AI agents in healthcare: Building an infinite care team

Summary

• Healthcare requires more human attention than it can hire or afford, with the global health workforce gap heading toward 11 million by 2030.¹ 
• AI agents in healthcare can now coordinate and complete multi-step care journeys end-to-end, at a marginal cost approaching zero.
• The final barrier is trust. An agent talking to a sick or anxious person has to clear a far higher safety bar than a back-office administrative tool.
• The care team of the future will be built agent by agent. The organizations that build on a healthcare-grade foundation now will get there first.

Soon, every person will have a team of AI agents working on their health. I’ve believed that for a while, but I’m more convinced of it today than I’ve ever been. At this point, the question worth asking is how fast we get there.

I’ve spent a lot of time this year in rooms with healthcare executives. Whether I’m sitting with a leader of a small regional health plan or a massive national health system, the conversation ends up in the same place: demand keeps climbing, while the workforce and the budget to meet it keep shrinking.

Last month, I sat down with the CEO of a globally ranked academic health system. His problem is one I hear constantly. He simply doesn’t have enough care coordinators for the number of complex patients his teams manage. When nobody has time to follow up, the default answer becomes “come back to the hospital.” That answer is incredibly expensive for the system and exhausting for the people working inside it. It’s also usually the wrong answer for the patient.

Leaders like him know what their members and patients need: one-on-one support to coordinate their care. What they can’t do is hire enough people to deliver it. 

That is the gap AI closes.

WHITE PAPER

The foundation underneath safe healthcare AI agents

Six capabilities your agents need to safely and effectively support members and patients.

Read the white paper

The math of human attention no longer works

Historically, healthcare has always grown the same way. When demand goes up, we hire more nurses, coordinators, and call center staff to meet it. That model has hit its limit.

In 2023, 76.4% of U.S. adults were living with at least one chronic condition,² and so were about 44% of Canadian adults.³ These are people who need steady attention all year long, and there are more of them every year.

At the same time, health spending keeps growing faster than the economy. U.S. national health spending reached an estimated $5.7 trillion in 2025, the third straight year of growth above 7%.⁴ Canada is seeing a similar trend.⁵

Yet we don’t have the labor we need to keep up with demand. The U.S. is projected to be short more than 140,000 physicians⁶ and more than 100,000 registered nurses⁷ by 2038. Globally, the shortfall is expected to reach 11 million health workers by 2030.¹

We pay for that gap every day, usually in the most expensive settings of the system:

• $32 billion a year in avoidable ER visits among privately insured patients alone.⁸
• $100 billion to $300 billion a year in avoidable spending tied to medication non-adherence.⁹
• An average of $16,300 for every 30-day hospital readmission.¹⁰

These are the costs of care arriving too late. A patient who could’ve stayed healthy at home ends up back in the hospital. We tend to look at quality scores, retention, medical cost, and cost to serve as four separate problems, but they share one constraint: human attention. Nobody can hire their way out of it.

Meeting demand without the limitations of labor

AI gives healthcare something it has never had: the ability to put a proactive care team around every member and patient, 24/7, with no limit on labor. At League, we call this the infinite care team. 

Agents can reach out before something goes wrong and get people to the right care at the right time. I wrote about this in The end of reactive healthcare, and we laid it out for health plans in our Intelligent payer series. What makes this possible now is a radical shift in both cost and capability. The marginal cost of AI-generated attention is approaching zero, and agents can now support someone through an entire, multi-step care journey.

To see what that means for one person, take Sara. She’s a composite Medicaid member, but her story is one health plans and health systems see every day.

Sara’s mammogram is overdue. A reminder letter comes in the mail, but booking it means calling during work hours. She puts it off, eventually forgets about it, and the screening never happens. Later, her Medicaid renewal packet goes to an old address, with a deadline she doesn’t know about. She doesn’t even realize her coverage has lapsed until a claim gets denied.

When her daughter gets a rash on Sunday night, the doctor’s office is closed. With no way to get advice on what to do, she spends hours in the ER, only to hear it could have been a primary care visit that week. When she calls to ask if a ride to her next appointment is covered, her break ends while she’s still on hold, so she hangs up and cancels the appointment.

Now, imagine that same scenario with an infinite care team.

An AI agent reaches out about the mammogram via text, books it, and confirms the visit. Another walks her through her renewal before the deadline, in her native language. When her daughter gets a rash on Sunday night, an agent asks about the symptoms and guides Sara to book a primary care appointment for the next morning instead of defaulting to the ER. If anything had pointed to an emergency, the agent would have directed her to the ER right away. Her question about transportation is answered by an agent that books the ride for her in minutes.

Each of these moments needed someone to coordinate care for Sara and her family. That someone always had to be a person, until now.

Every moment that got easier for Sara also moved the metrics her health plan and providers are measured on:

• The mammogram closes a care gap, raising HEDIS and state quality scores. 
• The on-time renewal means her plan keeps her capitation payment. 
• Her daughter’s rash becomes a PCP appointment instead of a costly ER visit.
• The ride booking is one less call that a human has to handle.

The hard part is trust

The conversation around AI agents in healthcare has officially shifted from adoption to trust. We talked about this in our recent Fierce Healthcare webinar.

Almost every executive I talk to already has AI running in the back office, and some have agents handling simple administrative tasks for members and patients. But very few have agents guiding people through more complex and clinically relevant use cases like medication adherence, chronic condition management, palliative care, or oncology. 

The safety bar there is different. Back-office AI can be wrong some of the time, but a member- or patient-facing agent has to be right 99.99% of the time for every member and patient, across different plans, languages, and channels. A wrong answer about coverage or symptoms can turn into a surprise bill, a grievance, or a delayed diagnosis. And the organization owns every word its agent says.

Any engineering team can stand up a convincing demo in an afternoon, but getting to production takes ongoing testing, guardrails, approved knowledge sources, escalation rules, and monitoring to make sure the agent is safe. In practice, you’re looking at over 100 subsystems working together under a single agent. 

The reality of building and maintaining that infrastructure leaves leaders with a difficult choice: sacrifice speed or compromise on safety. But organizations don’t have to make that tradeoff when they build agents on a healthcare-grade platform with the safety layer already built in.

Why we built Forge by League, a healthcare-grade AI platform

This vision is why we built Forge by League. In Greek mythology, Hephaestus’s forge turned raw metal into tools for the gods. Forge by League turns raw AI capability and a decade of health engagement data into tools healthcare relies on.

Our customers are bringing us increasingly complex challenges, including maternity, palliative care, chronic condition management, and Medicaid redetermination. They’re also bringing us ones that sound simple but aren’t, like health coaching and benefits navigation.

Many of the agents needed to solve these problems are already built and supporting members and patients today. At a leading hospital network, a palliative care agent answers families’ questions at any hour, bringing in a clinician when it’s needed. At one of the largest not-for-profit health insurers in the U.S., a health coaching agent answers members’ health questions within clinical guardrails and points them to the programs and support they need.

FORGE BY LEAGUE

What could agents do for your members or patients?

Bring the problem you most want to solve, and we’ll show you where  Forge by League could help first.

Get started

Forge by League is the healthcare-grade AI platform that makes this possible. You can start with agents we’ve already built, build your own with our forward-deployed engineers, or do both. Every agent runs on the same governed foundation, with six core requirements built in:

• A range of models, grounded in approved knowledge: Routed by task, running on infrastructure you control.
• Vertical evals and golden datasets: an expert-written answer key sets the bar before an agent ships, and your team can add its own test cases.
• Omnichannel tools: chat, SMS, voice, and WhatsApp all connect to the same reasoning engine.
• Shared member and patient context: every conversation picks up where the last one left off.
• Orchestration over time: a plan of record holds between sessions and prompts outreach when something needs attention.
• Guardrails and escalations: every turn is checked for harmful content, PHI exposure, and medical emergencies.

The foundation also compounds. Every conversation an agent has leaves a record of what it saw and did, which is reviewed and turned into new test cases. As the foundation improves, every agent on it gets better at delivering the right care at the right time.

This flywheel effect is another reason why moving early matters. An organization that starts deploying agents now starts collecting the real-world interactions that make its next ten agents safer and cheaper to launch. It also means your team spends less time on safety infrastructure and more time building the experience.

Building an infinite care team for families like Sara’s

Every health plan and system shares the same mandate right now: get people to the right care at the exact moment they need it. But there will always be more families like Sara’s than we have the human workforce to support. By building an infinite care team, we can finally close the gap between the care members and patients deserve and the capacity we actually have, so no one falls through the cracks.

So here’s my question: what use case would your team build an agent for first?

WHITE PAPER

Your checklist for deploying healthcare-grade AI agents

See what a safe agentic foundation requires, and how it helps organizations build an infinite care team their members and patients need.

Read the white paper

Sources

1. WHO, Health workforce levels & trends 2026 report, 2026.
2. CDC, Trends in Multiple Chronic Conditions Among US Adults, By Life Stage, Behavioral Risk Factor Surveillance System, 2013–2023, 2025.
3. Canadian Public Health Association, Chronic Disease and Public Health in Canada, 2024.
4. Health Affairs, National Health Expenditure Projections, 2025–34, June 2026.
5. CIHI, NHEX Trends, 2025 — Snapshot, November 2025.
6. HRSA, Physician Workforce: Projections, 2023–2038, December 2025.
7. HRSA, Nurse Workforce Projections, 2023–2038, December 2025.
8. UnitedHealth Group, 18 Million Avoidable Hospital Emergency Department Visits Add $32 Billion in Costs to the Health Care System Each Year, July 2019.
9. Duke Health, Medication Nonadherence Increases Health Costs, Hospital Readmissions, November 2018.
10. AHRQ, Characteristics of 30-Day All-Cause Hospital Readmissions, 2016–2020, September 2023.