Category: Opinion

A former health system CEO is making a blunt argument that many healthcare leaders have so far only implied: the fastest path to financial survival may be swapping large portions of today’s workforce for increasingly autonomous AI. In a STAT+ First Opinion, former Geisinger CEO Glenn Steele contends that health care is on a trajectory toward autonomy—and that systems that don’t embrace it will struggle to stay afloat.

The provocation isn’t the idea that AI will change workflows; that’s already happening. The provocation is the scale and certainty of the claim: that “huge numbers” of roles will be replaced, not merely assisted, and that this shift is not optional but existential. That framing matters because it resets the industry conversation from “AI adoption” to “labor substitution,” a far more contentious—and operationally complex—transition.

Why this matters now: margins, labor, and the limits of incremental automation

U.S. health systems have been trapped between rising labor costs, unpredictable payer dynamics, and growing clinical complexity. Even as patient volumes rebound in many markets, margins remain fragile—especially for community hospitals and safety-net providers. The traditional playbook for cost control (revenue cycle optimization, supply chain savings, incremental productivity gains) is running out of runway. Steele’s thesis, as reported by STAT News, is that autonomy—software that can execute end-to-end tasks with minimal human oversight—becomes the next lever big enough to move the balance sheet.

That’s plausible in a narrow sense. Healthcare is packed with “coordination work”: authorizations, scheduling, documentation, referral management, coding, billing follow-up, inbox triage, discharge planning logistics, and the endless handoffs that keep care moving. Much of this work is rules-based, repetitive, and measurable, which makes it a natural target for agentic AI systems that can read, write, summarize, route, and act across multiple software platforms.

But turning plausibility into reality requires a leap many organizations haven’t made: trusting AI not just to suggest, but to do. Most AI deployments today are “decision support,” living behind disclaimers and human review. Autonomy implies a different operating model—one where humans supervise exceptions, handle edge cases, and provide accountability, while AI performs the routine execution at scale.

Which jobs are most exposed—and which may become more valuable

If health systems pursue autonomy as aggressively as Steele suggests, the first wave of displacement will likely concentrate in administrative and operational domains: revenue cycle functions, prior authorization processing, call center operations, patient access, routine chart abstraction, and basic care coordination. These areas already rely heavily on digital inputs, templates, and standardized decision trees. AI doesn’t need a robot body to replace work that is fundamentally screen-based.

Clinical roles are more complicated. Some tasks within nursing, pharmacy, and medicine can be offloaded—drafting notes, reconciling meds, summarizing histories, monitoring remote patient data, and guiding patients through protocols. But replacing clinicians wholesale runs into constraints that aren’t just technical: licensure, malpractice liability, informed consent, and the reality that many “clinical” decisions are actually negotiations among patient preferences, uncertainty, and risk tolerance.

Paradoxically, autonomy may increase the value of certain human roles. As AI absorbs routine production work, the system will need more people who can audit outputs, investigate failures, validate model drift, and redesign workflows. “Clinical AI safety,” “model operations,” and “AI governance” functions—once niche—start to look like core infrastructure. The workforce may shrink in some areas while specializing and professionalizing in others.

What patients stand to gain—and what they might lose

Done well, autonomous AI could reduce the friction patients experience as “care logistics.” Imagine faster scheduling, fewer authorization delays, clearer instructions, better follow-up, and fewer errors caused by manual data entry. Autonomy could also make access more equitable if it scales scarce expertise—like triage and navigation—into always-on digital services.

But replacing people with software also changes the patient experience in ways that aren’t captured by throughput metrics. Patients already complain that healthcare feels impersonal and bureaucratic; a more automated system could amplify that sentiment if human touchpoints disappear. There’s also a trust gap: patients may accept AI that helps clinicians, but may be far less comfortable with AI that makes or executes decisions with limited human involvement—especially in moments of vulnerability.

The safety risks are equally real. Autonomy can fail quietly: an agent that misroutes a referral, misinterprets an exclusion criterion, or miscues follow-up can cause harm without a dramatic “AI error” moment. In an autonomous workflow, error detection becomes a design requirement, not an afterthought—continuous monitoring, audit trails, and clear escalation paths are non-negotiable.

The real constraint: governance, accountability, and the politics of workforce change

The hardest part may not be building the AI; it may be implementing it in organizations defined by professional boundaries and regulatory exposure. Health systems have to answer basic questions before autonomy can scale: Who is accountable when an AI agent acts? What documentation proves appropriate oversight? How do you measure and report safety? What’s the policy when an AI recommendation conflicts with clinician judgment or payer rules?

Then there’s the human reality. Large-scale labor substitution invites backlash—from employees, unions, clinicians worried about deskilling, and communities where hospitals are major employers. Even if autonomy helps the bottom line, leaders will need a credible transition plan: retraining pathways, role redesign, and clear communication about where humans remain essential. Without that, “AI transformation” becomes a morale crisis.

What comes next: from copilots to command centers

Steele’s argument, published as a First Opinion in STAT+, will resonate with executives who view autonomy as a last best chance to stabilize operations. But the winners won’t be the systems that automate fastest; they’ll be the ones that industrialize safety and governance while redesigning care around what humans do uniquely well.

Over the next few years, expect a shift from scattered copilots to centralized “AI operations” teams that manage fleets of agents the way hospitals manage pharmacy systems or labs: with standard operating procedures, validation, downtime plans, and relentless quality control. Autonomy is likely to arrive first in the back office, then creep into clinical pathways where oversight is easiest and outcomes are measurable.

Whether that future improves healthcare—or simply makes it cheaper to run—will depend on choices made now: what gets automated, what remains human, and how transparently systems explain those decisions to patients and staff.

Source: STAT News AI (First Opinion), “Former Geisinger CEO: U.S. health systems must replace huge numbers of people with AI,” April 7, 2026.

Healthcare leaders aren’t short on AI pilots. They’re short on AI that actually sticks. That’s the core warning in a recent piece from Healthcare Dive: the industry’s problem isn’t experimentation, it’s execution — and the widening gap is largely a data problem.

This is an uncomfortable inflection point for health systems, payers, and digital health vendors alike. Generative AI demos are easy to admire. But turning AI into durable clinical and operational capability requires something far less glamorous: a modern data foundation that can support reliable, governed, and reusable workflows.

From “cool pilot” to production: why the gap keeps growing

Over the last two years, many organizations have launched AI initiatives in parallel: a clinical documentation assistant here, a revenue cycle automation there, maybe a patient messaging bot. What they often discover is that each use case hits the same wall: fragmented data, inconsistent definitions, limited interoperability, and unclear ownership.

When data is scattered across electronic health records, imaging systems, claims platforms, lab vendors, and departmental databases, AI development becomes a one-off integration project every time. Teams spend months reconciling identifiers, aligning timestamps, mapping codes, and adjudicating “what counts” as a diagnosis, encounter, or readmission. AI then becomes expensive bespoke work — not a scalable capability.

As reported by Healthcare Dive, this execution gap is widening. That makes sense: organizations that already invested in data engineering, governance, and analytics maturity can deploy new AI tools faster, monitor them better, and prove ROI sooner. Everyone else faces a compounding disadvantage: the more pilots they run without fixing the data layer, the more technical debt they accumulate.

Why “data foundations” are suddenly strategic, not technical

For years, “data modernization” was treated as an IT line item. In an AI era, it is business strategy. That’s because AI performance and safety are only as strong as the data pipelines feeding models and the controls surrounding them.

Three forces are pushing data foundations to the front of the executive agenda:

1) AI now touches high-stakes workflows. The biggest value pools are clinical documentation, triage, utilization management, coding, denials, staffing, and care navigation — areas where mistakes can harm patients, create compliance exposure, or disrupt revenue.

2) Model monitoring depends on data discipline. If you can’t track data lineage, version datasets, or consistently define outcomes, you can’t reliably detect drift, bias, or failure modes. Without that, “responsible AI” is mostly a slide deck.

3) AI procurement is shifting toward platforms. Health systems are moving from buying isolated tools to selecting enterprise-grade platforms (from EHR vendors, cloud hyperscalers, and specialized startups). Platform bets amplify the importance of clean integration patterns, governance, and shared semantic layers.

What this means for clinicians and frontline teams

For healthcare professionals, the practical implication is simple: the AI tools that feel seamless will be those backed by coherent data. The tools that feel distracting — incorrect summarizations, missing context, odd recommendations, repeated clicks — often reflect upstream data issues more than “bad AI.”

Clinicians should expect a near-term period of uneven experiences. In a single health system, one department may see meaningful documentation relief while another sees limited benefit because its workflows and data capture are different. That can create skepticism and adoption friction, especially if leadership rolls out AI without aligning clinical operations, documentation standards, and governance.

There’s also a workload angle. Poorly integrated AI can shift effort onto clinicians: more time correcting notes, validating suggestions, and navigating exceptions. In contrast, AI built on trusted data can reduce rework and help teams focus on patient care rather than administrative cleanup.

What patients will feel — and what they won’t see

Patients won’t judge AI on benchmarks; they’ll judge it on access, clarity, and continuity. Strong data foundations can translate into tangible patient-facing improvements: faster prior authorizations, fewer billing surprises, smoother transitions between care settings, and better coordination across specialty, primary care, and ancillary services.

But patients may also experience the downside of weak execution: confusing portal messages, inconsistent care plans, delays caused by automation errors, or “AI says no” decisions that are hard to appeal because the logic is opaque. That’s why governance and traceability matter — not just for compliance, but for trust.

The winners won’t just “have AI” — they’ll operationalize it

The most important takeaway from the Healthcare Dive framing is that the next competitive divide won’t be between organizations that tried AI and those that didn’t. Nearly everyone is trying. The divide will be between organizations that built repeatable AI delivery systems — shared data products, standardized evaluation, monitoring, and change management — and those that keep running pilots as isolated projects.

In the coming 12–24 months, expect boards and CEOs to ask more pointed questions: Which AI use cases are scaling across service lines? What data domains are still “untrusted”? What are our measurable quality and safety guardrails? How quickly can we swap models or vendors without breaking workflows?

Healthcare is heading toward an era where AI is not a feature but an operating layer. Organizations that treat data foundations as the prerequisite for safe automation — rather than an afterthought — will move faster, spend less on rework, and earn more clinical confidence. Those that don’t may find that the AI inflection point is real, but not in their favor.

Source: Reporting and framing inspired by Healthcare Dive (“Healthcare’s AI inflection point: The organizations that win will be the ones with the strongest data foundations”).