5 Lessons Healthcare Learned the Hard Way – and Why Agentic AI Must Be Different

The following is a guest article by Aditya Bansod, Co-Founder and Chief Technology Officer at Luma Health

Over the last decade, the first digital wave in healthcare prioritized access and compliance over outcomes, creating patient portals and digital front doors that look modern and consumer-friendly.  

But underneath the hood, little actually changed. Patients can find information — a provider to suit their needs — but can’t act on it. They still have to pick up the phone and wait on hold to schedule an appointment. 

On the back end, staff face their own barriers. They can see patients’ EHRs, but referral workflows are still manual. Referrals sit in fax queues for weeks while patients wait or give up. 

Where is the efficiency that digitization promised to deliver? Digital front doors have done little to reduce the administrative burden or cost in healthcare and ultimately have increased operational friction. 

Now, as agentic AI explodes onto the scene, healthcare IT has a second chance: an opportunity to deliver on the promise of greater access, speed and efficiency, and to make things easy for patients (finally).  

But only if we learn from what went wrong the first time. 

Here are five lessons from the first digital wave in healthcare IT that can inform how health systems approach AI implementation. 

1. The Problem Isn’t Complexity; It’s Overlooking the Audience

Airlines let you book a cross-country, multi-city flight in seconds. Utility companies make it simple to track your usage and pay your bill. Both have immensely complex systems running behind the scenes, but users never see them and still get what they need. 

Healthcare is no more complex than those industries. They may look similar on the surface, but the functionality isn’t there. Analog workflows still drive basic tasks, and patients are left out of the loop.  

Case in point: when your PCP refers you to a specialist, the staff often makes the appointment without consulting you. Can’t make it? You have to call to reschedule, adding friction.  

Does that sound like patient-centered care? 

In the next wave, agentic AI must give patients autonomy, put the outcome at the center of the experience, and empower patients to take an active role in the process.  

2. Outcomes Over Interfaces

The first digital wave was all about connecting patients to information. Portals provided appointment data, visit notes, and test results. Then, generative AI offered summarization, ingesting, and analyzing data to answer basic questions.  

But do you just want information? Or do you, as a patient, want agency? 

Agentic AI shifts the deliverable from providing information to facilitating outcomes. Instead of showing patients a path, agentic AI walks them through it, enabling self-service like scheduling and rebooking appointments rather than canceling without human intervention. Anything less is just another digital dead end. 

AI can’t solve everything, but when systems are designed around clear outcomes, they can orchestrate workflows that move patients from intent to action, and improve throughput, efficiency, and revenue capture.  

3. Point Solutions are Unsustainable

The first digital wave was dominated by “best of breed” point solutions, and this is where most digital front doors and chatbots have fallen short. Most were never API-driven, and layering on AI or automated workflows has been extremely difficult because there’s no “plumbing” to connect them. Functions remain locked in siloes:  

• scheduling lives inside the EHR

• payer rules, claims, and prior authorization are stuck in payer-owned portals 

• referral workflows remain analog, through faxes and phone calls
  

Contrast that with platforms like Salesforce, where entire ecosystems are built on reliable, high-quality connectivity to a system of record. With advanced APIs that allow software to do anything an end user can, dozens of integrations can orchestrate real workflows at scale. 

Agentic AI demands the same platform approach. It can only work if it operates inside systems, not alongside them. Health organizations already have the systems of record in EHRs and EMRs. Now it’s time to bring on the ecosystem of solutions — and emerging standards like Model Context Protocols (MCPs) —  that can operate as a cohesive platform. Without it, AI is just a novelty — and another source of frustration for patients and providers. 

4. Tools Must Reduce Burden, Not Add to It

AI won’t eliminate humans from healthcare, but it can drastically change what they spend time doing. 

One of the biggest failures of early AI tools is that they add cognitive burden. When a patient walks into the Emergency Department, the provider might open a chart flooded with alerts, recommendations, and fragmented data. They must wade through 100 questions before treatment, wasting time and energy. 

The opportunity for agentic AI isn’t to surface more insights — it’s to reduce noise. Instead of handing clinicians a thousand data points, systems should align around relevance and goals. For the patient in the ED, for example, the system shows:  

• They’re at risk for Type II diabetes. What’s the status? Does it matter for this visit? 

• Their intake data is incomplete. What’s missing? What’s relevant?
  

Technology should lighten the load, not feel like one more thing to manage. By automating data gathering and orchestration, AI can reduce cognitive load, freeing clinicians to focus on the patient in front of them instead of the screen.  

5. Design for Adaptability, Not Perfection

The first digital wave taught us that technology is rarely static. Tools change, models improve, and capabilities evolve.

AI strategies that assume permanence will fail. Things are moving too fast, and froth and churn are inevitable. In this environment, there can be no sacred cows. Health systems need to get comfortable experimenting, iterating, and replacing tools every few years.  

That’s why moving away from point solutions toward platform thinking is essential: it allows flexibility in choosing tools that can expand, adapt, and integrate over time. It’s the only way to grow an ecosystem that will outlast the hype cycle.  

The Next Wave: Operational AI That Turns Insight Into Action 

Healthcare doesn’t need more interfaces. We need AI that actually does the work and meets patients where they are with tools that can evolve to serve future needs. Systems that build for adaptability will be ready for what comes next. 

About Aditya Bansod

Aditya Bansod is Co-Founder and CTO at Luma Health. With a lifelong passion for building software, Bansod leads Luma Health’s technical vision and strategic direction for building a platform that empowers healthcare providers to better serve their patients and improve healthcare outcomes. With over 15 years of experience as a product management leader developing mobile solutions at Adobe and Microsoft, and at venture-backed start-ups, Bansod made the transition from B2B software solutions to healthcare in 2015 in order to have a meaningful and measurable impact on how providers use mobile technologies to engage with and communicate with their patients.

The following is a guest article by Aditya Bansod, Co-Founder and Chief Technology Officer at Luma Health

Over the last decade, the first digital wave in healthcare prioritized access and compliance over outcomes, creating patient portals and digital front doors that look modern and consumer-friendly.  

But underneath the hood, little actually changed. Patients can find information — a provider to suit their needs — but can’t act on it. They still have to pick up the phone and wait on hold to schedule an appointment. 

On the back end, staff face their own barriers. They can see patients’ EHRs, but referral workflows are still manual. Referrals sit in fax queues for weeks while patients wait or give up. 

Where is the efficiency that digitization promised to deliver? Digital front doors have done little to reduce the administrative burden or cost in healthcare and ultimately have increased operational friction. 

Now, as agentic AI explodes onto the scene, healthcare IT has a second chance: an opportunity to deliver on the promise of greater access, speed and efficiency, and to make things easy for patients (finally).  

But only if we learn from what went wrong the first time. 

Here are five lessons from the first digital wave in healthcare IT that can inform how health systems approach AI implementation. 

1. The Problem Isn’t Complexity; It’s Overlooking the Audience

Airlines let you book a cross-country, multi-city flight in seconds. Utility companies make it simple to track your usage and pay your bill. Both have immensely complex systems running behind the scenes, but users never see them and still get what they need. 

Healthcare is no more complex than those industries. They may look similar on the surface, but the functionality isn’t there. Analog workflows still drive basic tasks, and patients are left out of the loop.  

Case in point: when your PCP refers you to a specialist, the staff often makes the appointment without consulting you. Can’t make it? You have to call to reschedule, adding friction.  

Does that sound like patient-centered care? 

In the next wave, agentic AI must give patients autonomy, put the outcome at the center of the experience, and empower patients to take an active role in the process.  

2. Outcomes Over Interfaces

The first digital wave was all about connecting patients to information. Portals provided appointment data, visit notes, and test results. Then, generative AI offered summarization, ingesting, and analyzing data to answer basic questions.  

But do you just want information? Or do you, as a patient, want agency? 

Agentic AI shifts the deliverable from providing information to facilitating outcomes. Instead of showing patients a path, agentic AI walks them through it, enabling self-service like scheduling and rebooking appointments rather than canceling without human intervention. Anything less is just another digital dead end. 

AI can’t solve everything, but when systems are designed around clear outcomes, they can orchestrate workflows that move patients from intent to action, and improve throughput, efficiency, and revenue capture.  

3. Point Solutions are Unsustainable

The first digital wave was dominated by “best of breed” point solutions, and this is where most digital front doors and chatbots have fallen short. Most were never API-driven, and layering on AI or automated workflows has been extremely difficult because there’s no “plumbing” to connect them. Functions remain locked in siloes:  

• scheduling lives inside the EHR

• payer rules, claims, and prior authorization are stuck in payer-owned portals 

• referral workflows remain analog, through faxes and phone calls
  

Contrast that with platforms like Salesforce, where entire ecosystems are built on reliable, high-quality connectivity to a system of record. With advanced APIs that allow software to do anything an end user can, dozens of integrations can orchestrate real workflows at scale. 

Agentic AI demands the same platform approach. It can only work if it operates inside systems, not alongside them. Health organizations already have the systems of record in EHRs and EMRs. Now it’s time to bring on the ecosystem of solutions — and emerging standards like Model Context Protocols (MCPs) —  that can operate as a cohesive platform. Without it, AI is just a novelty — and another source of frustration for patients and providers. 

4. Tools Must Reduce Burden, Not Add to It

AI won’t eliminate humans from healthcare, but it can drastically change what they spend time doing. 

One of the biggest failures of early AI tools is that they add cognitive burden. When a patient walks into the Emergency Department, the provider might open a chart flooded with alerts, recommendations, and fragmented data. They must wade through 100 questions before treatment, wasting time and energy. 

The opportunity for agentic AI isn’t to surface more insights — it’s to reduce noise. Instead of handing clinicians a thousand data points, systems should align around relevance and goals. For the patient in the ED, for example, the system shows:  

• They’re at risk for Type II diabetes. What’s the status? Does it matter for this visit? 

• Their intake data is incomplete. What’s missing? What’s relevant?
  

Technology should lighten the load, not feel like one more thing to manage. By automating data gathering and orchestration, AI can reduce cognitive load, freeing clinicians to focus on the patient in front of them instead of the screen.  

5. Design for Adaptability, Not Perfection

The first digital wave taught us that technology is rarely static. Tools change, models improve, and capabilities evolve.

AI strategies that assume permanence will fail. Things are moving too fast, and froth and churn are inevitable. In this environment, there can be no sacred cows. Health systems need to get comfortable experimenting, iterating, and replacing tools every few years.  

That’s why moving away from point solutions toward platform thinking is essential: it allows flexibility in choosing tools that can expand, adapt, and integrate over time. It’s the only way to grow an ecosystem that will outlast the hype cycle.  

The Next Wave: Operational AI That Turns Insight Into Action 

Healthcare doesn’t need more interfaces. We need AI that actually does the work and meets patients where they are with tools that can evolve to serve future needs. Systems that build for adaptability will be ready for what comes next. 

About Aditya Bansod

Aditya Bansod is Co-Founder and CTO at Luma Health. With a lifelong passion for building software, Bansod leads Luma Health’s technical vision and strategic direction for building a platform that empowers healthcare providers to better serve their patients and improve healthcare outcomes. With over 15 years of experience as a product management leader developing mobile solutions at Adobe and Microsoft, and at venture-backed start-ups, Bansod made the transition from B2B software solutions to healthcare in 2015 in order to have a meaningful and measurable impact on how providers use mobile technologies to engage with and communicate with their patients.