320080760942592-27FX?utm_source=share&utm_medium=member_desktop&rcm=ACoAACKkTu8BFJG1YHsaT6kWOskKDexaBBvfIn8

X: https://x.com/ArthurMacwaters/status/2037479229294874918?s=20

Press release: https://nypost.com/2026/03/27/business/artificial-intelligence-can-now-prescribe-mental-health-drugs

Phase 3: Fully autonomous care

Phase 3 is fully autonomous care in the future. Over time, we believe more workflows will move from clinician-led to AI-led with supervision, to fully autonomous in well-scoped pathways with hard guardrails, continuous auditing, and mandatory escalation when appropriate.

We do not think medicine jumps from zero to full autonomy overnight. We think it gets there one real workflow at a time.

What we launched

In Utah, Legion is rolling out AI-led renewals for eligible lower-risk psychiatric maintenance medications.

The workflow is intentionally narrow and safety-scoped. Patients explicitly opt in, are clearly told they are interacting with AI, go through identity verification, and complete a focused safety review grounded in medication and chart context. If anything is ambiguous, higher-risk, or out of scope, the case escalates to a clinician. Patients can also request human review at any point.

We started with renewals because renewals are a perfect example of what is broken in mental healthcare today, and because they are exactly the kind of routine, high-friction workflow that should eventually become software-handled by default inside a real care-delivery system.

Why is this such an important step in healthcare AI?

At first glance, a refill may sound narrow. We think that is exactly why it matters.

If AI cannot safely take on a constrained, repetitive, clinically meaningful workflow like renewals, then talk of autonomous care is still mostly theory. But if AI can safely execute a narrow clinical workflow in production, under explicit guardrails, with clinician escalation when needed, then the future starts to become concrete.

That is what this launch represents.

It is not just a faster refill. It is a proof point that care can start moving:

• from information to clinical action
• from copilot to autopilot
• from clinician-led workflows everywhere to clinician-led exceptions where they matter most
• from scarcity-based care to a system that can eventually support abundance

We think this is the next important step toward the real solution: mental healthcare that is always available, high-frequency, longitudinal, more personalized, more consistent, and able to scale far beyond clinician supply alone.

Why we had to build a clinic first

We did not start with a chatbot. We started by building the clinic.

That gave us something much more valuable than a thin AI layer: a unified data layer across records, prescriptions, scheduling, messaging, and follow-up; real operational feedback loops; clinician escalation pathways; and the infrastructure required to turn AI into actionable clinical resolution in the real world.

We think that is the real moat here.

The company that wins this category will not just have impressive models. It will own the default entry point to care, the patient relationship, the workflows, the regulatory trust, and the feedback loops required to make AI-native care delivery safe and real.

Why we started Legion Health

My co-founders and I have known each other for 12 years and were roommates at Princeton. I (Yash) worked on Medicare and Medicaid payment policy at the Congressional Budget Office, Danny brings deep AI and engineering experience, and Arthur brings product, operations, and design depth.

This problem is also deeply personal to us. We started Legion because we have seen how broken mental healthcare can feel when someone actually needs help, and we want to build the kind of system we wish existed for people we love.

Call to action

If you or a loved one wants early access to the easiest way ever to get incredible mental health care, join and/or share our waitlist here: https://ai.legionhealth.com/

If you’re a cracked engineer, and you care about working on hard problems that impact real humans, we want to hear from you.

If you’re an investor who believes in a future where humans and AI work together to solve humanity’s biggest problems, we want to hear from you.

If you work in digital health, payer partnerships, state/federal government, or AI-native care delivery and want to talk, we want to hear from you.

Contact: yash@legionhealth.com