The People AI Was Supposed to Replace Are the Ones Making It Work

Every tech executive with a keynote seems to agree on one thing lately: middle management is dead weight. Flatten the org. Let AI handle the coordination. Cut the translators and let the tools speak for themselves.

I keep waiting for someone to ask the obvious follow up question. If that strategy works so well, why do most AI rollouts still fail?

---

The Coordination Layer Nobody Wants to Pay For

What I think is actually happening: organizations are treating AI adoption like a technology problem when it’s fundamentally a translation problem. The C-suite sets the vision (“we’re an AI first company now”), the tools get licensed, the training decks get sent out. And then nothing changes on the ground floor because nobody is doing the messy, unglamorous work of turning strategy into practice.

That work has always been the middle manager’s job. Not because it says so in the job description but because someone has to sit between the people making decisions and the people living with them. Someone has to know that the sales team won’t touch the new CRM integration unless you frame it around their specific pipeline pain. Someone has to notice that the junior analyst is quietly drowning in AI generated outputs she doesn’t know how to verify.

A 2025 study on generative AI’s impact on middle management found something that should stop the “flatten everything” crowd in their tracks. Rather than making middle managers obsolete, AI is transforming their roles from traditional administrators and communicators into what the researchers call “AI orchestrators, meaning makers, and ethical guardians” [1]. The managers who succeed evolved into that role. They didn’t disappear.

---

The organizations cutting middle managers to fund AI tools are removing the exact people who make those tools actually work.

---

More Demanding, Not Less

There’s a persistent fantasy in AI discourse that automation makes management easier. Less coordination, fewer meetings, more time for “strategic thinking” (whatever that means in practice).

Research from the Journal of Service Research tells a different story. A case study of 25 middle managers leading AI integrated teams in financial services found that AI integration made their work significantly more demanding [2]. They were suddenly responsible for managing the relationship between humans and AI systems, balancing what the researchers described as “dialectical tensions” between efficiency and empathy, automation and judgment, speed and accuracy.

Think about what that actually means on a Monday morning. Your manager isn’t just running the team anymore. She’s also the person who decides when to trust the AI’s recommendation and when to override it. She’s the one who notices the model is surfacing biased candidate profiles before HR finds out. She’s the one who has to explain to a frustrated team member why the AI keeps rejecting their expense reports for reasons nobody can articulate.

That’s not dead weight. That’s the hardest job in the building.

---

What Actually Works

Stop cutting the translation layer. Before you flatten another level of management to fund an AI initiative, ask yourself who is currently doing the work of turning your AI strategy into daily practice. If the answer is “nobody,” that’s your problem. If the answer is “middle managers,” maybe don’t fire them.

Invest in the new skill set instead. The old middle management playbook (scheduling, reporting, information routing) is genuinely getting automated. The new one (AI orchestration, ethical judgment, meaning making across teams) requires serious investment. Train for the role that’s emerging, not the one that’s disappearing.

Let managers experiment first. Most AI rollouts push tools to frontline workers and hope for adoption. Flip it. Give your middle managers three months to live with the tools before anyone else touches them. They’ll find the edge cases, build the workarounds, and create the context that makes adoption possible for everyone else.

And measure translation, not just adoption. Tool usage metrics tell you nothing about whether AI is actually changing how work gets done. Ask your middle managers what’s working, what’s breaking, and what the tools can’t see. They’re your best sensor network and most organizations treat them like an expense line.

---

The conversation about AI and management keeps getting framed as a question of survival. Will managers make it? Can they adapt? But that framing misses the point entirely. The real question is whether organizations are smart enough to recognize that the people closest to the work are the ones best positioned to make AI work.

Every successful AI transformation I’ve seen up close had one thing in common. It wasn’t the budget or the tooling or the executive sponsor. It was a middle manager somewhere in the building who figured out how to make the technology make sense for real people doing real work.

Who is playing that role in your organization right now? And what happens if they leave?

I write about leadership, AI adoption, and the human side of organizational change on LinkedIn. Come say hi.

---

Sources

[1] Tughanbulut Kurtulush et al. (2025). The Evolutionary Influence Of Generative AI On Middle Management Roles And Competencies. IOSR Journal of Computer Engineering.

[2] Jonna Koponen et al. (2023). Work Characteristics Needed by Middle Managers When Leading AI-Integrated Service Teams. Journal of Service Research.