Building Tomorrow's Workforce, Today: Scaling Internal AI Academies

Summary

Three learning leaders compared notes on how they built AI training programs that reach an entire workforce, not just engineers.

The webinar, hosted by DataCamp's Richie Cotton, brought together Dr. Giorleny Altamirano Rayo (chief data scientist and responsible AI official, U.S. Department of State), Carolann Diskin (senior technical program manager, Dropbox), and Mike Baylor (chief digital and AI officer, Lockheed Martin) to discuss what it actually takes to run a company-wide AI training program. All three agreed on a starting premise: everyone needs some form of AI training, from the C-suite to frontline staff, though the depth and focus should match the role. The conversation moved through motivation, program design, promotion, engagement, and measurable payoff. Altamirano Rayo described a State Department program that cut report-writing time by 50% at one embassy, while Baylor detailed a framework for weighing the aggregate value of training tens of thousands of employees against training a smaller pool of specialists. Diskin outlined how Dropbox turns its most enthusiastic learners into internal champions who recruit their peers. The panel closed with practical advice for organizations just getting started: keep the language simple, mix training formats, and get visible buy-in from senior leadership before scaling.

Key Takeaways

• Everyone in an organization benefits from some level of AI training, but the depth and focus should match each person's role, from executives to engineers.
• Foreign Service Officers at the U.S. Department of State must complete AI and data courses as part of required "tradecraft" training, and completion counts toward their promotion file.
• Lockheed Martin built a job code and career pathway that lets employees move from a general AI/ML fundamentals course into a formal machine learning engineer role.
• One embassy in the State Department's "post data" program cut the time staff spend writing field reports by 50% using AI tools.
• Dropbox turns its most engaged learners, employees who top leaderboards or complete boot camps, into internal champions who recruit and mentor their peers by word of mouth.
• Lockheed Martin measures AI training ROI by modeling the aggregate value created across roughly 80,000 employees receiving general AI training, arguing it can exceed the value generated by a smaller group of around 5,000 specialized AI/ML engineers.
• Blended learning, mixing self-paced video, hands-on boot camps, and external platforms, works better than a single training format because employees absorb information differently.

Deep Dives

Who needs an AI training program?

All three panelists reached the same answer almost immediately when Cotton asked who inside their organizations needs AI training: everyone. Altamirano Rayo put it bluntly. "So in a word, everyone," she said of who needs training at the State Department, then backed the claim with numbers: over the past three years, the department has delivered more than 81,000 person-hours of training in data and AI. That training runs on three tiers. Six courses, run through the Foreign Service Institute, target self-identified "consumers and developers" of AI who are ready to go deep into technical material. A second tier folds data and AI content into required tradecraft courses that every Foreign Service Officer must complete, and which count toward their promotion file. A third tier serves ambassadors and deputy chiefs of mission, who need to understand what tools exist and how to apply them to mission-critical work.

Baylor described a similarly wide net at Lockheed Martin, framed around inevitability rather than obligation. "There won't be a corner of the corporation at Lockheed Martin that won't have AI touch it one way or another in the next, you know, five, ten years," he said. That belief translates into open access: anyone at the company can take the training, though some tracks are built for specific audiences, such as a generative AI course for business leaders and a separate track for people who want to build models themselves.

Diskin's answer at Dropbox reinforced the pattern: training that starts with senior leadership and cascades down. "It does start from the top down, really," she said, "from our senior leadership driving that, empowering our teams to expand our skill sets." Dropbox pairs online coursework with internal boot camps built by its own machine learning team, on the theory that hands-on application matters as much as theory. The consensus across all three organizations was that breadth beats exclusivity: an AI training program built only for specialists misses most of the workforce, and most of the opportunity.

Turning AI training into a career pathway

At Lockheed Martin, AI training isn't just an option. It connects to a specific career pathway. Baylor described a program built on the idea of "build once, use many": a foundational team develops reusable training materials and technology once, then rolls them out across the roughly 100,000-person corporation. One product of that team is an AI/ML fundamentals course that, once completed, qualifies an employee to move into a formal machine learning engineer job code. "We find a lot of those individuals that are currently employed and working in my organization came from that," Baylor said, "which is fantastic."

The appeal, Baylor argued, is that employees who take this route arrive with something universities can't teach quickly enough: domain knowledge. "Half of it's the technology piece, the other half is the domain," he said, adding that Lockheed employees typically bring the domain side already and learn the technical skills on their own time. That combination, he said, someone who already understands the business problem paired with new technical ability, produces what his team internally calls "unicorns." He also pointed to a difference in engagement between people who invest their own time to learn AI and those assigned mandatory training: "You get really good engagement in employees as an artifact of that, versus a mandated type of training."

Altamirano Rayo described a parallel structure at the State Department, where completing data and AI training inside the required tradecraft courses "goes through towards their promotion file." Checking that box isn't optional for Foreign Service Officers, but it carries a direct, visible payoff for their careers. Both organizations treat AI literacy less like a compliance exercise and more like a credential, something that opens a door to a different role or a stronger promotion case, which gives employees a reason to finish the training beyond a company mandate.

Driving employee AI adoption: demos, champions, and word of mouth

Getting people to sign up for an AI training program is one problem. Keeping them engaged once competing priorities show up is another, and each organization solved it differently. At the State Department, Altamirano Rayo's team leans on live demonstrations of a chatbot the department is currently beta-testing, built to safely process internal department data. "We demo things all of the time," she said, and letting employees interact with the tool directly does more than any lecture: it "inspires new ways of thinking or creativity about what other use cases could we envision."

Dropbox relies more heavily on peer champions. Diskin described employees who top the company's internal leaderboards volunteering to lead onboarding sessions for colleagues who are still on the fence about taking a course. "These are the folks that really will do the promotion for you without any effort really, to be honest, because they're so passionate and so driven," she said. Dropbox also runs boot camps taught by its own AI and ML experts, which Diskin said have generated enough interest that "we have multiple signups for our next one."

Lockheed Martin has seen similar organic spread, driven largely by engineers talking to other engineers about problems they solved with AI. "It's been a blessing that we haven't had to do a ton of internal promotion," Baylor said, though he acknowledged the company is now working through the early-adopter phase and turning its attention to employees less naturally drawn to new technology. All three organizations also lean on more conventional incentives: Dropbox hands out branded notebooks and coffee cups to learners who complete courses, and Diskin said the company shares internal success stories in newsletters and Slack channels to keep momentum visible.

Measuring the ROI of an AI training program

Justifying the cost of training an entire workforce requires a different argument than justifying a single engineer's course, and Baylor said Lockheed Martin built a specific model to make that case. His team split the corporation into two groups: roughly 5,000 high-end machine learning engineers building AI capabilities directly, and around 80,000 other employees who need broader, general AI literacy, things like basic generative AI use or retrieval-augmented generation techniques.

"Each one of those people might not be a huge percentage of value on the high end of the ramp with the 80,000 people per person," Baylor explained, "but in aggregate, the value is going to potentially exceed the, you know, hardcore AI/ML engineers when you look at the overall value." The framing matters, he said, because it lets leadership see training spend as a portfolio decision rather than a line item aimed at a small specialist group. "Articulating that story up the chain to executives to justify the expense is really important, because you can't get there without training."

Altamirano Rayo offered a more concrete example of return from the State Department's post data program, which brings data and AI capability to embassies around the world. At one embassy, staff now use AI to cut the time spent writing cables, the department's standard field reports, by roughly half. "Our embassy is able to save 50% of their staff time writing cables," she said, noting that staff still spend time fact-checking AI output given known risks around hallucination. Cotton called the result "a massive win," pointing out that most productivity gains in other contexts amount to a fraction of that. Both examples point to the same underlying argument: return on an AI training program doesn't show up only in the output of a handful of specialists. It shows up in time returned to a much larger group of people doing the same job slightly faster.

Getting started: practical advice for building an AI training program

For organizations that haven't started yet, or have stalled, the panel's closing advice converged on a few practical points. Diskin's guidance was direct: "I say go for it," she said, arguing there's no reason to wait given how central AI is becoming to every industry. She recommended starting at a foundational level, running demos and show-and-tells, bringing in outside experts to talk about industry trends, and securing visible buy-in from senior leadership early, since that support "resonates with your workforce."

Altamirano Rayo offered a three-ingredient recipe. "Make it intuitive and accessible" comes first, she said, followed by casting "a wide net" so training reaches people well beyond self-identified data specialists, and finally building enough of a foundation that employees can "incorporate additional knowledge on emerging technology" as the field keeps moving.

Baylor's advice focused on prioritization and where good ideas actually come from. "A lot of times training and these big activities get put on the back burner, but you need to make it a top priority," he said. He pushed back on the instinct to treat AI training as something mainly for engineers: "The best ideas are going to come from people buried within the organization that aren't AI experts, that get plugged in." On format, Diskin recommended blended learning, mixing self-paced video, hands-on boot camps, and external platforms like DataCamp, since people absorb training differently. "Bite-sized chunks, bring people in slowly and encourage them to keep learning," she said, beats a single long course.

Across a government department, a software company, and a defense contractor, the advice landed on the same basic model: start broad, make it easy to begin, tie it to something employees care about, and get leadership visibly behind it before scaling.