By now, many companies can tell a confident story about artificial intelligence (AI) adoption. Employees have the tools, and leaders have taken the training. Usage is climbing. By every metric companies usually track, the transformation is well underway.
New research from LHH brands General Assembly and EZRA, surveying more than 500 directors, vice presidents and C-suite leaders across the U.S. and UK, confirms this sentiment: 82% say their teams now use AI regularly, 79% of leaders have taken some form of AI training and 93% personally encourage their people to use it. The infrastructure for adoption is largely in place. But ask what leaders are truly doing with these tools, and the day-to-day reality is much narrower. Most use AI to search for information (69%), summarize documents (68%) and draft emails (58%). Far fewer use it for work that moves a business, like scenario planning (27%), organizational design (27%), financial modeling (28%) or governance and compliance (28%).
Thus, many senior-level workers are using the most powerful technology of their careers the way they’d use a savvy intern: hand it the busywork and check back later. There’s so much more that AI is capable of, but today, the way many talent leaders define “AI fluency” enables them to get away with the basics.
The disconnect has less to do with the tools themselves than with how organizations are adapting around them. It’s clear that AI has plenty of transformative potential. What’s missing is the managerial and operational capability to turn that access into a genuinely different way of working. Until that catches up, “AI adoption” will keep describing a fleet of tools that have been handed out but never really wielded to their full potential.
Here’s where you can train leaders to focus instead:
- Adoption isn’t the same as changed work, so stop measuring exposure.
Most organizations gauge AI progress through training completion rates, licenses distributed and usage rates. Those numbers climb reliably and tell you almost nothing about whether the work itself is different. The story is in the research. Despite near-universal adoption, only 52% of leaders say they’ve reworked team structures or workflows around AI. The rest have layered a new tool on top of an unchanged process, which is exactly how you get high usage without a change in results.
A more honest question is harder to answer but worth more: Where has AI moved past experimentation and changed how work gets done? You can see it when junior analysts take on work that used to require a manager’s oversight, when teams move from draft to decision faster, and when output improves without added time or headcount.
- The real bottlenecks are organizational, not technical.
When AI initiatives stall, the instinct is to blame the model. The data tells a different story. Among leaders who reversed or scaled back an initiative in the past year, 41% said the AI’s work didn’t meet quality expectations, 36% said their data wasn’t ready, 32% cited resistance to changing workflows, and 30% said it never saved meaningful time or money.
Almost none of the challenges with AI implementation are about the technology itself. They’re about quality standards, data infrastructure, change management and operational design, the unglamorous plumbing that determines whether a capable tool produces anything useful. The next phase of AI transformation depends far less on better models than on whether organizations can adapt around the ones they have.
- Managers need different skills, and training is the dividing line.
Our research found that 73% of vice presidents have taken any AI training, compared to 88% of directors. Only 39% have reworked workflows around AI, versus 71% of directors. And 68% don’t know what “vibe coding” is.
That’s a problem, because AI is rapidly changing the coordination, review and oversight functions that define much of middle management. The moment demands different muscles: workflow redesign, AI oversight, operational judgment and the change management skills to move a team from an old process to a new one.
Fortunately, this is teachable, and the research is blunt about what works. Leadership-specific AI training is the single sharpest dividing line in the dataset. Leaders who took it were far more likely to have restructured workflows (68% versus the 52% average), built an AI workflow themselves, and developed a rubric for evaluating AI-assisted work (71% versus 52%). Training doesn’t just raise confidence. It changes what (and how) leaders can build.
- Redefine what “ready” really means.
Most L&D teams still define AI readiness through participation — courses completed, engagement logged and pilots launched. Those metrics are easy to collect and easy to report to a board, which is exactly why they persist. They’re also why so many pilot programs remain pilots.
Readiness should mean something operational instead. Are workflows being redesigned? Are decisions getting better and faster? Are AI-enabled processes being reused rather than treated as one-offs? Can a manager point to a specific project where training clearly changed the approach? Those questions are harder to measure but they’re the only ones that correlate with true advantage.
The Real Test
Today, the “easy part” is behind us. Getting tools into people’s hands turned out to be a procurement problem and most companies are good at procurement. Licenses were bought, logins were created, training was scheduled and dashboards lit up green.
What comes next is harder, because it can’t be solved through procurement alone. The real test is whether teams using AI truly operate differently than they did 18 months ago — whether work moves faster, output improves and managers can point to processes that were fundamentally redesigned around the technology.
The organizations that pull ahead will do more than roll out new tools; they’ll redesign how work gets done. Everyone else will end up with an expensive AI stack and roughly the same results they had a few years ago.
