Part 2: Accelerate multi-agentic AI adoption to give doctors their day jobs back

US healthcare will continue to face various challenges, including cost, epidemics, disease prevalence, access, clinician shortage, and an aging population, all of which will worsen as demand increases and supply shrinks. Agentic AI will have to play an oversized role in managing some of these challenges addressed in Part 1 of this series.

The US is currently short of 55,000 doctors and 79,000 nurses. That number will increase to about 80,000 for doctors and remain at a high of 65,000 for nurses. The confidence of those metrics is limited given the level of uncertainties in societies and their implications. To add a twist to the shortages, there is a gap between the type of clinician needed and the demand (see Exhibit 1), e.g., by 2034, there will be more older people than younger, yet the clinician specializations don’t align with the needs of the time. Given that graduating more doctors will not bridge the gap, the only real path to maximizing clinician time is delegating all administrative processes to agentic so clinicians can focus on delivering care and helping health systems and hospitals stay financially healthy.

Exhibit 1: The onerous admin processes that burden clinicians will further worsen clinician shortages

Source: Bureau of Health Workforce, US Dept. of Health and Human Services, HFS Research, 2025

Agentic can put a dent in the administrative processes that have become cancerous

Administrative processes forced upon doctors by health plans, such as prior authorizations (PA) and regulators for various reporting requirements, can be significant. Several estimates indicate that doctors spend between 20% and 35% of their time on non-clinical functions, including billing, insurance processes, and compliance documentation. Spending eight to 14 hours of a 40-hour week regularly on these functions is detrimental to doctors (loss in income, lower satisfaction, limited access to patients) and patients (increased costs, reduced access, delayed care). In a landscape of increasing demand and declining clinical supply, administrative burdens are corrosive and harmful to care delivery and hospital financials as they divert clinical time to unnecessary tasks.

While healthcare funders (payers, government) will always raise the bar for access to their funds, clinicians now have a new tool to fight this battle for them—agentic AI (see Exhibit 2). Task-based agentic is now commonplace for auto-generated clinical notes, detecting fraudulent claims, and managing prior authorizations. However, by 2027, we should expect multi-agentic (multiple specialized agents that collaborate to perform complex, interrelated tasks autonomously) to have a significant footprint.

Exhibit 2: By 2027, doctors can likely practice 100% clinical care without the shackles of admin burden on their time

Source: HFS Research, 2025

Remain skeptical of the early examples, but embrace the possibilities

Marketers have been very busy painting anything remotely AI as agentic. There is a clear technical difference between AI agents and agentic AI. The agentic architecture, orchestration, dynamic registry, and memory layer are some attributes that separate it from AI agents, aside from the outcomes. So, conducting deep diligence on any agentic AI solution is critical to validate its authenticity.

Yet some examples have a multi-agentic feel (see Exhibit 3). While the hype of AI is powerful, there are many questions to answer about the multi-agentic properties of solutions. Key filters for multi-agentic AI include system freedom to reorder tasks, backtrack, or invent new subtasks, continuous learning, context accumulation, avoidance of repeat work, support parallel, multi-modal, or multi-domain workflows without manual prompts.

Exhibit 3: Claims of multi-agentic AI are beginning to grow, which can boost accelerated scaling

Source: HFS Research, 2025

The value of multi-agentic AI currently appears to be focused on addressing legacy challenges that are about increasing barriers to care (prior authorizations, appeals management), payments (claims, billing), and customer service to achieve cost savings by reducing headcount and completing processes faster, and by some measure, with greater levels of accuracy. While some legacy processes are deliberate, others, such as member services, can be proactively addressed. Most member service calls are to understand eligibility for benefits and the benefits in the plan. Similarly, patients in a post-acute setting need help to learn about support resources or clarification of medications. All of these can be addressed proactively and consistently with multi-agentic AI connecting earlier and often with members and patients through the channels of their choice and supporting their needs. Health plans and health systems have an opportunity to truly open their minds to the possibilities.

There are many barriers to adopting multi-agentic, but embrace it nevertheless…it will pay off

Like all new technologies, AI poses various risks that can fuel pessimism about its potential. Understanding those risks and developing specific remedies is essential to ensuring deliberate and timely actions. While the challenges may seem daunting, the end goal of maximizing clinician time is a huge prize for both health systems and society in general.

1. Tech-debt: Most payers still run COmmon Business Oriented Language (COBOL)-based core admin platforms, and many health systems sit on monolithic electronic health records (EHRs) closing on 2 decades in age. These stacks lack event-driven APIs, making it difficult for agent swarms to fetch real-time data or trigger downstream actions. Multi-agentic AI needs longitudinal, high-quality data (clinical, claims, social determinants of health, pharma, device feeds) to assign subtasks and reason over patient journeys.  It is time to cut the cord with legacy tech and embrace modern SaaS or BPaaS offerings while getting serious about data governance.
2. Cybersecurity: Agentic AI brings new concerns beyond typical LLM risks. Attackers can poison memory to alter an agent’s long-term behavior or use prompt tricks to hijack decisions. Since these agents connect to tools such as electronic health/medical record (EHR/EMR), Core Admin system (CAPS), and other databases, even a minor exploit can lead to real-world damage. When multiple agents work together, one compromised agent can quietly affect others. Data and AI governance is foundational, accompanied by Zero Trust identity, isolated runtimes, supply-chain integrity, prompt-layer guardrails, and continuous monitoring to ensure multi-agentic AI does not go rogue. Invest in the next generation CISO and security protocols that leapfrog contemporary posture.
3. Consumer adoption: There is significant fear of job loss, a lack of fluency in prompt-based workflows, and concern over algorithmic bias as a bigger hurdle than technology itself. Health plans and health systems have no evidence of strong change management in the past or collaboration with clinicians to co-design. This construct leads to perpetual experimentation and proofs of concept with no scaling, which is trapped in endless pilot cycles. Invest in collaborative change management to bring employees and consumers together to maximize agentic potential.
4. Actual cost of Agentic AI: The popular narrative is that AI will reduce the human workforce and reduce costs (operations, products, and services). However, there is no evidence to suggest that AI is inexpensive. All estimates of savings are precisely that: estimates. There is just not enough data to suggest that the total price of tokens, compute, power, etc., will be less expensive. There are ways to understand the true cost and avoid the sticker shock. However, the value proposition of AI is about being a catalyst to reimagine the possibilities; it’s the value that must be quantified, not just the cost.

The Bottom Line: Multi-agentic AI is a no-brainer for maximizing the ever-shrinking clinician time and improving hospital financials.

Clinician retirements, early exits due to burnout, and medical graduate output not meeting population demands are driving the depressed clinician count for the next decade and possibly beyond. Technology is the only way to bridge the gap between supply and demand. Agentic AI is promising enough to invest in and embrace sooner rather than later to strengthen hospital financials.