Agentic AI as the Brain of Personalized Drug Delivery Devices

Healthcare systems are under sustained workforce strain, with physician burnout still approaching 42%¹ and persistent staffing shortages placing increasing pressure on clinical delivery capacity.

In fact, Reuters reported² that the U.S. is expected to have upwards of an 86,000-plus physician shortage by 2036. Staff shortages and physician burnout rates contribute to medical errors, lower patient satisfaction, and reduced quality of care. Against this backdrop, the latest artificial intelligence (AI) technology may offer help to ease healthcare professionals’ workloads and stress levels.

In the past, drug delivery devices including insulin systems, infusion pumps, and wearable injectors were primarily focused on transmitting data, delivering alerts, and enabling clinical oversight. However, clinicians still had to interpret the information and decide on the next course of treatment. 

In 2026, that model is beginning to shift with advanced agentic AI solutions, which minimize human oversight in routine decisions. For original equipment manufacturers (OEMs), this shift has significant implications for design, validation, and governance over time. 

Algorithms Act, Not Just Advise

Historically, artificial intelligence in medical devices has played a largely passive role. Algorithms highlighted unstructured data, visualized trends, and supported clinical decision-making, but clinicians remained responsible for taking needed action. 

Agentic AI, however, unveils a different paradigm, as these systems are goal-driven agents that consistently evaluate sensor data while operating within predefined safety limits to maintain a desired clinical state. 

In drug delivery applications, this can mean automatically adjusting doses based on inputs like glucose levels and patient activities, without requiring continuous manual human approval. This autonomy is restricted. It is bounded by hardware capabilities, sensor reliability, safety mechanisms, and regulatory requirements defined for product development. 

The goal is not to replace clinical judgment, but to enable systems to manage regular therapy adjustments in real time and under clinical supervision. 

Agentic AI Smart Drug Delivery: The Final Frontier

While the Smart Drug Delivery Systems, which are primarily powered by AI, projects a rapidly expanding $88.66 billion market in 2026,³ the “final frontier” of medical technology is the autonomous brain now being injected into these devices. This shift is driven by an estimated $1.83 billion agentic AI market⁴ in 2026, which is expected to increase to $19.71 billion in 2034. 

As a result of this upward trend, the U.S. Food and Drug Administration (FDA) is adapting to this emerging technology shift. 

Although the FDA does not classify medical devices as “agentic AI,” in January 2026, it cleared an automated insulin delivery system algorithm that analyzes glucose levels in real-time and adjusts insulin doses, relieving the “mental burden associated with the daily management of Type 1 diabetes.”

Clinical and Operational Advantages

There are clear clinical and operational advantages of agentic systems, including:

• Less Administrative Rigidity: Agents address standard checks and inconsistencies of data without raising alerts and allow staff to focus on higher-value care.
• Faster Therapy Adjustments: The consistent assessment of various physiological factors facilitates the timely modification of doses or infusion rates within approved safety limits.
• Lower Alert Fatigue: The non-critical events are filtered out, and clinics are only alerted when intervention is required.
• Improved Therapy Consistency: Real-time input and constraint balancing facilitates stable treatment, particularly in home and chronic care.

Human-in-the-Loop vs. Human-on-the-Loop

Traditional medical software safety protocols rely on Human-in-the-Loop (HITL). AI systems suggest actions and require approval at each clinician stage. This is effective for many real applications, but HITL can slow down the value of real-time therapy. 

As a result, a new approach, Human-on-the-Loop (HOTL), is emerging. 

In this model, the systems execute predefined actions autonomously while clinicians monitor outcomes and step in only when required. For drug delivery devices, HOTL allows faster adjustments while keeping clinician control.

To make this model regulatory compliant, explainability and traceability must be built in from the outset so that when AI modifies a dose, it provides physiological signals and reasoning. Clarity in logs and data builds trust and also allows clinicians to supervise autonomy rather than micromanage system behavior.

Teams Enable Safe Autonomy in Drug Delivery

Development of smart or intelligent drug delivery systems places pressure on traditional organizational models. Agentic AI requires a fundamentally different development model, as this shift spans multiple challenges, including hardware constraints, software behavior, cybersecurity needs, and regulatory compliance. Traditional structures isolate development teams, who pass on all the work to one group after the other, often slowing down development and increasing risk.

Many OEMs prefer to create cross-functional teams that bring all the desired disciplines together early in the development process. Engineers work alongside quality and regulatory experts, who define the safety controls, post-marketing monitoring plans, and more. IT and security teams ensure the devices are safely connected to the intelligent systems. Additionally, workforce development also plays an important role requiring companies to search for talent capable of thinking across physical devices, and system-level software domains.

Build, Buy, or Hybrid: Strategic Choices for Drug Delivery OEMs

OEMs face three strategic approaches to integrating agentic AI capabilities into drug delivery systems: Build, Buy, or Hybrid:

• Build: Ensure you have skilled agentic AI developers in-house or through a solution provider to build a system like a smart drug delivery system. This method usually has higher initial costs and sometimes longer rollout times, but with the right tech provider, deployment can be fast plus you will own the intellectual property and minimize ongoing costs. 
• Buy: Healthcare organizations like clinics or hospitals may prefer to buy a third-party product that has already been market evaluated. This approach usually incurs lower initial costs and faster deployment, but it may include ongoing costly subscription fees and lack the necessary customization to meet specific operational needs.
• Hybrid: This is a common approach that combines proven AI or agentic AI products with device-centric customization provided by vetted software solution providers.

Each approach comes with distinct implications that can affect intellectual property ownership, regulatory accountability, and system management over the developmental lifecycle.

The Future is Agentic AI

Based on the growth in agentic AI, the healthcare industry will see a decided shift in drug delivery devices from passive tools to active participants in care delivery. In coming years, integrating these systems with digital models and continuous monitoring platforms can deliver personalized therapy.

For OEMs, the challenge is not only in integrating intelligence within the system but in engineering responsible autonomy. To achieve this, regulations, hardware and software rules, and team readiness must advance together. Thus, the devices are capable of delivering effective therapy, but the true value lies in autonomy that is safe, explainable, and closely supervised. 

The promise of agentic AI medical devices, specifically smart drug delivery systems, also eases the burden on the strained medical labor force while at the same time providing better patient outcomes because care is given in real time.

References
¹ bit.ly/mposoftware06261
² reut.rs/43qIKdf
³ bit.ly/mposoftware06262
⁴ bit.ly/mposoftware06263

MORE FROM THIS AUTHOR: Agentic AI as Embedded Decision Makers in Diagnostic Devices

Anshu Raj is a director of operations at Chetu, where he oversees the AI and engineering R&D portfolios, including high-tech medical devices. Raj, who holds certifications in PMP, Agile, and NetSuite Foundation, focuses on mid-market companies that want to accelerate innovation through on-demand engineering teams.