ILLUSTRATIVE SCENARIOS
The scenarios below are hypothetical composites that show how Birch is designed to work in typical practices. The practices, individuals, quotes, and figures are illustrative examples — not reports from actual customers.
Healthcare administrative burden has reached a crisis point. Physicians spend nearly two hours on paperwork for every hour of patient care, contributing to widespread burnout and decreased quality of care. But a quiet revolution is underway: AI agents are beginning to automate the repetitive, time-consuming tasks that have long plagued healthcare workers.
Birch's AI agents are built for practices spanning primary care, women's health, behavioral health, and specialty medicine. The goal is transformative — not just in efficiency metrics, but in the fundamental experience of healthcare delivery for both providers and patients.
This article walks through three illustrative scenarios that show how AI agents can reshape healthcare workflows today. But first, it's important to understand what makes these AI agents different from traditional automation—and why they're succeeding where previous technologies have failed.
Understanding AI Agents: Beyond Simple Automation
When most people hear "automation" in healthcare, they think of simple rule-based systems: if a patient checks this box, send that email. If a lab result exceeds this threshold, flag it for review. These systems have existed for decades, and while useful, they're fundamentally limited. They can only handle predetermined scenarios and break down when faced with anything unexpected.
AI agents represent a fundamentally different paradigm. Rather than following rigid if-then rules, they understand context, make decisions, and execute complete workflows autonomously. A recent study by researchers at Carnegie Mellon and Stanford Universities compared how AI agents and human workers approach the same tasks across multiple domains—and the findings reveal both the promise and limitations of this technology.1
The Programmatic Approach: How Agents Think
One of the most striking findings from workflow research is that AI agents take an overwhelmingly programmatic approach to work—even for tasks that humans typically complete through graphical user interfaces. When asked to create a presentation or design a website, humans open PowerPoint or Figma and work visually. AI agents, remarkably, write programs to generate the same outputs.
This programmatic bias extends across virtually all computer-based activities, with agents using programming tools 93.8% of the time. In healthcare, this means that when a Birch agent needs to extract patient information from multiple systems and compile it into a prior authorization form, it doesn't navigate through screens the way a human would—it directly queries databases, processes the information programmatically, and generates the completed form.
This approach offers significant advantages: it's faster, more reliable, and less prone to the kind of errors that occur when humans manually copy information between systems. An agent can process an entire prior authorization in 8 minutes versus the 45 minutes a human requires—not because it types faster, but because it takes a fundamentally more efficient path through the workflow.
The Augmentation vs. Automation Distinction
Research on human-AI collaboration reveals a critical distinction between two ways of deploying AI in workflows: augmentation and automation. This distinction has profound implications for healthcare implementations.
AI augmentation involves integrating AI agents into existing workflows with minimal disruption. The human worker maintains control and uses the AI as a tool for specific steps. For example, a nurse might use an AI agent to quickly summarize a patient's medication history before a consultation, but still reviews the summary and makes the final clinical decisions. Studies show that augmentation accelerates human work by 24.3% while maintaining 76.8% workflow alignment—meaning the fundamental way people work doesn't change dramatically.
AI automation, by contrast, hands entire workflows over to AI agents. The human's role shifts from doing the work to reviewing and debugging AI outputs. Surprisingly, research shows that full automation can actually slow work down by 17.7% compared to human workers alone. Why? Because humans spend significant time verifying AI outputs, catching errors, and fixing problems—often taking longer than if they'd done the task themselves.
The most successful healthcare implementations of Birch agents follow the augmentation model for complex clinical decisions while using full automation for well-defined administrative tasks. Prior authorizations, appointment scheduling, and routine patient communication are ideal candidates for automation because they follow clear protocols and have objective success criteria. Clinical decision-making, care planning, and complex patient interactions benefit more from augmentation.
Speed and Cost Advantages
The efficiency gains from AI agents are substantial. Across multiple work domains, agents complete tasks 88.3% faster than human workers and at 90.4-96.2% lower cost. In healthcare specifically, we project even larger improvements in certain workflows:
- Prior authorization processing: 82% faster (45 minutes to 8 minutes)
- Patient intake screening: 73% faster (45 minutes to 12 minutes)
- Routine patient communication: 95% faster (response time from hours to minutes)
- Insurance verification: 78% faster (20 minutes to 4 minutes)
These aren't just incremental improvements—they represent a fundamental shift in what's possible for healthcare operations. A practice that once needed two full-time staff members to handle prior authorizations can now process the same volume with AI agents running in the background, freeing those team members to focus on direct patient care.
Current Limitations and Honest Assessment
It's equally important to understand what AI agents can't do well—at least not yet. Research comparing human and agent work quality reveals several consistent limitations:
Visual understanding challenges: While humans excel at quickly parsing visual information—reading a scanned bill, understanding a hand-drawn diagram, or interpreting a complex chart—agents struggle with these tasks. In healthcare, this means agents work best with structured, digital data from EHRs and databases, but may have difficulty with faxed documents, handwritten notes, or visual assessments.
Context and nuance: Agents can miss subtle contextual cues that humans naturally pick up on. A human reading a patient message might detect underlying anxiety or confusion that warrants a phone call rather than a text response. Agents excel at following protocols but can struggle with the judgment calls that experienced healthcare workers make instinctively.
Format transformation: Agents work best with programmatic data formats (JSON, CSV, database records) but can struggle when needing to transform information into or from human-friendly formats. A human can easily convert a conversation into a structured note; an agent might produce technically correct but awkwardly formatted output.
The key to successful implementation is understanding these limitations and designing workflows that play to agents' strengths while keeping humans in the loop for tasks that require judgment, visual interpretation, or complex decision-making.
Real-World Healthcare Implementations
With this understanding of how AI agents work—and where they excel versus struggle—we can examine three healthcare practices that have successfully deployed Birch agents. These implementations demonstrate both the transformative potential and the practical considerations of bringing autonomous AI into healthcare workflows.
Scenario 1: Automating Prior Authorization in Primary Care
An illustrative 15-physician primary care practice
The Challenge
Picture a busy 15-physician primary care practice drowning in prior authorization requests. With each physician seeing an average of 25 patients per day, a practice like this processes over 200 prior authorizations weekly. Each request required staff to manually gather patient medical history, review insurance requirements, complete multiple forms, and follow up with insurance companies—a process that took 45 minutes on average.
In practices like this, two full-time staff members are often dedicated entirely to prior authorizations — and the team is still constantly behind. Patients wait days or weeks for medication approvals, and the staff burns out.
The Implementation
In this scenario, the practice deploys Birch AI agents to handle the entire prior authorization workflow. The agents integrate directly with the practice's EHR system and insurance portals to:
- Automatically detect when a prescription requires prior authorization
- Extract relevant patient medical history and diagnostic codes from the EHR
- Determine specific insurance requirements and documentation needs
- Complete and submit authorization forms to insurance companies
- Track submission status and automatically follow up on pending requests
- Alert staff only when human intervention is required
A rollout like this is designed to take about three weeks, including integration with existing systems and staff training. The agents handle straightforward authorizations automatically while escalating complex cases to human staff.
The Projected Results
Within the first month, the modeled impact looks like this:
The transformation this design targets: staff spend their time on patient care instead of paperwork, patients get their medications faster, and approval rates hold steady — the AI agents are built to be as accurate as human staff, only much faster.
Perhaps most importantly, a practice in this scenario can redeploy its prior authorization staff to higher-value roles — care coordination, patient education, chronic disease management.
Scenario 2: 24/7 Prenatal Care Coordination
An illustrative 8-physician OB-GYN group
The Challenge
Consider an OB-GYN group serving over 600 pregnant patients annually with comprehensive prenatal care. A practice like this faces constant challenges with patient communication: appointment reminders, lab result notifications, answering routine questions about pregnancy symptoms, and coordinating specialist referrals.
A group this size fields 50–60 patient calls per day. Most are routine questions — "Is this symptom normal?" "When should I schedule my next ultrasound?" "What do my lab results mean?" — but staff still have to be available to answer them, so phone lines stay busy and patients often wait hours for callbacks.
The Implementation
In this scenario, the group implements Birch AI agents to handle patient communication and care coordination. The agents work 24/7 through the practice's patient portal and SMS system to:
- Answer routine prenatal questions with evidence-based information
- Send personalized appointment reminders and handle rescheduling requests
- Notify patients of lab results with context and next steps
- Triage urgent concerns and alert on-call providers when necessary
- Coordinate specialist referrals and ensure follow-up appointments are scheduled
- Send week-by-week pregnancy education and wellness tips
The Projected Results
The design goal: patients get instant answers to their questions — especially in the evening, when anxiety peaks — and clinicians focus their time on complex medical decisions rather than routine communication.
The projected reduction in appointment no-shows matters most. With automated reminders and easy rescheduling, patients are more likely to attend appointments or give adequate notice when they can't — improving scheduling efficiency and continuity of care.
Scenario 3: Mental Health Intake Automation
An illustrative 8-therapist behavioral health group
The Challenge
Now take a group practice specializing in anxiety and depression treatment that struggles with its intake process. Initial patient screenings take 45 minutes per patient, requiring administrative staff to collect detailed mental health history, verify insurance, and assess the urgency of care needs.
The result in practices like this: a waitlist of 100+ people — not from lack of therapist capacity, but because intake is so time-consuming that only 4–5 new patients can be processed per week, and people in crisis wait weeks for their first appointment.
The Implementation
In this scenario, the group deploys Birch AI agents to handle initial patient intake. The agents conduct empathetic, structured conversations with new patients to:
- Collect comprehensive mental health history through conversational screening
- Verify insurance coverage and provide cost estimates
- Assess symptom severity and treatment urgency using validated clinical scales
- Detect crisis situations and immediately alert on-call clinicians
- Match patients with appropriate therapists based on specialties and availability
- Schedule first appointments and send intake paperwork
The agents are designed with clinician input to ensure empathetic, trauma-informed communication. They maintain strict HIPAA compliance and escalate any concerning responses to human clinicians.
The Projected Results
The access-to-care impact this design targets: from a 3-week wait for an intake appointment to same-day or next-day intake, people in crisis identified immediately and connected with urgent care, and therapists receiving comprehensive intake information before the first session — so that time goes to actual therapy rather than data collection.
Published research on conversational intake suggests many patients actually prefer an initial screening with an AI agent, finding it less intimidating than talking to a human stranger about sensitive mental health issues.
Key Insights and Common Patterns
Across these scenarios — and the published research on human-agent collaboration — consistent patterns emerge in how AI agents successfully transform healthcare workflows:
1. Focus on High-Volume, Rules-Based Tasks
AI agents excel at tasks that are repetitive, time-consuming, and follow clear protocols. Prior authorizations, appointment scheduling, and intake screening all fit this profile. These tasks don't require clinical judgment but do require meticulous attention to detail and significant time investment.
2. Augmentation, Not Replacement
The most successful implementations use AI to augment human workers, not replace them. Agents handle routine work while escalating complex cases to humans. This allows healthcare professionals to focus on tasks that require empathy, clinical judgment, and creative problem-solving.
3. 24/7 Availability Creates Unexpected Value
The ability for AI agents to work around the clock provides value beyond simple efficiency gains. Patients get immediate responses to questions, appointments can be scheduled outside business hours, and urgent issues can be triaged in real-time. This level of accessibility was previously impossible without significant staffing costs.
4. Integration Is Critical
The success of these implementations depended on seamless integration with existing systems—EHRs, patient portals, insurance systems, and communication platforms. Agents that work within established workflows are far more successful than those requiring new processes or systems.
5. Staff Redeployment, Not Reduction
Practices that successfully implemented AI agents didn't reduce staff—they redeployed them to higher-value work. Administrative staff moved into care coordination, patient education, and other roles that directly improve patient outcomes. This approach maintains employment while improving both job satisfaction and patient care.
The Science of Implementation: What Makes These Deployments Successful
The three scenarios above describe typical practices, not exceptional ones. But success isn't automatic. Research on human-agent collaboration reveals several critical factors that determine whether an AI agent deployment transforms operations or creates new problems.
Workflow-Level Integration, Not Task-Level Replacement
One of the most common mistakes in AI deployment is thinking in terms of individual tasks rather than complete workflows. A prior authorization isn't just "fill out a form"—it's a workflow that includes gathering patient history, understanding insurance requirements, crafting medical justification, submitting documentation, tracking status, and following up. Agents succeed when they can execute entire workflows autonomously, not when they're deployed as glorified form-fillers.
The prior-authorization scenario works because the Birch agent owns the entire workflow from detection to approval. Staff members don't need to babysit the process or manually hand off between steps. The agent monitors incoming prescriptions, detects authorization needs, executes the full workflow, and only surfaces to humans when it encounters something it can't handle.
Strategic Task Delegation Based on Programmability
Research on agent capabilities suggests a useful framework for deciding which tasks to delegate: programmability. Tasks fall into three categories:
Readily programmable tasks follow clear rules and work with structured data. Prior authorizations, appointment scheduling, insurance verification, and routine patient communication all fit this category. These tasks are ideal for full automation—agents can complete them faster and more accurately than humans, with minimal supervision.
Partially programmable tasks have some structure but require judgment or visual interpretation. Patient triage falls into this category—protocols exist, but deciding whether a symptom description warrants urgent attention requires contextual understanding. These tasks benefit from augmentation: agents can gather information and apply initial protocols, but humans make final decisions.
Minimally programmable tasks depend heavily on interpersonal skills, complex judgment, or visual assessment. Delivering difficult news to patients, adjusting treatment plans based on nuanced patient feedback, or interpreting unclear medical images all fall here. These tasks remain squarely in human domain—at least for now.
The most successful practices match agent capabilities to task characteristics. They automate the programmable, augment the partially programmable, and keep humans centered in everything else.
The 24/7 Advantage: Redefining Accessibility
One unexpected benefit of AI agents is their ability to work around the clock without fatigue, overtime costs, or burnout. This isn't just about efficiency—it fundamentally changes what's possible for patient care.
In these scenarios, prenatal patients get answers to routine questions at 2 AM when anxiety peaks, intake screenings happen on weekends when people finally have time to seek help, and prior authorizations get submitted immediately rather than waiting for the next business day. This level of responsiveness was previously impossible without significant staffing costs.
Research shows that immediate response times don't just improve patient satisfaction—they improve health outcomes. Patients who get quick answers to questions are more likely to follow treatment plans. Patients who can schedule appointments at their convenience are less likely to miss them. The 24/7 availability of agents creates a fundamentally more accessible healthcare system.
Staff Redeployment: The Human Element
A common fear about AI automation is job loss. But the model Birch is built around doesn't reduce headcount—it redeploys it. When a practice automates prior authorizations, it doesn't need to lay off its authorization specialists: one can move into care coordination, directly improving patient outcomes by helping people navigate complex treatment plans; another can focus on patient education for chronic disease management, driving better medication adherence and fewer complications.
This is the pattern the research supports. Staff freed from repetitive administrative work move into roles that require human judgment, empathy, and complex problem-solving—exactly the skills that healthcare desperately needs more of. The result is the same number of jobs, but higher job satisfaction, better patient care, and more sustainable workflows.
Research on human-agent teaming suggests this is the optimal model: agents handle high-volume programmable work with 96.4% fewer actions and 88.3% faster completion, while humans focus on tasks requiring clinical judgment, emotional intelligence, and creative problem-solving. This division of labor optimizes for both quality and efficiency.
Looking Forward: The Next Wave of Healthcare AI
These three scenarios represent just the beginning of how AI agents will transform healthcare delivery. Current capabilities focus largely on administrative workflows—tasks that, while critical, don't directly involve clinical decision-making. As agent capabilities improve, we expect to see expansion into more clinically adjacent areas.
Medication management: Agents could monitor patient medication adherence through smart packaging or app integration, proactively reaching out when doses are missed, automatically arranging refills, and flagging potential drug interactions when new prescriptions are added.
Chronic disease monitoring: For conditions like diabetes or hypertension that require continuous monitoring and frequent adjustments, agents could track patient-reported symptoms and biomarker data, flag concerning trends, and even suggest protocol-based treatment adjustments for physician approval.
Care coordination: Patients with complex conditions often see multiple specialists, creating coordination challenges. Agents could ensure that all providers have current information, schedule follow-ups based on treatment plans, and make sure test results reach the right people at the right time.
Preliminary diagnostic support: While diagnosis will remain a physician responsibility, agents could gather comprehensive symptom histories, order appropriate screening tests based on protocols, and prepare detailed pre-visit summaries that help physicians make faster, more informed decisions.
But regardless of how sophisticated these capabilities become, the fundamental value proposition will remain the same: freeing healthcare workers from administrative burden and routine workflow execution so they can focus on what they do best—providing compassionate, skilled care to patients who need it.
The Implementation Imperative
The practices in these scenarios aren't exceptional—they're deliberately typical. A suburban primary care practice, an eight-physician OB-GYN group without a large IT department, a therapy group with a waitlist problem and a limited budget. With the right tools and implementation strategy, practices like these are exactly who this technology is designed for.
What success requires isn't resources or technical sophistication. It's a willingness to rethink workflows, a clear-eyed assessment of where automation helps versus hurts, and a commitment to using technology to support rather than replace human healthcare workers.
The question isn't whether AI agents will transform healthcare workflows—they already are. The question is which practices will adapt quickly to take advantage of these capabilities, and which will continue struggling with unsustainable administrative burdens while their staff burns out and their patients wait.
Based on the research and the workflow economics above: the practices that move first will have a significant competitive advantage. They'll be able to serve more patients with the same staff, provide faster and more accessible care, and create work environments where healthcare professionals can focus on the aspects of their jobs that drew them to healthcare in the first place—helping people.
References
1 Wang, Z.Z., Shao, Y., Shaikh, O., Fried, D., Neubig, G., & Yang, D. (2025). How Do AI Agents Do Human Work? Comparing AI and Human Workflows Across Diverse Occupations. Carnegie Mellon University and Stanford University. arXiv:2510.22780v2 [cs.AI]
