Most of Your Executives Think AI Is Tearing the Company Apart

Sharon Gai
Jun 1
8 min read

The short version: In a 2026 survey, 54 percent of executives said adopting AI is tearing their company apart, even though 91 percent of businesses use AI and only about 29 percent see real returns. The cause is structural, not technical: powerful AI dropped into org charts built for human throughput. The fix is naming one accountable human, a beekeeper, for each AI workflow.

Key takeaways

54 percent of C-suite leaders say AI adoption is tearing their company apart, and only about 29 percent report real ROI.
The models work; the bottleneck is organizational design, not the technology.
The chaotic free-for-all comes from pouring two-role (AI plus human) work through a six-tier org chart.
Fix it by naming one accountable human per AI workflow and stripping out the tiers between them and the outcome.

I read a lot of survey data and most of it is forgettable. This one was not. In a recent study of 2,400 knowledge workers and executives, 54 percent of C-suite leaders said adopting AI is tearing their company apart. Not slowing them down. Not creating growing pains. Tearing apart.

The rest of the numbers tell you why. 79 percent of organizations report real challenges adopting AI, a double-digit jump from the year before. 56 percent say AI has created power struggles and disruption inside the organization. 78 percent of executives say it has created tension between IT and the rest of the business, and 55 percent describe their own company's AI use as a chaotic free-for-all. Meanwhile, only about 29 percent report meaningful return on their generative AI investment, and nearly half call the whole thing a disappointment.

So let me state the paradox cleanly, because it is the most important thing happening in enterprise AI right now. Adoption is near universal. 91 percent of businesses say they use AI in at least one capacity. And results are nearly absent. Over 80 percent of firms report no measurable bottom-line impact. Everyone is doing it. Almost nobody is winning at it. Why?

The models are not the problem

I want to clear one thing off the table immediately, because executives love to blame the technology and it lets them off the hook. The models work. They are good enough, today, to transform real workflows. The 60 percent hallucination reduction in the latest frontier models, the agents that now run multi-step tasks unsupervised, the open models approaching the quality of closed ones, all of it means the capability ceiling is not what is holding companies back.

The problem is structural, and the survey practically screams it. Power struggles. IT versus the business. A chaotic free-for-all. Those are not phrases you use to describe a technology failure. Those are phrases you use to describe an organization that took a powerful new capability and dropped it into a structure that was never designed to hold it.

Picture what actually happens inside most companies. Leadership announces an AI mandate. Every team buys its own tools. Nobody agrees on who owns the outputs, who is accountable when something goes wrong, or which workflows are supposed to change. IT wants control for security and governance. The business wants speed and autonomy. Both are right, both are fighting, and the result is exactly what the survey found. Fifty-five percent describe it as a free-for-all because that is literally what it is.

Why bolting AI onto an old structure produces chaos

Here is the mechanism, and once you see it you cannot unsee it. A traditional org chart is built around human throughput. Work flows up and down through layers of people, each one adding a little, each one a checkpoint, each one moving at human speed. That structure evolved over a century to coordinate human labor, and it does that job reasonably well.

Now you introduce a capability that produces work at machine speed and machine volume. You have not changed the structure, so all that machine output has to flow through the same human checkpoints, the same approval layers, the same coordination meetings. The result is a traffic jam where you expected leverage. The AI generates ten times the volume, and the human structure around it cannot route, judge, or approve ten times the volume, so the whole thing seizes up and people start fighting over who is in charge of the mess.

That is the tearing-apart that 54 percent of executives are describing. They added a machine engine to a chassis built for human legs, and now the thing is shaking itself to pieces. The fix is not a bigger engine. It is a different chassis.

The two-role company

The structure that actually holds AI is something I call the hive, and its defining feature is that it has only two roles. There are bees, which are the AI systems doing volume work at machine speed. And there is the beekeeper, the human who directs the bees, judges their output, and owns the outcome. That is the whole org chart for an AI-native workflow. Two roles.

The reason most companies are in chaos is that they have a hive's worth of bee labor running through a structure with six tiers of humans, none of whom have clearly been designated as the beekeeper. When nobody owns the output, everybody owns the output, which means nobody is accountable and everybody is fighting. The free-for-all is not a cultural problem. It is the predictable result of pouring two-role work through a six-tier chart.

So the fix is concrete and almost boring. For each workflow where you have introduced AI, name one human as the beekeeper. One person, accountable for the output, with the authority to direct the systems and the judgment to decide what good looks like. Strip out the tiers between that person and the outcome. Let the AI do the volume. Let the beekeeper own the result. The power struggles dissolve because there is no longer ambiguity about who is in charge.

Governance is not the enemy of speed

One more thing the survey surfaces, the IT-versus-business tension, deserves a direct answer, because it is where a lot of AI initiatives go to die. IT wants governance. The business wants velocity. Most companies treat these as opposites and pick a side, and whichever side loses sabotages the initiative out of spite.

They are not opposites. The free-for-all is what happens when you have velocity without governance, and the stalled pilot is what happens when you have governance without velocity. The beekeeper model resolves this because it puts accountability in a named human, which is exactly what governance requires, while removing the tiers that slow the work down, which is exactly what velocity requires. You can have both, but only if you redesign the structure instead of arbitrating the fight.

The gap between the leaders and everyone else is widening

If you want evidence that this is a design problem rather than a technology problem, look at the spread between the companies winning and the companies stuck. Recent data shows that frontier firms now use 3.5 times more AI intelligence per employee than typical firms, with the largest gaps in advanced agentic workflows. Everyone has access to roughly the same models at roughly the same price. The thing separating the leaders from the laggards is not what they bought. It is how they organized around it.

That same survey carried a darker statistic that I cannot ignore, because it shows how badly the chaos can curdle. 60 percent of companies said they plan to lay off employees who will not adopt AI. Read that alongside the finding that more than half describe their AI rollout as a chaotic free-for-all, and you see the trap forming. Leadership creates a structureless mandate, the rollout descends into confusion and power struggles, results do not materialize, and then management blames the employees and threatens their jobs. The structure was never built to succeed, but the people get held responsible for its failure.

I have seen this pattern up close and it is corrosive. You cannot threaten your way out of an organizational design problem. If the workflow has no beekeeper, no clear owner, no redesigned process, then telling employees to adopt AI harder is like telling drivers to merge faster into a traffic jam. The problem is the road, not the drivers.

The Double 11 lesson the winners already learned

Years ago at Alibaba, during Double 11, the largest shopping event on the planet, my team faced a volume problem no amount of human effort could solve. We needed an enormous quantity of tailored marketing creative in a compressed window, far beyond what any human team could produce. We turned to generative systems, and they delivered at a scale that reset my understanding of what was possible.

But the reason it worked was not the technology. It was that we redesigned the work around it. The humans stopped producing the creative and started directing the systems, judging the output, and owning the campaign results. We did not bolt a machine onto the old process and hope. We built a new process where the machine did the volume and the humans did the judgment, and we were explicit about who owned what. That is the move the 29 percent of companies reporting real return have made, and it is the move the 80 percent reporting nothing have skipped.

The companies tearing themselves apart are running the old process with a new engine attached. The companies pulling ahead rebuilt the process so the engine has somewhere to go. The technology is identical. The redesign is everything.

And the redesign is not a one-time project. The reason the leaders keep their 3.5x advantage is that they treat workflow redesign as a continuous practice, revisiting it every time the models get more capable, which is now every few months. Each capability jump opens a new set of tasks the bees can absorb, which means a new set of decisions about which humans move up to beekeeping. The laggards run one big AI initiative, declare victory, and freeze. The leaders never stop asking what the machines can newly do this quarter that they could not do last quarter, and they restructure accordingly. That habit, more than any single tool, is what compounds into a gap competitors cannot close by buying the same software.

The uncomfortable read on your own company

If you are an executive and these numbers feel familiar, I want to offer the uncomfortable version of the diagnosis. The chaos you are experiencing is not a sign that AI is too immature for your company. It is a sign that your company's structure is too immature for AI. The technology is exposing an organizational design problem that was always there, just hidden, because human throughput moves slowly enough to paper over a messy chart.

The companies that will pull away from the pack this year are not the ones with the best models. Everyone has access to roughly the same models. They are the ones that redesign first, that name their beekeepers, that strip out the tiers, that decide who owns what before the volume hits. The 29 percent reporting real return are almost certainly the companies that did this work. The 80 percent reporting no impact are almost certainly the ones that bought tools and changed nothing else.

So here is my challenge to you. Pick the one AI workflow in your company that is causing the most friction right now, the one where the power struggle is loudest. Ask a single question: who is the beekeeper here? If you cannot name one person, you have found the source of the chaos, and you have also found where to start. Will you name them this week, or wait until the tension shows up in your attrition numbers?

Sharon Gai is an AI transformation strategist, keynote speaker, and author of How to Do More with Less Using AI. She advises Fortune 500 companies on AI adoption and organizational redesign.