SENA Insights
Posted on November 9, 2026
By Anthony J. Wehbe, DO, MBA, FACOI | Founder & CEO, SENA Health | Practicing Internist, Senacare | Former Health-System Physician Executive
I hear the same question from almost every direction in healthcare right now.
A physician group president asks me what AI can safely take off the front desk. An administrator tells me they have open positions they cannot fill. A board member wants to know why the organization is not moving faster. A practice manager has just seen another impressive demonstration showing an AI agent answering calls, scheduling appointments, and working around the clock.
The pressure naturally becomes: Which tool should we buy, and how quickly can we turn it on?
I think that is the wrong first question.
I still practice medicine through Senacare, caring for patients across the Philadelphia region in their homes, assisted-living communities, nursing facilities and group homes. And that experience has made me more optimistic about AI — but also much more demanding about how we deploy it.
I see what happens when a seemingly simple administrative task does not go as planned.
A nurse in a group home may need an order clarified. A family caring for an elderly parent may be trying to get a medication refilled. A facility may need a specialist appointment scheduled for a patient who cannot navigate the system independently. A patient may call about something that sounds administrative until, two questions later, it becomes clear that clinical judgment is required.
Those situations have taught me that the real challenge is rarely just answering the interaction.
It is knowing what can safely be automated, what requires a person, and when the system needs to change course.
That is why I believe patient access should start human-led.
AI should then earn its way into the workflow, task by task, based on evidence that it can reliably finish the patient’s need. And when it cannot, a person should be able to step in immediately.
That is the same standard physicians already use throughout medicine.
We do not adopt a new protocol because it looked impressive in a demonstration. We define what success looks like, test it, observe the outcomes, and expand our trust as the evidence accumulates.
AI at the front desk deserves the same discipline.
Physicians are not inherently resistant to AI.
In a survey of nearly 1,200 physicians conducted through November 2024, the American Medical Association found that 57% identified reducing administrative burden through automation as the greatest opportunity for AI in medicine. In the same survey, increased oversight was the regulatory action physicians most often ranked first as a way to increase trust in AI.
That matches what I hear directly from physicians.
Most physicians I speak with are not asking, “How do we stop AI?”
They are asking:
Where exactly is it acting?
What happens when it gets something wrong?
When does a human take over?
Can my patient still reach a person?
And perhaps most importantly:
Who is accountable when the workflow fails?
Those are reasonable questions.
In medicine, trust is rarely granted all at once. It is earned through repeated evidence.
I think AI should earn trust the same way.
One of the biggest misconceptions in healthcare technology is that the workflow lives in the EHR.
It often does not.
A great deal of operational knowledge lives in people.
The experienced scheduler knows that one physician will see a certain type of patient at 4:30 on a Friday while another will not.
The front-desk employee knows which payer requires a phone call before the authorization moves.
The nurse knows which patient will answer the daughter’s phone but not their own.
The practice administrator knows that one office follows a different scheduling rule because of something that happened five years ago.
And in my Senacare practice, the nurses and staff caring for people in group homes and nursing facilities know things no algorithm could infer from the chart alone.
They know which resident communicates best through a particular caregiver.
They know which patient becomes distressed when routines change.
They know when a family needs a phone call rather than another portal message.
They know the difference between a patient refusing something behaviorally and a patient whose condition is truly changing.
That operational knowledge is real.
The problem is that much of it is undocumented.
I saw the same thing earlier in my career at Kennedy and Jefferson. When an experienced employee left, organizations sometimes discovered that a critical part of the workflow had been living in that person’s head.
That is dangerous whether the replacement is another employee or an AI agent.
So when we start working with an organization, I do not want the first question to be, “What can we automate?”
I want the first question to be:
How does the work actually happen today?
Healthcare-trained coordinators can surface those workarounds, document them, identify which ones are necessary, and expose which ones exist only because the process is broken.
That is where the real redesign begins.
I think about AI adoption almost like graduated responsibility in clinical training.
You do not ask someone to perform every responsibility independently on the first day.
You start with well-defined work.
You establish supervision.
You watch performance.
You expand responsibility as competence is demonstrated.
Patient-access AI should follow the same pattern.
Start with frequent, rule-bound work where the expected result is clear.
Then define what success means before the AI goes live.
Did the task actually get completed?
Did the patient have to call back?
Did the work return to staff?
Did the AI escalate appropriately when the request fell outside the expected workflow?
Those measures matter more to me than a raw automation percentage.
I would rather automate 30% of a workflow extremely well than automate 80% and quietly create more work for nurses, medical assistants, schedulers and physicians.
This is where I think healthcare should borrow from how clinicians build trust.
When something is new, we supervise more closely.
Human teams should review AI-handled requests while the organization learns where the system performs well and where it fails.
As performance stabilizes, the level of review can decrease.
But oversight should never disappear entirely.
And every organization should name a person who has the authority to say:
“This task is not performing the way we expected. Move it back to people.”
That sounds obvious, but many organizations adopt technology without ever defining who has that authority.
If physicians begin reporting that something is going wrong, there should be a direct path to someone who can act.
That is how trust is preserved.
In my own practice, I see why this boundary has to be explicit.
A patient may call because they want an appointment.
That sounds administrative.
But then they mention worsening shortness of breath.
Or a caregiver calls asking whether a medication was renewed, but in the course of the conversation explains that the patient has stopped eating and has barely urinated all day.
The workflow has changed.
AI may help gather information, summarize it, and route it efficiently.
But clinical interpretation, triage, and treatment decisions belong with appropriately qualified clinicians.
The system has to recognize when the administrative workflow has crossed into clinical territory and hand it off with context.
That is not an AI failure.
That is successful orchestration.
Patients do not judge the architecture of the system.
They judge what happens at the moment something gets difficult.
If the AI cannot finish the request, what happens next?
Does the patient hear:
“Someone will call you back”?
Or does a trained person step into the interaction with the conversation already in front of them and continue solving the problem?
That difference matters.
I have watched patients become frustrated because they have to repeat the same story three times to three different people.
I have also watched physicians lose confidence in access systems because their patients eventually bypass the system and contact them directly when something fails.
That is one of the hidden costs of poor access design.
The unresolved work always goes somewhere.
Frequently, it ends up back with the physician.
A good handoff prevents that.
It brings a person in before frustration grows, preserves the context, and gives that person responsibility for seeing the request through.
And every handoff teaches the organization something about the AI’s limits.
That information should guide what the AI earns next.
This philosophy comes directly from my experience as both a physician and an operator.
At Senacare, I see healthcare at the patient level.
I see the elderly patient at home.
I see the nurse in the group home.
I see the nursing facility trying to coordinate care across physicians, pharmacies, hospitals and families.
At SENA, I see the same issues at organizational scale while working with physician leaders and administrators trying to redesign access for thousands of patients.
And earlier in my career at Kennedy and Jefferson, I saw how easily an organization can mistake a workflow problem for a staffing problem — or a technology problem.
Those experiences have convinced me that the answer is not people or AI.
It is people and AI, deliberately designed around the work.
That is why SENA’s Access Command Center can begin human-led.
Healthcare-trained coordinators first understand the workflow, handle the work, document the exceptions and establish the performance baseline.
Then AI agents take on specific tasks as they demonstrate that they can perform them reliably.
Routine work increasingly moves to technology.
People increasingly focus on exceptions, empathy, judgment and problem-solving.
Clinical issues move to clinicians.
And if the technology stops performing, the work can move back.
The goal is not the highest automation rate.
The goal is the best operating model.
When I discuss AI with physicians, the questions are remarkably consistent:
Where does AI act?
When does a person step in?
Can my patient always reach someone?
Who do I call when something is not working?
If leadership can answer those questions clearly, physician skepticism usually changes.
Because most physicians are not afraid of technology.
They are afraid of losing control of something that affects their patients.
The best way to build trust is not to ask them for more faith.
It is to show them evidence.
Start human-led.
Measure the baseline.
Give AI one well-defined task.
Watch what happens.
Let it earn the next one.
I believe AI will eventually perform a very large share of administrative work in healthcare.
That is not the question anymore.
The more important question is how we get there without losing the judgment, accountability and trust that make the healthcare system work.
For me, the answer is straightforward:
Start with people. Let AI earn the work. And never lose sight of who is accountable for the patient at the end of the workflow.
Want to talk through your organization’s front end? Schedule a conversation with Dr. Wehbe.