Automating clinic workflow, community screening and referral
Agentic AI in Public Health: What It Is, Why It Matters, and How to Get Started
Posted by Dr. Bashir Ssuna | March 2026
Artificial intelligence has been part of public health for years, mostly as a passive tool. You feed it data, it gives you a prediction or a classification, and you act on the result. That model has been useful, but it has limits. The next wave is different. Agentic AI does not wait for instructions at every step. It reasons, plans, takes actions, retrieves information from multiple sources, and adjusts its approach based on what it finds, all with a defined goal in mind and minimal human intervention between steps. For those of us working in epidemiology, biostatistics, and implementation research in low- and middle-income settings, this shift is worth paying close attention to.
What Makes AI “Agentic”?
The U.S. Government Accountability Office defines AI agents as systems that can “make and adjust plans when the actions required to accomplish a goal are not clearly defined by a user.” Unlike a conventional model that classifies a chest X-ray or predicts a fasting glucose level, an agentic system can be given a broader task such as “conduct a literature review on TB-diabetes co-management in Sub-Saharan Africa,” and then go ahead and search databases, synthesize findings, flag inconsistencies, and produce a structured report on its own. Using machine learning algorithms, agentic AI adapts to real-time environments. Unlike conventional AI, which depends on fixed rules, agentic AI acts on its own to achieve goals and continuously updates its behavior as new information comes in.
The architecture behind this typically involves four core patterns: the agent reflects on its own outputs and iterates, it uses external tools like APIs and databases, it plans multi-step processes, and in more advanced systems, multiple specialized agents collaborate to divide complex tasks.
Where Agentic AI Is Already Being Applied in Public Health
The CDC is one of the most prominent institutions actively exploring agentic AI for public health work. CDC’s internal exploratory evaluation during 2025 found that deep research tools (a class of agentic AI capabilities) are particularly well-suited for literature synthesis for emerging health threats, policy and legal scans across jurisdictions, strategic planning and scenario analysis, public health communications, and comparative analysis of programs, interventions, or regulations.
In clinical and research settings, a recent scoping review published in npj Digital Medicine examined agentic AI systems across healthcare domains. The review identified studies spanning emergency medicine, oncology, radiology, and rehabilitation, with systems demonstrating features such as autonomous operation, goal-directed behavior, action initiation, and multi-agent collaboration. Reported outcomes included high accuracy in cancer diagnosis, treatment planning, alert generation, coaching, and workflow optimization. You can read the full review here: npj Digital Medicine, 2026.
In drug discovery, Genentech worked with AWS to build an agentic research tool called gRED Research Agent that automates manual literature and database searches. What makes this solution notable is its use of autonomous agents that can break down complicated research tasks into dynamic, multi-step workflows, adapting their approach based on information gathered at each step, accessing multiple knowledge bases, and executing complex queries through internal APIs.
For those interested in the broader healthcare perspective, a recent open-access article in Frontiers in Digital Health titled “AI with Agency: A Vision for Adaptive, Efficient, and Ethical Healthcare” is a solid entry point. It is available at PMC. A preprint systematic review titled “The Role of Agentic Artificial Intelligence in Healthcare” is also available on ResearchGate and is worth bookmarking as the evidence base matures.
What This Means for Research in African Settings
Much of the published work on agentic AI comes from high-income healthcare systems with robust infrastructure. But the opportunities for research-constrained settings are significant and arguably more urgent. Consider some of the bottlenecks that agentic AI could address:
Systematic review preparation, which typically takes months, could be accelerated by agents that autonomously search PubMed, Embase, and Scopus, screen abstracts, extract data, and produce draft evidence tables. Surveillance data analysis, where an agent could monitor multiple disease registries and flag unusual patterns without requiring a statistician to manually query each source. Grant narrative development, where agents could review existing literature and draft background sections aligned with specific funder priorities. Clinical risk stratification, where an agent does not just run a model but retrieves updated patient data, applies the model, checks for data quality issues, flags high-risk cases, and logs the decision, all within a single workflow.
These are not speculative use cases. The building blocks exist now. What is needed in settings like Uganda is the capacity to adapt, validate, and deploy them responsibly.
The Ethical Terrain
Agentic AI introduces risks that are qualitatively different from conventional models. When a system takes autonomous actions, errors can compound across steps before a human catches them. In public health, where decisions affect populations, this is consequential. The CDC’s guidance on agentic research tools identifies human oversight, scientific integrity, and mission alignment as non-negotiable principles. It recommends that a qualified reviewer verify sources, facts, and conclusions before any output is acted upon or released publicly.
Data sovereignty is another concern that deserves particular attention in African research contexts. When agents interact with cloud services and external APIs, it is important to understand where data flows, who has access to it, and whether the terms of those platforms are compatible with research ethics obligations to participants and communities.
How to Start Learning Agentic AI
If you are a public health researcher or biostatistician who wants to build practical skills, here are some of the most accessible paths:
For a free, no-code conceptual introduction, the Vanderbilt University course on Coursera titled “Agentic AI and AI Agents: A Primer for Leaders” covers how agents work and how to think about deploying them without requiring prior programming experience. It is free to audit.
For a more hands-on, Python-based foundation, DeepLearning.AI’s Agentic AI course walks you through the four core design patterns (reflection, tool use, planning, and multi-agent coordination) with practical implementation using Python.
For a structured certificate path, Johns Hopkins University’s Certificate Program in Agentic AI covers agentic architectures, reasoning models, the Model Context Protocol, and multi-agent systems, with mentored sessions and a portfolio project.
IBM also offers a free Agentic AI Hands-On learning path through Cognitive Class that takes you through frameworks including CrewAI, LangGraph, AutoGen, and PydanticAI, with healthcare use cases included in the curriculum.
For those who prefer books, AI Engineering by Chip Huyen and The LLM Engineering Handbook by Paul Iusztin and Maxime Labonne are both highly regarded starting points that provide conceptual depth before diving into frameworks.
A Note on Integrating This Into Your Practice
At MESC, we are actively exploring how agentic workflows can support the kind of work we do: building predictive models for maternal and fetal outcomes, conducting systematic reviews on TB and HIV comorbidities, and developing risk stratification tools for populations in Uganda. The IFG and hypertension risk stratification tools featured on this website are early examples of what happens when epidemiological models are made accessible through interactive interfaces. Agentic AI represents the next step: systems that do not just present a result but take that result and act on it within a defined, auditable, and human-supervised workflow.
We are at an early point in this transition. The evidence base is still thin, most studies are exploratory, and robust clinical validation in African settings is nearly absent. But the direction is clear, and the window to shape how these tools are designed and governed in our contexts is open now, not later.
The MESC team works at the intersection of epidemiology, biostatistics, and implementation science in Uganda. Follow our YouTube channel at youtube.com/@makepistat for tutorials on R, Python, data analysis, and applied machine learning in public health.