Everyone completed the AI training. The dashboard is green. The completion rate is 98%. And nothing about how the work gets done has changed at all.
This is training theater, and it is the default outcome of most corporate AI upskilling. A module is assigned, a video is watched at double speed, a quiz is passed, a certificate is filed. The organization can now say it has trained its people. Whether anyone can actually do anything new is a separate question that the dashboard does not ask.
The gap between plans and results
The intent is real. The World Economic Forum's 2025 survey found that 77% of employers plan to upskill their workforce, and that 63% name the skills gap as the single biggest barrier to transformation. The problem is not commitment; it is method. Training that stays abstract, generic, and disconnected from a person's actual tasks tends to evaporate within days, no matter how many people complete it.
The evidence on what does change behavior points the other way, toward practice embedded in real work. The randomized BCG trial that measured large AI productivity gains did not just hand people a tool; its design treated hands-on use, and a dedicated prompt-training condition, as the variables that mattered, because using the tool on real tasks is what builds the skill. And the meta-analysis of job-crafting interventions, which reshape how people actually do their jobs, found real if modest gains in engagement and performance, with the strongest designs being those that combined organizational goals with the individual's own. The pattern across both: change comes from doing the work differently, not from hearing about it.
Certifications you can buy
As Ricardo Semler has told it, he once sat on the jury of a best-employer award and watched it go to a bank where the tellers were not allowed to sit down while a vice president served kombucha to guests. His conclusion about culture certifications was cutting: they are paid for, the company runs its own survey, pays the fee, and hangs the plaque. Semco never had a plaque. It had a hammock in the middle of the office.
AI training theater is the plaque version of upskilling. The completion certificate is the artifact; the capability is the thing it is supposed to represent, and the two have almost nothing to do with each other. An organization can accumulate a wall of green dashboards while its people quietly conclude that AI is not for them.
What upskilling that works looks like
Real AI upskilling has a recognizable shape. It starts from the tasks people already do, so the practice transfers the moment the session ends. It is hands-on rather than watch-only, because the skill lives in the doing, in seeing a prompt fail and fixing it. It is paced by peers as much as trainers, riding the sideways diffusion that already moves faster than any central program. And it is paired with explicit permission to change the workflow, because a person who learns a better way but is not allowed to adopt it has gained nothing the organization can use.
It also measures the right thing. A completion rate measures attendance. What matters is whether the work changed: fewer hours on the drudgery, higher-quality first drafts, tasks that used to bottleneck on one expert now handled by more of the team. Deloitte's 2025 finding that organizations capture far more value from AI when workers derive personal value from it is, read carefully, an argument about training too. Upskilling that lands on the worker as a better week compounds. Upskilling that lands as a checkbox does not.
There is a timing point too. Because AI competence spreads fastest peer to peer, the highest-leverage thing a formal program can do is not deliver content but create the conditions for people to teach each other: protected time, a place to share what works, and recognition for the colleagues already doing it. A program that ignores that current and tries to be the sole source of learning is not just theater, it is slower than the informal alternative it displaces.
Burn the plaque
The test for any AI upskilling program is simple and unforgiving: three weeks later, is anyone doing their job differently? If the honest answer is no, the program was theater, however high its completion rate. Capability is not conferred by a certificate any more than culture is conferred by an award. It is built in the work, by people given real tasks, real practice, and real permission to change how they do them.
Sources
- World Economic Forum (2025). Future of Jobs Report 2025. 63% of employers name the skills gap as the top barrier to transformation; nearly 40% of required skills will change by 2030 and 77% of employers plan to upskill their workforce. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Dell'Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, Lakhani et al. (2025). Navigating the Jagged Technological Frontier (Organization Science). RCT with 758 BCG consultants: on tasks inside AI's frontier, GPT-4 users did 12.2% more tasks, 25.1% faster, ~40% higher quality; on tasks outside it they erred more. https://pubsonline.informs.org/doi/10.1287/orsc.2025.21838
- Oprea, Barzin, Virga, Iliescu & Rusu (2019). Effectiveness of Job Crafting Interventions: A Meta-Analysis (EJWOP 28). Across 14 intervention studies, structured job-crafting programs raised work engagement (g=0.31) and contextual performance (g=0.39); effects are real but small. https://doi.org/10.1080/1359432X.2019.1646728
- Deloitte (2025). 2025 Global Human Capital Trends. Organizations are roughly 6x more likely to see significant financial benefit from AI when workers derive personal value from it, yet 77% are doing nothing meaningful about it. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html