Skill You Can See: From Game Traces to Veteran Training Pipelines

Four new signals point the same way. Skill is easier to prove when we watch people act, and employers are still catching up.

Reading skill from actions, not scores

A June 2026 arXiv preprint from the Federal University of Uberlandia, Brazil, describes a multi-agent LLM architecture for stealth assessment in a serious game. An LLM classifier labels each player action in a 2D platformer against the OECD/INFE financial-literacy framework. Four domain agents and per-competency Bayesian Knowledge Tracing estimate mastery. An expert judge agent then combines their views.

The team tested it on 193 students across 264 sessions. Mastery estimates correlated with learning gain (r=0.276) and with the post-test (r=0.333), but not with the pre-test. That pattern is what convergent and discriminant validity should look like. Agreement with human experts reached Fleiss kappa 0.624. The multi-agent design delivered roughly three times the predictive validity of a single LLM, which was not significant on its own.

The sample is small and the domain is not veterans, so treat this as early validation. Still, the principle matters for PowerTechs. A veteran's behavior inside a realistic task is evidence, and it can be scored and checked. Read the paper.

A corporate bet on veteran engineers

Cognizant awarded a grant to VetsinTech to launch an AI for Engineering Training Program for veterans and military spouses. Three cohorts begin in Q1 2027. The program combines AI-focused technical training with career readiness, mentoring and direct connections to employers. Target roles span engineering, aerospace, defense, manufacturing and tech.

This fits a growing cluster of earn-and-learn efforts that route service members into critical industries through practice rather than a diploma. For employers, the open question is how to see a candidate's real ability when the program ends. Source.

Everyone wants AI skills, few can build them

The 2U report, The State of Enterprise Learning 2026, surveyed more than 500 learning and workforce leaders at companies with 500 or more employees. It found that 83% of employers prioritize AI skills, yet only 29% have mature AI capabilities. And 78% admit nobody holds a clear mandate to lead AI upskilling.

The gap is not about interest. It is about the ability to develop skills and confirm they exist. Hands-on practice with measurable outcomes is the missing piece. Source.

Workers cannot see the skills they need

WTW published an analysis arguing that uncertainty about future skills is becoming the next workforce risk. Employees struggle to know which skills they need and how fast they must build them. The share of workers who say their manager helps them develop their abilities slipped from 84% to 78% over four years.

When neither the worker nor the manager has a clear picture, an objective measure of skill helps both sides. For a veteran changing careers, it also gives a plain answer to a hard question: where do I stand today? Source.

Takeaway

Interest in skills is high everywhere. What is scarce is a trusted way to observe them. Game traces, simulated work and employer-ready records can close that gap, and veterans stand to gain the most.

For veterans: see roles that match your military experience For employers: book a 30-minute call