Editor-in-Chief: Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors: These health AI insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy healthcare executive.

Insights and Agentic AI Action Plans for the week of June 29, 2026 include:

• Enterprise AI Is Becoming a Management System, not a Tool Portfolio

• AI Governance Is Becoming an Operating License, not a Compliance Add-On

• Frontline AI Is Shifting from Assistance to Managed Autonomy

• AI Is Reshaping Healthcare Economics Before It Reduces System Costs

• AI Risk Is Increasingly a Third-Party Resilience Problem

Read more

Patients still fall through the cracks. Referrals remain incomplete. Consults are delayed. Prior authorizations stall. Discharge plans wait for missing information. Staff spend countless hours bridging gaps between disconnected systems.

The next generation of AI value will come from connecting these workflows rather than simply making individual steps faster. The exciting news is that agentic AI provides a practical way to build this orchestration layer.

Think Like Building a Care Team

Hospitals already know how to organize people into multidisciplinary teams. Complex patient care rarely depends on a single clinician. Instead, physicians, nurses, pharmacists, social workers, care coordinators, therapists, and case managers each contribute specialized expertise while collaborating toward a common goal.

Agentic AI should be engineered the same way. Instead of building one large general-purpose agent, organizations should build a team of specialized agents coordinated through a harness and managed by a central orchestration agent.

In this post, we outline the team of AI agents that could be built to assist with workflow integration of optimizing operations. At the end of the post, we provide an even deep dive into the routing agent to provide some detail about how these agents should be constructed. We provide an example prompt, the necessary skills and knowledge, and data sources and tools needed for the agent to perform its task.

Read more

Most conversations about agentic AI focus on the agents themselves. We ask what models they use, what tools they can access, what workflows they automate, and how much productivity they create. But as agents begin taking actions across complex systems, another issue becomes just as important: how should humans supervise them without becoming overwhelmed, disengaged, or reduced to rubber-stamp reviewers?

This is the emerging problem of supervision engineering. In this post, we explain supervision engineering and provide some examples.

Read more

At its core, loop engineering is the design and optimization of iterative cycles that allow agents to improve their work before producing a final result or taking action. Rather than treating AI as a one-shot prediction engine, loop engineering treats intelligence as a process of continuous refinement.

This idea may sound simple, but it represents a profound shift in how AI systems are designed. In many cases, the performance gains associated with modern agentic AI come not from larger models but from better loops.

What Is Loop Engineering?

Loop engineering is the practice of designing, managing, and optimizing the cycles of generation, evaluation, critique, revision, and validation that occur within an agentic workflow. A key feature is that agents generate the prompts rather than humans.

An agentic AI loop often begins with a prompt engineered by specialist prompting agents, but it does not end there. A typical agentic loop may look something like this:

1. Generate the initial prompts

2. Produce a draft output

3. Evaluate the output

4. Critique weaknesses

5. Revise the output

6. Validate quality

7. Repeat 1-7 if quality criteria and unit tests not met

8. Deliver the final result

The goal is not simply to obtain an answer. The goal is to obtain a trustworthy and accurate answer. Just as human experts rarely solve complex problems in a single attempt, agents can often improve their performance through structured iteration.

Read more

Editor-in-Chief - Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors - These health AI insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy healthcare executive.

Insights for the week of June 22, 2026 include:

• AI scale is becoming an enterprise management discipline, not a technology rollout

• Workflow integration is overtaking point automation as the real source of AI value

• AI-enabled documentation is becoming both a financial lever and a contracting flashpoint

• Clinical AI performance is improving faster than enterprise evaluation methods

Read more

Much of the industry remains focused on comparing GPT, Claude, Gemini, Llama, and other foundation models. While model selection certainly matters, organizations are beginning to discover that models alone are not enough. Even the most capable large language model is surprisingly limited if it cannot access the systems, data, and tools required to perform useful work.

This is where the Model Context Protocol, or MCP, enters the picture.

For healthcare executives trying to understand agentic AI, MCP may be one of the most important developments to emerge in the past year. While it receives far less attention than new models, MCP has the potential to dramatically expand what AI agents can accomplish by giving them standardized access to tools, data, and enterprise systems.

In many ways, MCP is helping transform AI from something that merely talks into something that can actually work.

Why Agents Need Tools

Imagine hiring a highly intelligent employee and placing them in an empty room. They have no computer, no phone, no internet connection, no access to databases, no scheduling system, no email, and no organizational knowledge.

Despite being intelligent, they would struggle to accomplish meaningful work. The same is true for AI.

Large language models are powerful reasoning engines, but they are often disconnected from the systems where work actually occurs. They may understand how a hospital operates, but they cannot schedule appointments. They may understand insurance appeals, but they cannot access payer portals. They may understand clinical research, but they cannot retrieve data from a warehouse.

To become useful digital workers, agents need tools. Just as humans use computers, phones, EHRs, calculators, spreadsheets, and search engines, agents need access to external capabilities that allow them to interact with the world.

The Problem Before MCP

Before MCP, connecting agents to tools was often messy. Every AI platform developed its own way of connecting to external systems. Every software vendor created proprietary integrations. Organizations repeatedly built custom connectors to databases, APIs, file systems, and applications. As a result, AI ecosystems became fragmented.

An agent built on one platform might not easily access tools developed for another platform. Organizations often found themselves rebuilding the same integrations repeatedly. The situation resembled the early days of computing before common networking standards emerged. What the industry needed was a common language.

What Is MCP?

The Model Context Protocol (MCP) is an open standard that allows AI models and agents to connect to external tools, data sources, and services through a consistent interface.

Originally introduced by Anthropic in late 2024, MCP has rapidly gained support across the AI ecosystem. Major AI platforms, open-source frameworks, and software vendors are beginning to adopt MCP as a standard method for exposing tools to agents.

At a high level, MCP defines a standard way for agents to discover available tools, understand what they do, send requests, and receive results. Think of MCP as a USB port for AI agents. Rather than creating a custom integration every time a new tool is needed, organizations can expose tools through MCP and allow many different agents to use them. The result is a growing ecosystem of reusable capabilities.

What Does an MCP Tool Look Like?

An MCP tool can be almost anything.

Examples include:

• Search engines

• Email systems

• Calendars

• Databases

• EHR systems

• Scheduling platforms

• Knowledge bases

• Financial systems

• Analytics platforms

• Research repositories

• Cloud services

• Internal applications

When an agent needs information or needs to perform an action, it can invoke an MCP tool rather than relying solely on its internal knowledge. This dramatically expands what agents can accomplish.

Healthcare Examples

Healthcare may ultimately become one of the largest adopters of MCP-enabled agents because clinical and operational work spans so many disconnected systems.

Consider a patient navigation agent. Through MCP, the agent could potentially access:

• EHR records

• Appointment scheduling systems

• Provider directories

• Insurance eligibility services

• Referral management platforms

• Patient messaging systems

Rather than merely answering questions, the agent could coordinate care across multiple systems.

Similarly, a clinical research agent could access:

• Data warehouses

• Research protocols

• Literature databases

• Statistical analysis tools

• Knowledge graphs

• Regulatory documentation

The agent becomes a coordinator of resources rather than simply a conversational interface.

Share

Why MCP Matters for Agentic AI

Agentic AI differs from traditional chatbots because agents perform work. Work requires tools. A scheduling agent needs scheduling tools. A financial agent needs financial tools. A research agent needs research tools. A clinical operations agent needs clinical operations tools.

This is why MCP is so important. It provides a standardized mechanism for connecting intelligence to capability. Without tools, agents remain largely informational. With tools, agents become operational.

What Executives Should Watch

Healthcare executives should not think of MCP as a technical standard. They should think of it as an infrastructure layer for digital workers. The strategic question is not whether your organization will adopt MCP specifically. The strategic question is whether your organization is preparing for a future in which agents interact directly with enterprise systems.

Organizations will increasingly deploy agents that:

• Schedule appointments

• Process referrals

• Review claims

• Coordinate care

• Monitor quality metrics

• Conduct research

• Generate reports

• Support operations

All of these workflows require access to tools. MCP is rapidly emerging as one of the leading mechanisms for providing that access.

It is the tools, skills, and knowledge that move agentic AI well beyond the obvious limitations of large language models.

These are important questions, but they are not the most important questions.

The more important question is whether we can trust these systems to operate safely, accurately, and consistently in environments where mistakes can directly affect patient outcomes. In healthcare, an incorrect recommendation is not merely an inconvenience. It can delay treatment, compromise patient safety, violate privacy regulations, or erode trust between patients and providers.

This is why guardrails have become one of the most important concepts in modern agentic AI engineering.

What Is a Guardrail?

A guardrail is a constraint, policy, rule, or validation mechanism designed to guide agent behavior and prevent undesirable actions. Think of guardrails on a mountain road. The goal is not to prevent movement. The goal is to ensure movement occurs safely within acceptable boundaries.

Similarly, AI guardrails do not eliminate agent autonomy. Rather, they define the limits within which agents can operate.

Examples of healthcare guardrails might include:

• Prohibiting agents from making independent diagnoses

• Preventing access to protected health information without authorization

• Restricting medication recommendations to approved clinical guidelines

• Requiring human approval before contacting patients

• Escalating uncertain situations to clinicians

• Preventing actions that exceed defined confidence thresholds

As agents become more autonomous, guardrails become increasingly important.

In the rest of this post we discuss why guardrails matter, how to build a guardrail, and provide a healthcare example. We then discuss why guardrails aren’t enough.

Read more

The real challenge is no longer generating intelligence. It is engineering intelligence.

As organizations deploy increasingly sophisticated agentic AI workflows, success depends less on the underlying model and more on how agents are designed, organized, governed, evaluated, and integrated into existing business processes. This emerging discipline is often referred to as agentic AI engineering, and it may ultimately become as important as software engineering itself.

To help organizations navigate this rapidly evolving landscape, I created the Best Practices for Agentic AI Engineering Cheat Sheet. Think of it as a practical guide for building reliable, scalable, and trustworthy digital workers.

The cheat sheet is provided below along with a description of the major topics.

Read more

Most organizations today focus on models. They debate whether to use GPT, Claude, Gemini, Llama, or another foundation model. Yet as agentic AI matures, many organizations are discovering that the model itself is often not the primary determinant of performance. Instead, success increasingly depends on the quality of the skills that agents possess.

Just as human employees rely on training, experience, and specialized expertise, AI agents rely on skills. These skills define how agents perform tasks, interact with tools, access data, follow organizational policies, and generate outputs. Developing these capabilities requires expertise that extends beyond traditional software engineering or prompt engineering. It requires a new professional role focused on transforming organizational knowledge into reusable digital expertise.

Enter the Agentic AI Skills Engineer.

What Is an Agentic AI Skills Engineer?

An Agentic AI Skills Engineer is responsible for designing, developing, testing, maintaining, and optimizing the skills used by AI agents.

A skill is much more than a prompt. It is a structured package of expertise that may include instructions, reasoning frameworks, tool usage guidance, data access patterns, quality checks, safety mechanisms, guardrails, ethics guidelines, domain knowledge, evaluation criteria, escalation rules, and organizational policies.

For example, a clinical literature review skill might teach an agent how to search biomedical databases, assess study quality, identify biases, summarize findings, and generate evidence-based conclusions. A referral management skill might teach an agent how to evaluate referral requests, identify missing information, prioritize urgency, and route cases appropriately.

The Skills Engineer transforms institutional expertise into reusable digital assets that can be shared across many agents and workflows.

In this deeper dive we discuss why this role is needed and cover key qualifications. A detailed job description is presented along with evaluation criteria.

Read more

Editor-in-Chief: Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors: These health AI insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy healthcare executive.

Insights for the week of June 15th, 2026 include:

• Ambient Documentation Is Emerging as Healthcare’s AI Beachhead, but the Real Value Is Cognitive Relief, Not Just Time Saved

• AI Governance Is Shifting From Policy Language to Procurement Discipline and Evidence Thresholds

• Administrative AI Is Creating a New Payer-Provider Battlefield Over Coding Intensity, Affordability, and Margin

• Cybersecurity and AI Governance Are Converging Into a Single Operational Resilience Agenda

Read more

History suggests that transformative technologies rarely eliminate work altogether. Instead, they change the nature of work and create new specialties. The rise of electronic health records created clinical informaticians, EHR trainers, implementation specialists, and health IT leaders. Cloud computing created cloud architects, DevOps engineers, and cybersecurity specialists.

Agentic AI is likely to follow a similar path.

As organizations move beyond chatbots and begin deploying coordinated teams of specialized agents, entirely new roles will emerge focused on designing, managing, governing, evaluating, and optimizing digital workers. The future workforce may include not only human employees but also large populations of AI agents requiring oversight, coordination, and continuous improvement.

In this Looking to the Future post, we provide details of more than 20 new skilled positions that will be needed in the agentic AI era.

Read more

Healthcare is no exception. Across the industry, leaders are exploring agentic AI for patient navigation, scheduling, documentation, care coordination, revenue cycle management, research support, and operational optimization. However, relatively few health systems have successfully scaled these capabilities beyond isolated demonstrations. The technology appears increasingly capable, but implementation continues to lag.

The question is why.

The conventional explanation focuses on technology. Organizations are often told they need better models, more mature platforms, or additional vendor solutions. While these factors matter, they may not be the primary barrier. The deeper challenge is that agentic AI is fundamentally a workflow problem rather than a software problem. Success depends not only on the intelligence of the agents themselves, but on how well those agents understand the specific operational environment in which they are expected to function.

This distinction is particularly important in healthcare. Clinical and operational workflows vary considerably across institutions. Referral processes, scheduling rules, escalation pathways, governance structures, staffing models, and regulatory requirements are rarely standardized. Even organizations using the same electronic health record frequently implement workflows in dramatically different ways. As a result, a solution that performs well in one institution may struggle in another despite appearing identical on the surface.

Many vendors underestimate this complexity. Most agentic AI products are designed to be broadly applicable across organizations. However, healthcare operations are highly contextual. Effective patient navigation at one institution may depend on local scheduling practices. Care coordination workflows may reflect unique staffing structures. Escalation procedures often evolve from years of operational experience. These nuances are difficult for external vendors to fully understand, yet they are often essential for success.

This helps explain why many organizations remain stuck in pilot mode. The technology itself may function as intended, but the workflow assumptions embedded within the solution fail to align with local reality. What appears to be a technology problem is often a workflow design problem.

For this reason, healthcare organizations should consider developing internal agent engineering capabilities rather than relying exclusively on external vendors. One promising model is the creation of an Agentic AI Rapid Response Team (ARRT): a small, multidisciplinary group focused on identifying workflow bottlenecks, developing agentic solutions, and rapidly deploying and evaluating targeted interventions.

The purpose of an ARRT is not to build enterprise-scale software platforms. Instead, its mission is to translate organizational knowledge into operationally useful agentic systems. By working directly with clinicians, operational leaders, researchers, and information technology teams, the ARRT develops a detailed understanding of local workflows and designs agents that operate within existing institutional structures.

This local knowledge becomes a strategic advantage. Unlike external vendors, internal teams understand how referrals are processed, how patient messages are escalated, how operational decisions are made, and where bottlenecks actually occur. They understand the organization’s culture, governance processes, ethical standards, staffing limitations, and performance goals. These insights provide essential context for designing agents that can meaningfully improve operations.

Equally important, internal teams are uniquely positioned to engineer agents with the right skills. Many current AI systems function as general-purpose assistants. While useful, they often lack the domain-specific expertise required for complex healthcare workflows. High-performing agentic systems increasingly depend on specialized skills such as referral review, prior authorization support, care gap identification, scheduling optimization, literature review, policy interpretation, and quality improvement analysis.

Over time, these skills become institutional assets. They capture organizational knowledge in reusable forms that can be shared across multiple workflows and service lines. Rather than repeatedly reinventing solutions, organizations can build libraries of validated skills that reflect their own best practices and standards.

The same principle applies to tools. Agents derive much of their value from their ability to interact with enterprise systems. A referral management agent may require access to scheduling systems, provider directories, communication platforms, and referral queues. A research agent may need access to PubMed, institutional data repositories, and analytical tools. Determining which tools agents require, and under what circumstances they may use them, is a local engineering challenge that is difficult to outsource effectively.

Perhaps most importantly, internal teams are best positioned to implement appropriate governance mechanisms. Healthcare organizations operate under unique ethical, legal, and regulatory constraints. Decisions about privacy, patient safety, bias mitigation, accountability, and human oversight cannot simply be delegated to vendor defaults.

An effective ARRT can address these concerns by embedding governance directly into agentic workflows. Critic agents can review outputs for quality and consistency. Safety agents can identify potential patient risks. Privacy and compliance agents can evaluate regulatory concerns. Ethics agents can assess fairness, transparency, and unintended consequences. Human review checkpoints can be integrated at critical decision points. Together, these mechanisms create a system of checks and balances that reflects the institution’s own values and governance framework.

The practical advantages of this approach are significant. Rather than spending months evaluating large-scale enterprise solutions, an ARRT can rapidly identify a high-friction workflow, develop a prototype, deploy a narrowly focused team of agents, and measure results. Early successes create organizational confidence while generating reusable skills, governance patterns, and implementation experience that can be applied to future projects.

This incremental approach may prove more effective than attempting to deploy large, generalized agentic platforms from the outset. Healthcare organizations rarely struggle because they lack AI. More often, they struggle because they lack a mechanism for translating AI capabilities into operational improvements. An ARRT provides that mechanism.

The broader implication is that the future of agentic AI in healthcare may depend less on acquiring technology and more on developing institutional capability. The organizations that realize the greatest value from agentic AI are unlikely to be those that simply purchase the most advanced platforms. Instead, they may be those that learn how to engineer, govern, and continuously refine teams of digital workers that reflect their own workflows, standards, and strategic priorities.

The challenge facing healthcare is therefore not a shortage of agentic AI solutions. It is the absence of organizational structures capable of adapting those solutions to local needs. Agentic AI Rapid Response Teams offer one practical path forward. By combining workflow expertise, agent engineering, governance, and rapid experimentation, these teams may become an essential component of the healthcare organization of the future.

https://www.theregister.com/ai-and-ml/2026/06/05/agentic-ai-hype-races-ahead-as-enterprises-remain-stuck-in-pilot-mode/5251711

They ask which model is being used.

Is it GPT? Claude? Gemini? Does it have a large context window? How does it perform on benchmarks?

These questions are understandable, but they often miss what determines whether an agentic system succeeds when applied to complex problems. As models become increasingly capable and interchangeable, the differentiator is shifting away from model selection and toward agent and harness engineering. The quality of an agentic system increasingly depends on how agents are organized, what skills they are given, how work is divided, what tools are available, how decisions are validated, and how governance is enforced.

In other words, the question is no longer simply whether the AI is intelligent. The question is whether the system is well managed.

We present here 12 questions every healthcare executive should ask when presented with an agentic AI solution or plan.

1. Has the Problem Been Properly Decomposed?

Every successful agentic workflow begins with problem decomposition.

One of the most common mistakes in agent engineering is asking a single agent to solve an entire problem from start to finish. Complex work rarely happens that way in human organizations, and it rarely works well in agentic systems either. Research, analysis, verification, communication, and decision-making are often distinct activities that benefit from specialization.

When evaluating a workflow, ask how the problem was broken apart and why those particular boundaries were chosen. The answer often reveals whether the team has thoughtfully designed the workflow or simply wrapped a large language model around an operations process.

Read more

Editor-in-Chief - Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors - These Agentic AI Strategy Insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy manager.

Insights for the week of June 8th, 2026 include:

• Operational Precision Is Becoming the Primary Defense Against Reimbursement Volatility

• AI Adoption Will Be Limited More by Trust Architecture Than by Model Capability

• The Real Staffing Strategy Is Workflow Capacity Creation

• Digital Dependence Is Raising the Cost of Third-Party Risk

Strategic Insight 1

Operational Precision Is Becoming the Primary Defense Against Reimbursement Volatility

Hospitals are being pushed toward more precise, data-driven operational management because blunt cost cutting increasingly damages the capabilities needed to recover margin. In an environment of reimbursement uncertainty, rising labor and supply costs, and rating-agency scrutiny, operational control has become a strategic financial capability.

For leadership, this means margin defense is no longer just about budgets. It depends on a hospital’s ability to manage denials, patient flow, asset utilization, staffing efficiency, and service-line performance in near real time. The CFO and COO increasingly need a shared operating view, because capital access, credit perception, and financial resilience are becoming linked to how well the institution can demonstrate operational controllability.

The forward implication is that health systems with stronger operational instrumentation will be better positioned to absorb payer shocks, preserve flexibility, and allocate capital more intelligently. The benefit is not only cost containment, but stronger resilience, better access protection, and improved credibility with boards and lenders.

Executive Action Plan

Stand up a Margin Control Tower Lite for 60 days. Publish one shared weekly dashboard for the CFO, COO, and service-line leaders using only 8 measures: denials trend, days in A/R, overtime, agency usage, asset rental spend, discharge delay hours, avoidable manual touches, and service-line contribution drift. Keep it simple, visible, and action-oriented. This can create cross-functional discipline without the cost or delay of launching a full command center.

Agentic AI Plan

Use a small multi-agent operating review assistant:

• a finance variance agent to explain weekly margin movement

• an asset utilization agent to identify underused or rented equipment

• a throughput agent to detect discharge or bed-flow bottlenecks

• a denials agent to summarize avoidable denial patterns and draft owner assignments

This approach can automate much of the weekly review packet and make shorter operating cycles practical for executive teams.

Source References

• HFMA — OBBBA Medicaid cuts increase credit risk for NFP hospitals

• Healthcare Business Today — Beyond budget cuts: A smarter way to respond to volatility in healthcare reimbursement rates

• MarketsandMarkets Blog — AI in Hospital Operations Market Growth 2030: Transforming Healthcare Efficiency with Automation, NLP & Predictive Analytics

Read more

The problem is rarely a lack of insight. The problem is execution.

Patients move between primary care, specialists, scheduling teams, call centers, patient navigators, care managers, and multiple technology systems. At each handoff, opportunities for delay, confusion, and patient leakage emerge. The result is a fragmented experience in which everyone knows what should happen next, but no one is consistently responsible for making it happen.

This is where agentic AI becomes useful. Rather than functioning as another departmental productivity tool, agentic AI can serve as a care progression layer that continuously monitors patient journeys, identifies breakdowns, and coordinates actions across people and systems. The goal is not to replace staff. The goal is to ensure that care keeps moving forward.

Start with One Pathway

The fastest way to demonstrate value is to focus on a single high-value service line such as oncology intake or cardiology referrals. The objective is to identify where patients stall and create a lightweight agentic workflow that continuously monitors those gaps.

For a pilot, track only three metrics:

• Time to next appointment

• Referral completion rate

• Patient drop-off rate

These measures are simple, meaningful, and directly tied to operational performance.

The Agent Team

A common mistake is to build one giant agent responsible for everything. A much more effective approach is to create a small team of specialized agents coordinated by a manager agent. We present below a detailed agentic AI plan.

Read more

Across hospitals and health systems, clinicians, administrators, and operational leaders encounter dozens of repetitive tasks every day that could potentially be improved with agentic AI. Documentation bottlenecks, inbox overload, referral coordination, scheduling inefficiencies, prior authorization delays, policy retrieval, and reporting workflows are all ripe for improvement. Yet most organizations lack a clear mechanism for identifying these opportunities, rapidly building solutions, and safely deploying them.

Many health systems respond by launching large AI initiatives, creating lengthy governance processes, or pursuing expensive vendor engagements. A more practical approach may be to start small.

Rather than building a large AI department, health systems should consider creating an Agentic AI Rapid Response Team (ARRT). The ARRT is a small, highly skilled group focused on rapidly identifying operational pain points, building lightweight solutions, and teaching departments how to solve similar problems themselves.

The goal is not to centralize AI development. The goal is to accelerate innovation.

We outline here a plan for building a lean ARRT that can engineer teams of agents in hours.

Read more

Editor-in-Chief: Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors: These health AI insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy healthcare executive.

Insights for the week of June 1st, 2026 include:

• Revenue Cycle AI Is Becoming a Competitive Necessity, Not an Experiment

• Governance Is Becoming the Real Barrier to AI Scale

• AI Orchestration Will Outperform Point-Solution AI

Read more

For decades, healthcare organizations have built security programs around human identities. Employees receive credentials, role-based access, training, audits, and oversight. Identity and accountability are tightly connected. If a physician signs an order, a financial analyst approves a transaction, or a scheduler modifies an appointment, the organization knows who performed the action and can apply policies, supervision, and accountability accordingly.

Agentic AI changes this model. Digital workers can access multiple systems, invoke tools, retrieve information, trigger workflows, interact with patients, and operate continuously across organizational boundaries. Unlike a human employee, an agent can combine permissions from multiple systems, perform actions at machine speed, and persist indefinitely without fatigue. As organizations move agents from answering questions to taking actions, a new governance challenge emerges.

The problem is no longer simply who has access. The problem is what they are allowed to do.

In this deeper dive we discuss AI agent passports as way to expand your institutional governance structure to be able to trust digital entities working autonomously in your sensitive data ecosystems.

Read more

This makes them ideal targets for agentic AI.

Instead of asking, “What is the most advanced AI use case we can deploy?” health systems should ask a more practical question:

Where is work getting stuck, how much is it costing us, and which AI intervention can produce measurable value within 90 days?

That is the purpose of an AI Friction Index.

Goal

Build a lightweight agentic AI system that helps a hospital identify, rank, and prepare operational AI opportunities across five workflows:

1. Clinical documentation

2. Scheduling and access

3. Denials and revenue cycle

4. Patient messaging

5. Pharmacy operations

The system should identify bottlenecks using four simple metrics:

• weekly volume

• labor time

• rework rate

• downstream cost

The output should be a ranked list of two or three AI opportunities with clear owners, measurable outcomes, and a governance-ready implementation brief.

Recommended Agentic AI Team

1. Manager Agent

The Manager Agent coordinates the workflow. It receives submissions, assigns tasks to specialist agents, checks that each output is complete, and produces the final executive summary.

Its role is not to make the final decision. Its role is to organize the evidence so leaders can make faster, better decisions.

2. Workflow Discovery Agent

This agent reviews pain-point submissions and operational data. It looks for repeatable bottlenecks in inboxes, denial logs, scheduling queues, call-center tickets, pharmacy exception reports, help desk tickets, and documentation burden reports.

Its output should include:

• workflow name

• bottleneck description

• affected staff

• weekly volume

• estimated manual time

• rework rate

• downstream operational impact

3. Clustering Agent

The Clustering Agent groups similar pain points. For example, multiple submissions about appointment rescheduling, referral delays, and prior authorization delays may all reflect a broader access-management problem.

This prevents leadership from reviewing dozens of fragmented complaints and instead helps them see enterprise patterns.

This would require adding a database or spreadsheet such as Airtable to the n8n workflow shown above along with the Clustering Agent to analyze the data.

4. ROI Estimator Agent

The ROI Estimator Agent converts workflow friction into rough value estimates.

It should estimate:

• labor hours consumed

• avoidable rework

• revenue leakage

• turnaround-time impact

• staffing burden

• potential savings or protected revenue

• expected payback period

The goal is not false precision. The goal is directional prioritization.

5. Governance Prep Agent

The Governance Prep Agent prepares each candidate use case for review by IT, compliance, privacy, legal, clinical leadership, and operations.

It should identify:

• data needed

• systems touched

• vendor involvement

• privacy risks

• cybersecurity concerns

• human oversight requirements

• monitoring needs

• implementation complexity

6. Ethics Agent

The Ethics Agent reviews whether the proposed AI intervention could create unfairness, inequity, opacity, or inappropriate automation.

It should ask:

• Could this workflow affect vulnerable patients differently?

• Could automation worsen access or communication disparities?

• Is there a risk of patients being deprioritized unfairly?

• Are staff roles being changed without appropriate transparency?

• Does the use case require patient-facing disclosure?

7. Safety Agent

The Safety Agent focuses on operational and clinical safety.

It should ask:

• Could an error delay care?

• Could the AI misroute or suppress an important message?

• Could a pharmacy exception or denial be handled incorrectly?

• Where must a human remain in the loop?

• What escalation pathway is required?

• What should trigger shutdown or review?

8. Implementation Agent

This agent converts the approved use case into a practical execution plan.

It defines:

• operational owner

• technical owner

• pilot scope

• data sources

• workflow integration points

• success metrics

• timeline

• training needs

• monitoring plan

Step-by-Step Workflow

Step 1: Collect workflow pain points

Create a simple intake form for operational leaders.

Ask for:

• workflow area

• description of bottleneck

• weekly volume

• staff time required

• rework rate

• downstream cost or consequence

• systems involved

• current workaround

• desired improvement

This can be done with Microsoft Forms, Google Forms, Airtable, REDCap, ServiceNow, or an n8n form trigger.

Step 2: Feed submissions to the Manager Agent

The Manager Agent checks whether each submission is complete. If key fields are missing, it generates follow-up questions for the operational owner.

Step 3: Run the Workflow Discovery Agent

The Discovery Agent turns each submission into a structured operational profile.

Example output:

{
  "workflow": "Patient messaging",
  "bottleneck": "High volume of routine portal messages requiring manual triage",
  "weekly_volume": 4200,
  "labor_time": "85 staff hours/week",
  "rework_rate": "18%",
  "downstream_cost": "delayed response, clinician inbox burden, patient dissatisfaction"
}

Step 4: Cluster related bottlenecks

The Clustering Agent groups submissions into strategic categories such as:

• access friction

• documentation burden

• denial rework

• inbox overload

• pharmacy exception handling

• duplicate data entry

• avoidable call-center volume

This allows leadership to see where friction is systemic rather than isolated.

Step 5: Estimate value

The ROI Estimator Agent ranks opportunities using a simple score:

• volume

• labor intensity

• rework burden

• financial impact

• feasibility

• time to measurable result

The output should be directional:

• high-value / low-complexity

• high-value / high-complexity

• low-value / low-complexity

• low-value / high-complexity

The first 90-day pilots should come from the high-value / low-complexity quadrant.

Step 6: Run ethics and safety review

Before recommending any AI intervention, the Ethics Agent and Safety Agent review the candidate.

They should flag:

• patient harm risk

• equity concerns

• privacy exposure

• unclear accountability

• excessive automation

• lack of human override

• insufficient monitoring

Any use case with high safety or ethics risk should require additional review before pilot approval.

Step 7: Generate a governance-ready one-page brief

The Governance Prep Agent creates a concise approval document for the COO, CFO, CIO, CMIO, CAIO, and relevant operational leader.

Each brief should include:

• problem statement

• baseline metrics

• proposed AI intervention

• expected value

• affected systems

• privacy/security considerations

• safety/ethics concerns

• human oversight model

• owner

• 90-day success metrics

Step 8: Select two or three pilots

Leadership should approve only a small number of pilots.

Good early candidates might include:

• portal message triage

• denial letter summarization

• scheduling backlog prioritization

• documentation draft support

• pharmacy exception routing

• call-center knowledge retrieval

Avoid pilot sprawl. The objective is measurable impact, not AI theater.

Step 9: Implement with n8n or Claude Code

A simple n8n implementation could include:

• form trigger for submissions

• database or spreadsheet for storage

• AI Agent nodes for discovery, ROI, governance, ethics, and safety review

• Slack or email approval workflow

• dashboard output in Airtable, Notion, Google Sheets, or Power BI

A Claude Code implementation could create a lightweight internal application with:

• intake form

• database

• agent prompts

• scoring logic

• review dashboard

• exportable one-page briefs

The n8n version is faster to prototype as shown in the workflow above. The Claude Code version may be better once the workflow stabilizes. Note that n8n workflows can be created using Claude Code using an n8n MCP server.

Step 10: Monitor outcomes

Each pilot should track only a few metrics:

• baseline volume

• labor hours saved

• turnaround time

• rework rate

• error rate

• user satisfaction

• financial impact

If the pilot does not show measurable improvement within 90 days, stop or redesign it.

Why This Matters

This approach changes AI strategy from a technology-selection exercise into an operational improvement discipline.

Instead of asking executives to fund speculative AI projects, it gives them a practical mechanism to identify where AI can reduce friction, protect revenue, improve access, and relieve staff burden.

The real value of agentic AI is not just that it can automate tasks. It can help organizations make better decisions about where automation should happen first.

In a margin-constrained health system, that may be the difference between AI experimentation and AI transformation.

Editor-in-Chief: Jason H. Moore, PhD, FACMI, FAMIA, FIAHSI, FASA

Authors: These health AI insights were researched, analyzed, written, critiqued, edited, and communicated by a collaborative team of 10 expert AI Agents for the busy healthcare executive.

Insights for the week of May 25th, 2026 include:

• AI Value Will Be Won in Operations First, Not in Moonshot Clinical Use Cases

• Clinical AI Is Becoming a Capacity Management Problem More Than a Diagnostic Problem

• AI Governance Is Merging with Enterprise Risk and Cyber Resilience

• Margin Pressure Is Forcing a More Disciplined, Evidence-Based AI Investment Model

• AI Adoption Will Stall Unless Hospitals Treat Trust as an Operational Asset

Strategic Insight 1

AI Value Will Be Won in Operations First, Not in Moonshot Clinical Use Cases

The most durable near-term AI value in hospitals is emerging in operational workflows, not in headline-grabbing autonomous clinical use cases. The pattern is clear across documentation, scheduling, revenue cycle, pharmacy operations, and patient communications: AI is most useful where work is repetitive, digitally observable, labor-intensive, and financially measurable. That matters because most health systems do not currently have the margin, staffing depth, or governance maturity to scale speculative AI bets with long timelines and uncertain payoff.

For executive leadership, this means the right AI strategy is not to chase the most advanced model, but to redesign the operating core of the enterprise. Operational AI should be viewed as a force multiplier for margin recovery, clinician retention, throughput, patient access, and future digital readiness. Leaders that sequence AI through operational wins can create a self-funding cycle in which early gains produce the trust, liquidity, and change capacity needed to support more advanced clinical AI later.

Executive Action Plan

Create a 90-day AI Friction Index across five workflows: documentation, scheduling, denials, patient messaging, and pharmacy operations. Have each function identify one measurable bottleneck using four metrics only: weekly volume, labor time, rework rate, and downstream cost. Approve only the two or three AI interventions with the shortest time to measurable enterprise impact and a named operational owner. This creates focus, prevents pilot sprawl, and ties AI directly to performance improvement.

Agentic AI Plan

Ideas for Accelerating Change with Agentic AI

A hospital IT or digital innovation team could assemble a lightweight internal triage system in a few hours using:

• a Workflow Discovery Agent to review inbox categories, denial logs, scheduling backlogs, help desk tickets, and pharmacy exception queues;

• an ROI Estimator Agent to convert workflow pain into labor savings, revenue protection, and turnaround-time improvement;

• a Governance Prep Agent to produce a concise risk, privacy, and oversight brief for each candidate use case.

Workflow:

• Operational leaders submit workflow pain points in a simple form.

• The discovery agent clusters and ranks bottlenecks.

• The ROI agent estimates value and payback timing.

• The governance agent creates a one-page review summary for COO, CFO, CIO, and CAIO approval.

This makes AI prioritization faster, less political, and more evidence-based without a major consulting engagement.

Source References

• How Hospitals are Implementing AI for Smarter Healthcare Services

• AI is about to break healthcare’s scarcity model — if we let it

• Hospitals of all sizes are using AI to improve revenue cycle operations

• A doctor shortage is coming. AI could be the only realistic fix

Read more