Most companies are running two AI strategies.
The official one lives in slide decks, steering committees, and enterprise agreements.
The real AI strategy runs in the tabs that get minimized when the boss walks by.
After letting teams run wild with expensive tools, Microsoft started canceling most Claude Code licenses across a major division and Uber burned through its entire 2026 AI budget in four months.
In May, Community Bank (a subsidiary of CB Financial Services) had to file a material incident report with the SEC after an employee fed customer Social Security numbers into a personal AI account while polishing a slide deck. The approved tools were too slow or too limited for the actual job. At the same time, the official AI program was still forming its first working group.
This split appears everywhere in the data, including Optro’s 2026 research on the AI oversight gap. Companies announce big investments in approved tools while the real activity happens outside the system. Visibility is almost nonexistent. The numbers on paper look modest because the dashboard only counts what IT signed off on. And now the backlash is here. The same places that opened the spending taps are slamming them shut when the token bills arrive.
People are not doing this for the thrill of breaking rules. The approved tools are too slow, too locked down, or simply do not match the actual friction on their desk that afternoon. The person staying late pasting a transcript into Claude is not rebelling. They are extracting themes before the official research team has scheduled its first meeting.
But the real problem isn’t the defiance. The shadow use is the best customer research a company will ever get for free. It shows exactly where the operating model leaks time and where AI actually creates value.
A few companies have figured this out. BBVA watched employees already using consumer tools on their own and responded by giving them a safe enterprise version with room to build. People created thousands of their own tools on top of it. Adoption was real because the tools solved real problems, not the ones a committee had imagined.
Most places do the opposite. They spot the unsanctioned activity, declare a crisis, and roll out one blessed platform that solves none of the problems that sent people into the shadows in the first place.
The better move is to watch what people are already doing. Make the safe version of that the easiest path. Give teams a fast way to ask for the next capability when the current one falls short. You can call it an audit if it helps get the budget approved. It mostly just means looking at how the work actually gets done instead of how the slide deck says it should.
Leaders who can see the gap between the strategy in the deck and the one already running in the tabs will have the edge. The best adoption starts at ground level anyway. Individuals get better at their own jobs with the tools that actually help them, then patterns worth building on start to appear. Everyone else will keep wondering why their big transformation program keeps delivering polished decks instead of real progress.
How did you like this post?
Your feedback helps me create better content in the future.