The Part of Management AI Can't Automate

I’ve been watching the conversation around AI and middle management for a few months now and I think a lot of companies are about to learn something expensive.

Jack Dorsey just announced that Block is moving toward a model where AI systems handle coordination, project tracking, task assignment, and real time business insights. Gartner is predicting that 20% of organizations will use AI to flatten their structures by the end of 2026, eliminating more than half of current middle management roles. And employers already posted 12.3% fewer middle manager jobs last year compared to the year before.

The message is loud and clear. Middle management is overhead, and AI can do it cheaper.

The conclusion is wrong.

The Part Nobody Is Talking About

The tasks AI is replacing (status updates, project tracking, scheduling, reporting) were never the real value of a good manager. They were the administrative scaffolding around the actual work. And yes, AI handles scaffolding beautifully. No argument there.

The problem is that most organizations never clearly defined what middle managers were supposed to be doing beyond that scaffolding. So when you strip the scaffolding away, it looks like the whole role is gone.

It isn’t.

Floyd and Wooldridge (1994) studied exactly this during the reengineering wave of the 1990s and what they found is worth paying attention to now. When companies automated and flattened middle management back then, they lost critical strategic functions. Middle managers were the ones interpreting market signals, translating executive strategy into something teams could actually execute, and building the organizational capabilities that let companies adapt. The researchers called it the difference between ‘dinosaurs’ and ‘dynamos’ and the dynamos were the ones doing work that couldn’t be reengineered away.

We’re running the same experiment again. Just with better technology.

The tasks AI automates were never the hard part of management. The hard part was always the human work, and that’s exactly what most organizations stopped investing in.

What This Means for L&D

If you’re in learning and development right now, this is the moment to make the case for investing in manager capability rather than eliminating managers altogether.

Tomaszewski et al. (2025) found that when healthcare organizations invested in structured leadership development for middle managers, engagement scores improved significantly. But more importantly, the participants reported higher team trust and better leadership effectiveness. Those aren’t things you get from a project management bot.

If I were leading an L&D function right now, three moves would be on the top of my list.

The first is auditing what your managers actually spend time on, separating the coordination tasks (which AI can handle) from the human development tasks (which it can’t). Most organizations have never done this clearly, and that’s why the “replace them all” argument sounds so compelling.

The second is rebuilding manager development around the skills that remain. Coaching, feedback, navigating conflict, building psychological safety, translating strategy into meaning for their teams. These are trainable skills, but most manager training programs still focus on process and compliance instead.

The third is repositioning managers as the people who make AI adoption actually work. Someone has to help teams figure out how to use these tools, manage the anxiety around role changes, and maintain trust during the transition. That someone is a manager, not an algorithm. Give that redefined role a concept your organization can rally around. ‘Human leadership layer.’ ‘Development manager.’ Whatever lands. When something has a name, it becomes harder to accidentally eliminate.

The irony is that the companies cutting middle management to save money are going to spend more rebuilding the capabilities they lost. We saw it in the 90s with reengineering and we’re going to see it again. The 42% of employees who expect their roles to change significantly need someone to help them through that change. AI can surface the data. It can’t sit across from someone and help them figure out what they’re actually afraid of.

I don’t think middle management is dead. I think it’s being forced to become what it should have been all along. If you’re trying to figure out what that looks like in practice for your organization, that’s the kind of work we do at Peak Potential Consulting.

What are you seeing in your organization? Are managers being supported through this shift or quietly shown the door?

I write about leadership and AI at the intersection of human development on LinkedIn. Come say hi.