Rahul Sharma of HSBlox on Why Agentic AI Will Redefine Healthcare Software Architecture

Agents are autonomous or semi-autonomous codebases that can perform three critical sub-tasks: They can access different data sources and synthesize data in real time, they can automate decision-making processes based on analysis of the data, and they can automate routine tasks by leveraging process automation tools, integrations to and from different systems, and orchestrate workflows.

Single as well as multiple agents can collaborate on interconnected tasks to arrive at an outcome. Single-agent systems can be used to automate standalone processes like claims validation, patient scheduling, and appointment reminders.

Meanwhile, multi-agent systems are required to handle more complex episodic events and workflows across multiple teams and systems. One example would be a care transition for knee surgery that involves hospitals, payers, different physician teams, and community health team members. A multi-agent system can help do better coordination between payers, providers, CBOs, patients, and all entities involved in the episodic care of the patient.

Agentic AI-based systems can be reliable partners for the healthcare workforce. They give our physicians, nurses and caregivers the enhanced capabilities of diagnosis, knowledge, and automating of tasks while maintaining the knowledgeable assistant-like feature to preserve the human aspect of healthcare – thus ensuring the best care for patients.

Agentic AI brings automation, personalization, and adaptive learning to healthcare. This transforms traditional SaaS tools into proactive care solutions. Instead of just presenting insights, Agentic AI acts on them, improving efficiency and patient outcomes.

There are several key technologies that make Agentic AI transformative for healthcare. Large language models (LLMs) can process unstructured data, such as medical notes, and automate communication. Computer vision allows Agentic AI to analyze medical imaging (another form of unstructured data), such as X-rays and MRIs. In both instances, clinicians and researchers have access to data that previously was inaccessible or difficult and costly to extract. Quality data at scale is needed to drive healthcare innovation, and that’s what Agentic AI can provide organizations.

Another powerful capability of Agentic AI, or Agent as a Service (AaaS), is reinforcement learning; that is, the ability to learn from outcomes. This helps clinicians to optimize care pathways based on real-world data. And Robotic Process Automation (RPA) frees up organization staff from repetitive tasks like data entry and appointment booking.

While there are multiple use cases for Agentic AI in healthcare, I’ll just cite a few here. A major use case is chronic conditions management. Treating chronic conditions accounts for the vast majority of healthcare spending in the U.S. AaaS can automate identification of the patient, prepare and communicate personalized dietary advice, order blood tests (if needed) and alert the care team if a patient’s condition does not improve.

Agentic AI also excels at enabling transitions between care teams. Most task hand-offs today are manual in nature across care teams, and workflow-based tools do not integrate across care settings. AaaS-based platforms can facilitate real-time coordination, making seamless transitions for inpatient, outpatient and post-acute settings.

Finally, AaaS platforms can accelerate claims processing by automating validation of claims, identifying any missing pieces of information, triggering any workflows that require resolution, and reducing denials. AI agents can utilize LLMs for clinical document interpretation and extraction/matching for coding accuracy.

In response to the first part of the question, it can, but for specific use cases. With advancements in AI, a lot of the business logic layer for appropriate use cases is going to be handled completely by AI agents. Once that happens, there is really no need to have a traditional SaaS-based model. AI agents will be able to understand what users want or need, anticipate the requests, and remove the need for the current model of SaaS applications.

A traditional SaaS implementation gives healthcare teams needed data and insights (analytic outputs), while an AaaS-based implementation can analyze, decide, and then act on the data, essentially automating much of the process. So, yes, these advanced capabilities should make AaaS a preferable option for healthcare organizations.

What this means for healthcare is both better clinical outcomes and reduced costs. Agentic AI automates repetitive tasks, personalizes the patient experience, and proactively prevents serious issues from arising.

In addition, platforms, solutions, tools, and utilities that utilize Agentic AI architectures can enhance productivity, reduce errors, and mitigate physician burnout. It’s important to note that these technologies are not replacements for healthcare workers. Rather, they are meant to be assistants that help provide better healthcare.

The success of AaaS-based platforms and solutions will depend upon the measurable results aligned with the needs of the business, while the biggest challenges are going to be in co-existing with current applications during the transition for specific use cases.

AaaS excels at data integration and management. Healthcare data often is scattered across multiple systems – electronic health records (EHRs), pharmacy data, lab results, and even patient-reported outcomes.

SaaS typically integrates with EHR systems using APIs but relies heavily on pre-defined data pipelines. Consequently, staff may need to manually curate and validate data – a cumbersome and time-consuming process for clinicians who would prefer to be engaging with patients.
Using autonomous data retrieval methods, Agentic AI can pull data from multiple sources in real-time, identify missing or incomplete records (and request corrections automatically), and recognize patterns without needing explicit rules for each data source. This helps clinicians quickly interpret data at the point of care.