Imagine a newly promoted supervisor at a Tennessee manufacturing facility. One of her employees keeps skipping a safety procedure because it slows down production.
She's completed her training. She knows the policy. She passed the assessment.
Now she has to confront a colleague who has worked at the facility for 20 years, explain why the procedure matters, handle his objections, and make a decision about whether the work can continue.
There's no multiple-choice answer on the factory floor.
This is an overlooked part of workforce development. We spend considerable time teaching people what they need to know. We have fewer opportunities to observe whether they can apply that knowledge when a situation becomes uncomfortable, unpredictable, or emotionally difficult.
With artificial intelligence becoming part of nearly every discussion about the future of work, Tennessee has an opportunity to examine how that final stage of training is delivered.
And there's already money moving toward the problem.
Tennessee Is Building Its Next Generation of Workforce Training
In November 2025, Tennessee's Artificial Intelligence Advisory Council released its first statewide AI Action Plan. Among its four priorities was workforce development, including AI literacy, worker reskilling, and partnerships with educational institutions [1].
The plan also called for measurable AI pilot projects across state government.
That distinction matters. Training people to understand AI is one task. Using AI to improve how people acquire and demonstrate occupational skills is another.
Tennessee is now pursuing both.
On October 13, 2026, the state launched the TNWorks Talent Accelerator, a program that connects employers with training providers and funding opportunities. Its TN Talent component provides access to $5 million in federal workforce funding focused on nuclear energy, AI infrastructure, and advanced manufacturing. Rural Works supports workforce training in rural communities, initially emphasizing healthcare [2].
Two weeks later, the Tennessee Board of Regents announced a $7.6 million U.S. Department of Labor grant involving all 13 of the state's community colleges. The four-year award will help develop short-term career certificate programs in five priority sectors, including healthcare, information technology and AI, manufacturing, logistics, and energy [3].
Then, on October 29, the Board of Regents introduced its first 22 AI Fellows. These faculty members will develop classroom experiences focused on practical AI competencies, including evaluating AI-generated information, verifying results, and applying human judgment. Pilots are planned for spring 2027 [4].
Much of this work began under Governor Bill Lee. As Tennessee prepares for a new administration, Governor-elect Marsha Blackburn has also identified community colleges and Tennessee Colleges of Applied Technology as institutions whose career-training programs she wants to expand [5].
The policy direction is increasingly clear. Tennessee wants workers who can adapt to changing technology and employers who can find people prepared to do the work.
The harder question is how we will know whether that preparation is working.
A Certificate Tells Us Something. It Doesn't Tell Us Everything.
Consider a healthcare worker in rural Tennessee.
She might have completed a course on motivational interviewing, reviewed the relevant clinical guidance, and correctly answered questions about patient-centered communication.
But what happens when a patient refuses a recommended treatment?
Can she identify the patient's actual concern? Can she explain the clinical reasoning without becoming defensive? Does she listen long enough to understand the objection before moving to the next recommendation?
Those skills are difficult to evaluate with a written examination.
The same problem appears outside healthcare.
A construction worker may understand a safety protocol but struggle to challenge an experienced coworker who ignores it. A new manager might know the organization's conflict-resolution policy but have little experience conducting a difficult performance conversation.
These situations involve technical knowledge, judgment, and communication at the same time.
Traditional instruction can establish the foundation. Role-play and simulation offer a way to observe what happens when someone must use that foundation in practice.
Human instructors, peers, and trained actors have provided this experience for decades. They remain valuable, particularly when learners need detailed coaching or when an assessment carries serious consequences.
The difficulty is providing enough opportunities for practice.
Live exercises require scheduling, experienced instructors, and time away from other responsibilities. A program may be able to give every learner one supervised encounter. Providing five attempts, with detailed feedback after each, is considerably more demanding.
AI-powered virtual humans offer another way to provide those additional attempts.
What the Research Actually Shows
The evidence for AI-supported simulation is encouraging, although it is still developing.
A systematic review published in the Journal of Medical Internet Research in May 2026 examined 15 studies involving 645 participants across several healthcare disciplines. Three were randomized controlled trials.
The authors found promising results for clinical reasoning, communication, and performance in some controlled comparisons. They also identified weaknesses in study design, short intervention periods, and limited evidence about longer-term learning [6].
A separate umbrella review published in BMC Medical Education examined nine systematic reviews of virtual-patient and immersive communication training.
The researchers reported potential benefits for short-term communication performance and learner engagement. However, all nine reviews received critically low methodological-quality ratings. Evidence concerning long-term retention and transfer to clinical practice remained uncertain [7].
A more recent randomized controlled trial provides an example of what can be measured more directly.
In a study published October 20, 2026, researchers at the University of Rochester evaluated an AI virtual patient used to teach serious-illness communication. Fifty-one students and practicing clinicians participated.
One group completed approximately 30 minutes of interactive virtual-patient training with automated feedback. A control group studied written materials covering the same communication framework.
The virtual-patient group showed greater improvement in communication scores assessed by standardized patients and independent blinded raters. Participants also reported higher confidence [8].
This was a relatively small trial examining a specific clinical skill. It doesn't establish that the same results will occur across other occupations, nor does it demonstrate lasting improvements in real-world patient outcomes.
Still, the study offers an example of moving beyond satisfaction surveys and knowledge checks.
The researchers measured how participants performed a conversation.
That's closer to the question employers and educators ultimately need answered.
What We've Learned at Xuron
We've spent years working on this problem from the technology side.
At Xuron, we develop AI-powered virtual humans that allow learners to practice realistic conversations and receive feedback on their performance. The system combines educational content, pre- and post-assessments, interactive practice, and analysis of the learner's actual responses.
An instructor can build a scenario around an existing case, define the tasks the learner should complete, and specify how the conversation will be evaluated.
A learner might practice explaining a new diagnosis, speaking with a skeptical colleague, or responding to a patient who is reluctant to follow a treatment plan.
Afterward, the feedback can identify which tasks were completed, point to specific statements from the conversation, and explain where the learner could take a different approach.
Importantly, educators retain responsibility for reviewing the scenario and deciding whether the feedback is appropriate.
In an evaluation conducted across four organizations, we examined 273 first-session completers participating in 11 Xuron learning activities between July 2024 and July 2025.
The activities combined subject-specific education, virtual-human conversations, assessment, and feedback. Knowledge and confidence measures increased after the combined activities, while conversation records provided additional information about where learners struggled to apply what they had learned [9].
That study had limitations. There was no control group, and because the educational content and simulation were delivered together, the results cannot isolate the effect of the virtual-human practice itself. We also did not establish long-term behavior change.
Those distinctions are important to us.
We want to know whether someone has learned something. But we also want to understand what they can demonstrate, where they still need help, and whether additional practice improves their performance.
Built in Tennessee, Reaching Patients Across the Country
There's a personal side to this work for me.
I was born and raised in Nashville. Tennessee is home, and being able to build a technology company here that serves clinicians and learners across the country means a great deal to me.
Xuron grew out of Tennessee's technology and venture capital community. Our team had already spent years building immersive environments and virtual-human systems before much of today's generative AI technology existed.
Local investors helped us continue that work. Nashville-based 3LS Ventures announced its investment in Xuron in January 2025, and we're also part of InvestTN's portfolio of Tennessee companies [10, 11].
We've had the opportunity to build a company here while serving organizations across the United States.
Clinicians taking Xuron activities report how many patients they see in an average week with the condition or situation covered by their course. Annualized, those self-reported patient volumes represent well over one million U.S. patient encounters [12].
A conversation practiced during an educational activity can prepare a clinician for situations they may encounter repeatedly during their career. The potential reach extends well beyond the number of people who complete the training.
I'm proud that we're building this in Nashville. I'm equally proud that Tennessee's technology community has helped make it possible.
And I'd like to see more of that work come home.
Where This Could Fit in Tennessee's Workforce Programs
Healthcare is a natural starting point because simulation already has an established place in professional education.
Tennessee's Rural Healthcare Workforce Development Initiatives currently support healthcare training through hospitals, community colleges, universities, and other organizations across the state. Recipients for the July 2026 through June 2027 period include Lipscomb University, Columbia State Community College, and several regional health systems [13].
These programs address different occupational needs. Some prepare new workers. Others expand existing training capacity or help people obtain professional credentials.
Interactive simulations could supplement portions of that training where communication or applied judgment is part of the work.
For example, a rural healthcare training program could give learners additional opportunities to practice a difficult patient conversation before an instructor observes them in person.
A manufacturing program could include a scenario in which a new employee must recognize a safety concern, explain it clearly, and respond appropriately to a supervisor.
A public-sector training program could help newly promoted managers rehearse workplace conversations that would otherwise occur for the first time with an actual employee.
The cases would be different, but the educational method could remain similar.
First, establish what the learner knows. Then observe them applying it in a simulated situation. Provide feedback, allow another attempt, and examine whether their performance changes.
For programs already using an LMS, these activities could be delivered as an additional component of existing instruction.
None of that means every workforce grant can fund AI simulation. Eligibility depends on the specific award, permitted expenses, program goals, and procurement requirements.
The educational question comes first: which occupational skills would benefit from additional practice, and how would we evaluate whether that practice helped?
Measuring Readiness Without Losing Human Judgment
If Tennessee institutions begin testing these approaches, I think the evaluation design deserves as much attention as the technology itself.
A useful pilot would start with a specific training problem. Perhaps a rural nursing program wants to improve how learners respond to patients who are hesitant about treatment. Or an employer wants new supervisors to handle safety concerns more consistently.
Before developing a simulation, the training team should define what successful performance looks like.
That means identifying observable behaviors, designing an assessment rubric, and deciding who will review the results.
Learners could complete an initial encounter, receive feedback, practice again, and undertake a later assessment. Where practical, trained human evaluators should independently review a sample of conversations to determine whether automated scoring is reliable.
The program should also examine how the system performs across different learner populations, communication styles, and relevant language or accessibility needs.
AI-generated feedback can be incorrect. A score that looks precise is not necessarily valid. An automated system should not become the sole basis for employment, credentialing, or high-stakes clinical competency decisions without appropriate validation.
And a stronger simulation score is still not proof that performance improved at work.
That requires follow-up evidence.
For workforce programs, the measures might include observed communication performance, knowledge retention, instructor time, training delivery costs, and later assessments of job-related behavior.
These measures would give training providers a clearer picture of what their investment accomplished.
The Opportunity in Front of Us
Tennessee has employers who need skilled workers, colleges developing new training programs, and public agencies investing in workforce readiness.
AI creates an opportunity to offer more practice and more detailed feedback than many training programs can currently provide.
The next step is to test that proposition carefully, with real learners, occupationally relevant scenarios, and measures that distinguish completing a course from demonstrating a skill.
As someone who grew up here and is now building a company in this field, I'd welcome the opportunity to work with Tennessee educators, employers, and workforce leaders on that question.
We have the technology to let more people practice before the stakes are real.
Now we need to find out how much that practice helps.
References
- Tennessee Department of Finance and Administration. (2025, November 24). Tennessee Sets Bold Course for AI Leadership.
- Tennessee Department of Labor and Workforce Development. (2026, October 13). TNWorks Talent Accelerator Creates New Workforce Training Opportunities.
- Tennessee Board of Regents. (2026, October 27). Tennessee's Community Colleges Awarded $7.6 Million Grant to Expand Short-Term Career Certificate Programs.
- Tennessee Board of Regents. (2026, October 29). TBR Names Inaugural Class of AI Fellows.
- Praino NS. (2026, November 2). Marsha Blackburn, Jerri Green Talk Business. Nashville Post.
- Jiang J, Ye MZ, Kwok TTO, Wong JYH. (2026). GenAI-Supported Virtual Patients in Health Care Education: Systematic Review. Journal of Medical Internet Research, 28, e82756.
- Kalmring M, Chmelar S, Greimel P, et al. (2026). Teaching Clinical Communication Skills Through Virtual Patient-Based Learning: An Umbrella Review of Systematic Reviews. BMC Medical Education, 26, 1082.
- Haut KG, Hasan M, Carroll T, Epstein RM, Sen T, Hoque E. (2026). Effects of a Generative AI Virtual Patient on Serious Illness Communication Skills: Randomized Controlled Trial. JMIR Medical Education, 12, e93034.
- Xuron. (2025). What 11 Activities Showed About Learning: Multi-Course Evaluation. Evaluation of 273 first-session completers across four organizations, July 2024–July 2025.
- 3LS Ventures. (2025, January 15). 3LS Ventures Invests in Xuron to Drive Innovation in Healthcare with AI-Powered Solutions.
- InvestTN. Portfolio: Xuron.
- Xuron. (2026). Learner-reported patient volumes collected in Xuron activities. Clinicians report the average number of patients they see per week with the condition or situation covered by their respective course. Annualized self-reported volumes represent well over one million U.S. patient encounters. Unpublished learner-response data.
- Tennessee Department of Labor and Workforce Development. Tennessee Rural Workforce Initiatives: July 2026–June 2027 Grantee Awards.