AI can automate a growing amount of administrative work currently done by managers. Gartner reported in January that CEOs expect AI to delayer middle management, while arguing that agentic AI can automate managers' administrative work. McKinsey makes a similar observation from the other direction: agents can take administrative tasks away from managers.
It's easy to see the implication. If agents can absorb some of the work that sits within management today, perhaps companies need fewer layers of management.
That may turn out to be true. But I think there is another reading of what is happening.
Customer Zero
One of the phrases I like at Culture Amp is Customer Zero. If we are building products that help other companies work differently, we should be willing to experience that change ourselves.
That feels particularly relevant to AI. I work at a company whose product is concerned with how people experience work. We are building AI into that product, which means participating in changes to how other people work. At the same time, our own ways of working are becoming increasingly agentic.
My perspective on AI adoption comes from somewhere in the intersection of those three things: thinking about people's experience at work, helping build products that are changing it, and experiencing some of the same change myself.
I am not a manager, so my view of management is from the other side of the relationship. And one thing that has stood out to me is how little simply having access to AI tells us about whether people will actually adopt it.
The AI Daily Brief recently covered research from Section, an AI education company that studies how AI is being adopted at work. Its June 2026 AI Proficiency Report, based on a survey of 5,026 U.S. knowledge workers across industries and functions, found a fairly stark gap: 69% said their organisation had taken some action on AI agents, but only 16% actually used the agentic tools they had access to.
Microsoft's 2026 Work Trend Index finds a similar mismatch from a different angle. Only 19% of AI users in its survey sat in what it calls the Frontier: high individual capability matched by high organisational readiness. Another 10% had strong individual capability but lacked the organisational conditions to apply it.
Having AI and adopting AI are different problems.
The Human Part of Adoption
That distinction matters because the change isn't evenly distributed.
Some people are eager to hand more work to agents. Others are cautious. People have different levels of confidence with the technology, different ambitions for how they want to work, and different concerns about what increasingly capable agents mean for their role.
An organisation can set a strategy for AI adoption. It can buy tools, publish guidance and provide training. None of that means the change lands the same way for every person.
This is where my experience makes me hesitate about the simple delayering story.
The best managers I have worked with were valuable because they understood their people. They knew when someone needed more context, when they needed room, what they were trying to get better at, and when something wasn't working even if the normal signals still looked fine.
That's not a new theory of management. Google's Project Oxygen found years ago that effective managers coach, communicate, empower and support development. Culture Amp's own research on the manager experience lands in similar territory.
AI doesn't obviously make those things less useful. During adoption, I can see an argument that it makes them more useful.
There is some emerging evidence for that interpretation. Research published in Harvard Business Review, based on interviews across two major consulting firms, found middle managers taking on new responsibilities as AI spread through their organisations: validating outputs, identifying errors and helping their teams build AI skills, while their existing delivery expectations remained.
That study doesn't tell us what management will look like in an AI-native organisation. It does show that AI adoption can create new management work at the same time as agents automate some of the old work.