Context
Applied AI is a top-priority critical technology area for the DoD, yet even the most well-funded initiatives are unlikely to deliver lasting impact if key lessons from early adoption efforts go uncaptured. Across industries, organizations are discovering the same uncomfortable truth: AI upskilling for the digital workforce isn't just about tool training and adoption. It's about addressing the new skills needed to manage the ripple effects these technologies have on human employees, team structures, and workflows. As Dr. Dennis Brown, Chief Engineer in Software Acquisition for the U.S. Air Force, recently noted: "AI may accelerate artifacts, but it does not replace accountable decisions. Speed without judgment becomes a delivery risk." A team of Airmen developers from the U.S. Air Force Software Factory, BESPIN, recently learned this firsthand — and we are here, with them, to share their story and experience of the unexpected pain points and gaps that shaped new strategies for AI training and adoption, while also highlighting areas that still need solutions.

Case Study
In September 2025, roughly a dozen Airmen completed a week-long AI bootcamp and were placed directly onto a TAK-based app development project — a forthcoming mobile situational-awareness platform — with approval to use AI to the fullest extent possible. They had no well-defined path, no guardrails, no shared spec discipline. They were taught how to start the car, but not how to drive it in traffic.

Four challenge areas quickly emerged. The foundational gap: the team lacked understanding of the "why" behind AI-generated code, leaving them unable to debug what they didn't build. The judgment gap: they couldn't reliably distinguish a hallucination from a correct answer — "AI is trained to make humans happy, not to be correct." The collaboration gap: as the team went heads-down with AI agents, the human loops that make teams work — design consultation, peer review, cross-functional input — quietly dissolved. The human gap: with transactional work automated, Airmen were suddenly expected to handle stakeholder communication and requirements gathering in roles they'd never been trained for.

The gaps forced an intervention. The team abandoned wholly AI-driven development after three sprints, recalibrating to an AI-as-partner model. They ran structured norming sessions to build an explicit human-machine division of labor, developed working guardrails, and made AI a standing sprint retrospective topic — continuously tuning their process rather than setting policy once and walking away. The result: an estimated 40% increase in user-story throughput, a shared teaming model that replaced per-engineer improvisation with common practice, and a team that felt genuinely grounded in and proud of their work.

Some gaps, however, remain open. The collaboration loops AI dissolved were worked around, not rebuilt. Foundational knowledge gaps in newer Airmen were compensated for, not closed. And the MQT program that onboards Airmen into BESPIN hasn't been redesigned for an AI-integrated environment. As one Airman put it plainly: "Absolutely nobody is asking those questions."

In Closing
The BESPIN team didn't set out to build a training model. They set out to build a product — and surfaced something the field hasn't fully named yet: the difference between what in-project experience can teach you, and what needs to be taught before the project starts. The four gaps here are not unique to one software factory. They are emerging wherever AI tools arrive before training frameworks do, which is nearly everywhere. 

This session, told in part by the Airmen who lived it, is an invitation to start the conversation. For each gap we surface, we offer a field-tested intervention and/or a concrete recommendation for how to combat that gap before it has a chance to emerge. Attendees will walk away with a shared language for challenges their teams may already be experiencing, and key strategies for building AI training and adoption approaches that actually prepare people for what comes after the tools arrive. The most important thing we can do is keep asking better questions, so the next generation of Digital Warriors can thrive with AI rather than scramble alongside it.

Conference Review Committee Please Note: 

• This will be an interactive presentation focused on audience participation and engagement.
• Presentation length ~20-30 mins, with ~10-15 min Q&A time with Airmen (this can be manipulated to be longer or shorter depending on conference needs)