Why It Matters
For public, academic, and state-affiliated health systems, a tool that performs well technically can still be unusable because of governance rules the buyer cannot negotiate around. The Arkansas example shows how that plays out. Autonomous coding is specifically designed to finalize codes without a human review step, which puts it directly at odds with a human-in-the-loop requirement.
Terri also noted that payers are starting to ask her to document the technology and vendors her operation uses, which she suggested may be heading toward contract language. That raises the stakes on vendor diligence. Buyers who evaluate AI only on performance and ROI can end up with sunk implementation spend, disrupted workflows, and operational gaps when a tool has to be removed.
Key Takeaways
- A human-in-the-loop requirement forces human oversight of an AI-enabled workflow rather than allowing it to operate fully autonomously.
- Arkansas state regulations covering state entities require a human in the loop for AI use, which disqualified an autonomous coding tool at the University of Arkansas after it had already been implemented.
- Autonomous coding is one of the clearest examples of a technology category that can conflict with human-in-the-loop requirements because it is built to run without a human review step.
- Governance and regulatory viability belong on the RCM vendor evaluation checklist alongside performance, integration, and workflow fit.
- Removing an AI tool after purchase creates workflow disruption and lost investment that earlier governance review can prevent.
Expert Perspective
Terri described the Arkansas situation directly on the RCM Reframed podcast. After arriving at the University of Arkansas, she worked to install safety-net technologies the system had been missing, including autonomous coding. Because Arkansas is a state entity, state regulations required a human in the loop for AI use. The autonomous coding tool she implemented had no human review step in its workflow, so she had to pull it out.
Her broader AI strategy is consistent with that philosophy. She described her strategic plan as redefining burden with AI, machine learning, and RPA while keeping staff and moving them from repetitive transactional work into critical thinking and oversight roles. She summarized the shift as going from burden to bandwidth. A human-in-the-loop requirement fits that model because it assumes AI augments the workforce and preserves human accountability for decisions that affect patient accounts and clinical documentation.
The practical implication for RCM technology buyers is to add a governance checkpoint early in the vendor evaluation. Before signing a contract or starting an implementation, confirm whether state, system, or institutional policy requires human review of AI output. Then confirm whether the vendor's workflow includes a native human review step or whether the product was built to run autonomously. In a state-entity environment, that single question can decide whether a tool is viable at all.
Frequently Asked Questions
What does human-in-the-loop mean in the context of healthcare AI? In this context, it means an AI-enabled workflow must include a human oversight step rather than running fully autonomously. The human remains accountable for reviewing the output before it takes effect.
Does human-in-the-loop apply to all healthcare organizations? Not uniformly. Terri's example is specific to Arkansas as a state entity, where state regulations required a human in the loop for AI use. Any provider evaluating AI should check what applies to their own organization based on state law, system policy, and institutional type before buying.
Can autonomous coding tools be used in environments with human-in-the-loop requirements? Not without a human review step in the workflow. Terri described implementing autonomous coding at the University of Arkansas and then removing it because there was no human in the loop, which conflicted with Arkansas requirements for state entities.
How should RCM buyers evaluate AI tools against human-in-the-loop requirements? Add a governance checkpoint early in the evaluation. Confirm whether state, system, or institutional policy requires human review of AI output. Ask vendors whether their product includes a native human review step, how that step affects throughput, and whether the operational case still holds with a human in the loop.
