A University of Alabama at Birmingham workshop demonstrated that healthcare professionals can build functional AI agents in just three days, even without coding experience. The Marnix E. Heersink Institute for Biomedical Innovation hosted the intensive "AI Automation in Healthcare" program August 21-23, 2026, bringing together clinical, administrative, and research staff for a hands-on journey from AI fundamentals to deploying working agents in real healthcare workflows.

What Skills Did Participants Actually Gain?

The workshop structured learning progressively across three days, starting with foundational concepts and building toward practical deployment. On day one, participants learned how large language models (LLMs) work, including their limitations around hallucination, bias, and data sensitivity in clinical settings. They then studied the CRAFT framework, a prompt engineering approach that stands for Context, Role, Ask, Format, and Tweaks, which gives clinicians a repeatable method for getting better results from AI tools.

Day two focused on hands-on building. Participants mapped out a real healthcare workflow, then converted it into an actual AI agent. They started with no-code tools in Microsoft's Copilot Studio, then progressed to a more advanced coding agent built with the free, open-source OpenAI Codex CLI. By the end of the day, each participant had created an agent capable of handling multi-step tasks, tested it on sample data, and recorded a demonstration of it in action.

The final day covered advanced frameworks, including an introduction to the Model Context Protocol and on-device deployment using Ollama, an open-source tool for running models locally. Critically, the workshop also dedicated time to governance, guardrails, and compliance, ensuring participants understood how protected health information must move through UAB Health System's approval pathway before any tool goes live.

How to Build Your Own Healthcare AI Agent: Key Steps

• Start with Foundations: Learn how large language models work, their limitations, and the CRAFT framework for effective prompting before attempting to build anything.
• Map Your Workflow: Identify a real healthcare task you want to automate, such as scheduling, documentation, or prior authorization, and document each step.
• Use No-Code Tools First: Begin with no-code platforms like Microsoft Copilot Studio to prototype your agent before moving to more complex coding approaches.
• Progress to OpenAI Codex CLI: Once comfortable with no-code tools, use the free OpenAI Codex CLI to build agents capable of handling multi-step tasks and complex workflows.
• Test and Govern: Test your agent on sample or de-identified data, and ensure it complies with your organization's governance and data protection requirements before deployment.

Every exercise in the workshop relied on de-identified or synthetic data, and every tool used was free or already included in UAB's Microsoft license. This approach removed cost and data-risk barriers for participants experimenting with AI technology for the first time.

The workshop was led by Rubin Pillay, M.D., Ph.D., Sandeep Bodduluri, Ph.D., Heather Milam, and Abhi Pudhota of the Marnix E. Heersink Institute for Biomedical Innovation, with special guest faculty Anthony Chang, M.D., and Alfonso Limon, Ph.D., rounding out the teaching team.

What Are Clinicians Actually Using These AI Agents For?

Participants left the workshop with concrete, deployable tools. Mark Williams, M.D., vice president of the Learning Health System and Quality Improvement at UAB Medicine, highlighted the importance of collaboration between clinicians and AI programmers.

"The workshop demonstrated how the partnership between clinicians with expertise and AI programmers is essential to build the AI agents needed to improve patient care at UAB Health System," said Williams.

Mark Williams, M.D., Vice President of the Learning Health System and Quality Improvement at UAB Medicine

Jamie Wade, director of Outpatient Rehabilitation Service at UAB Medicine, found immediate practical applications. She identified two sessions that reshaped her thinking: the CRAFT model, which gave her a repeatable approach to prompting AI for better results, and the discussion of AI agents and their potential to streamline work and support decision-making.

"One of my biggest takeaways is how AI can serve as a practical tool to improve efficiency. Since the workshop, I have been using it to reduce administrative workload, analyze complex data more quickly, and summarize themes and insights across a variety of projects," said Wade.

Jamie Wade, Director of Outpatient Rehabilitation Service at UAB Medicine

Wade sees momentum continuing to build across her field. "By automating routine tasks and simplifying data collection, they can provide more timely insights, support performance improvement efforts, and help teams make more informed decisions," she noted.

Williams added that the technology's trajectory points toward broader healthcare transformation. "AI and automation will be essential to deliver high-quality, cost-effective care to patients and increase rapid access to patient education with confirmation of comprehension," he said.

Williams

What distinguishes this workshop from typical AI training is its emphasis on real-world application. Participants didn't just learn theory; they left with a working agent built around a genuine healthcare workflow, a governance and rollout plan for actual deployment, and ongoing connections to peers and faculty. This structure transforms a weekend of intensive training into sustained projects, measurable efficiency gains, and a clearer vision for how AI and automation will reshape healthcare delivery.