I spent six years on one SAP stack. I learned it inside out. I knew where every bone was buried. And I’m not unusual. Plenty of people I’ve worked with built entire careers on a single system, some for a decade or more. That was the deal. You went deep, and depth paid.
Then the pace changed. On-premise to cloud. Cloud to AI. AI to generative AI. Now agentic AI. Each wave landing before anyone had fully absorbed the last one. The ground doesn’t just shift anymore. It shifts faster than you can retrain for it.
It turns out that feeling has a price tag. And it’s enormous.
The number nobody wants to own.
In 2026, 82 per cent of enterprise leaders say their organisation provides some form of AI training, according to DataCamp’s survey of more than 500 US and UK business leaders. And yet 59 per cent of those same leaders still report an AI skills gap.
Sit with that for a moment. Most companies are already doing something about this. And it still isn’t working.
The reason isn’t hard to find. Only 35 per cent of leaders report having a mature, organisation-wide upskilling programme. The rest is fragments: a course here, a workshop there, optional modules that nobody follows up on. Training exists. Capability at scale does not.
And the cost of the wider tech skills shortage this sits inside? IDC estimates $5.5 trillion by the end of 2026 in product delays, quality problems and lost competitiveness. That forecast was originally $6.5 trillion. IDC trimmed a full trillion off it because AI coding tools are picking up some of the slack. Which tells you something in itself: the tools are moving faster than the people.
The enablement illusion.
Gartner surveyed more than 12,000 employees and managers in the first quarter of 2026, and gave the core problem a name: the enablement illusion. In their Q1 2026 survey of 12,000 employees across 40 countries, they found that most leaders are mistaking basic access to AI tools for actual transformation.
Almost everyone gets access to AI tools now. But 73 per cent of the people who are genuinely, highly productive with AI are managers and executives, not the individual contributors who actually do the majority of automatable work. The people with the most to gain from AI get a login and a link to a course. The coaching, the context, the support? All of that pools at the top.
It gets worse when you look at what leadership is measuring. Many executives track AI success by hours saved. Yet 19 per cent of employees told Gartner they’ve saved no time at all with AI. Time saved is the wrong yardstick. What actually predicts productivity is deep, varied use of AI across different tasks, in ways that improve the quality of the work.
And here’s the number that hit me hardest. In a separate Gartner survey of nearly 200 senior business leaders in December 2025, only 27 per cent of executives said they have a comprehensive AI strategy. Just 20 per cent believe their workforce is truly AI-ready. We’re pouring money into tools and training while the structural foundations aren’t there.
Nobody is redesigning the work.
Deloitte’s 2026 State of AI in the Enterprise report, which surveyed over 3,200 leaders across 24 countries, makes the pattern impossible to ignore. 84 per cent of organisations have not redesigned jobs or workflows around AI. Not one job. Not one workflow. Despite most of them expecting a significant share of their work to be automated within three years.
Look at where the effort actually goes. The most common response is education: 53 per cent are training the broader workforce, 48 per cent are running upskilling and reskilling programmes. But only 33 per cent are redesigning career paths, and only 30 per cent are rethinking how work is actually organised around AI.
Companies are reaching for training first and restructuring last. Education is the easy lever. You can point at a completion rate and call it progress. Redesigning how work flows is harder, slower, and it’s the thing that actually determines whether any of this pays off.
A December 2025 Gartner survey adds the social layer: 37 per cent of employees with access to AI simply aren’t using it, for one simple reason: their team isn’t using it either. Skills don’t spread through courses. They spread through the people around you. That’s social inertia, and no training programme fixes it.
This is a trust problem wearing a skills costume.
Here’s the stat that ties everything together. Gartner found that employees with a positive outlook toward AI are 3.4 times more likely to be highly productive with it.
Three point four times. That is not a training number. That is a culture number.
You can see the tension generationally too. Research from Writer and Workplace Intelligence found that 41 per cent of Gen Z and millennial employees admit to actively working against their company’s AI strategy. The people most willing to adopt AI are pushing against systems that weren’t built for them. Old approval chains. Old ways of measuring good work. Meanwhile, the people who designed those systems sign off on another round of training.
Gartner’s Future of Work research puts it plainly: the winners of this transition will be process pros, not tech prodigies. The advantage doesn’t go to whoever knows the most about AI. It goes to whoever changes how the work is done.
McKinsey’s Global Tech Agenda points at the same divide from another angle: a widening maturity gap between companies that build AI capability inside, with their own people on their own work, and those still treating it as something you outsource to a vendor. That gap compounds.
What actually worked for us: Explore, Execute, Embed.
I didn’t get my answer from a report. I lived it.
A small team of us started exploring agentic AI in parallel with our day jobs. No pressure, no deadline. Just structured time to poke at it. That’s Explore.
Then one person crossed over properly. They went and built an AI agent into one of our flagship products. Real work, real stakes, with experienced support close by. That’s Execute.
Then they came back. Not as someone who’d completed a course, but as the in-house expert everyone else now goes to. The skill stopped being a certificate and became part of how the team works. That’s Embed.
And there’s data behind why this shape works. Overall, only 21 per cent of leaders see significant ROI from AI investments. But that doubles to 42 per cent for organisations with mature, structured upskilling programmes. Structure matters more than spend.
The version 1.0 mindset.
The uncomfortable truth underneath all of this: most organisations are still running a version 1.0 mindset in a world that’s shipped several major releases since. Train once, certify, deploy, repeat. That model was built for skills with a ten-year shelf life. AI skills have a shelf life measured in months.
So the question isn’t whether your company offers AI training. 82 per cent already do. The question is whether anything about the work itself has changed. Whether anyone redesigned the job, the incentives, the career path. Whether your best people are exploring, executing and embedding, or just collecting certificates while the ground moves underneath them.
Because right now, most companies are teaching people to swim by showing them videos of water.