Do organizations still need managers?

You can remove a layer on an org chart in a single afternoon. You cannot replace the coordination, context, and coaching that layer quietly provided without redesigning how the work flows. That is the mistake many teams are making as they rush toward flatter structures. The goal is sound: get closer to the customer, move faster, and let AI handle the paperwork. The failure is structural: companies are deleting the role before they replace the system.

Managers have become an easy scapegoat for bloat, yet the best research continues to show that strong management correlates with stronger performance. The reason is not mysterious. Good managers make local judgment calls that keep the machine running during ambiguity. They translate big goals into day to day tradeoffs. They coach people so capability keeps pace with complexity. When you cut the role without rebuilding these functions elsewhere, you do not create agility. You create hidden coordination debt that will surface as delays, rework, and churn.

So what does a manager actually do when the job is designed well for modern work. First, they hold judgment. Not rule following. Judgment. The ability to decide when to bend a process and when to enforce it. The ability to weigh short term speed against long term reliability. The ability to read culture and context so the same policy lands the right way across different teams. Judgment cannot be downloaded from a tool. It is built through exposure, feedback, and stakes that are real but survivable.

Second, they carry the human side of performance. In a market where skills decay faster and new tools arrive monthly, people need coaching, not generic training blasts. Coaching is not a vibe. It is a sequence. Clarify outcomes. Observe work as it happens. Give feedback that is specific and rehearsable. Create a path to practice under pressure. Great managers do this on a cadence that compounds. Remove this cadence, and people look busy while capability quietly stalls.

Third, they design work. AI is changing task mix and information flow in every function. Someone must re-scope roles, rethread processes, and decide what the human should still own once an agent or model enters the loop. If that design work is nobody’s job, you will see automation create fragmentation. Handoffs multiply. Exceptions spike. The tool looks powerful in demos and destructive in production.

Finally, they enable actual agility. Real agility is not a standup ritual or a slogan about empowerment. It is the repeated act of moving decision rights closer to the edge without losing coherence at the center. That requires local leaders who understand strategy with enough precision to act without waiting, and who escalate when the pattern shifts beyond their remit. Remove that layer and top leadership becomes a help desk. Speed dies in the queue.

If those are the functions, then the solution is not to defend a legacy manager archetype. The solution is to reinvent the role so these functions are explicit, measurable, and supported by AI rather than replaced by it. Start by separating two kinds of leadership that many teams lump together. Leadership of people and leadership of work. In some environments one person can do both. In others, splitting the role is cleaner. A leader of people focuses on capability building, performance conversations, and mobility across teams so talent does not calcify. A leader of work focuses on delivery flow, process quality, and cross team integration. When a company triples team span without adjusting this split, coaching collapses first. Engagement drops next. Turnover follows.

Next, design decision rights like you would design a product. Write a simple decision charter for each team. What decisions can this team make independently. What decisions require consultation. What decisions require approval. What are the value thresholds that trigger escalation. Decisions about pricing, compliance, security, and brand carry different risk weights. Bind each to a clear boundary rather than vague trust. This is how you push ownership out without creating policy roulette.

Now bring AI into the picture with intent. AI should shrink the cost of the role’s overhead so judgment and coaching can expand. Give managers a single pane to surface team sentiment, workflow bottlenecks, cycle times, and error clusters. Use models to draft performance feedback from real work artifacts rather than memory. Use scenario sims to rehearse hard conversations in a low risk environment. None of this removes the need for human presence. It simply moves the admin off the critical path so the human work can happen at the right depth and frequency.

Because judgment is the foundation, build it like a skill instead of hoping it arrives with seniority. Create rotational exposure that gives managers line of sight into upstream and downstream effects. Pair that with short cycle post mortems that focus on the decision, not the outcome alone. A good decision can produce a bad result when the world is noisy. A bad decision can hide behind a lucky result. Train managers to separate those cases. This is how a culture learns without fear and grows without mythology.

Measure what matters in the role. If you only count delivery, managers will become project chasers. Add coaching quality to their scorecard. Look at how often direct reports exceed goals after targeted feedback. Track internal mobility and time to ramp when people move across teams. Reward leaders who create connectors rather than empires. Connector managers produce outsized performance because they reduce search cost inside the company. They know who to pull in and when. They build networks that move information faster than hierarchy ever could.

Be deliberate about span of control. A large span can work when the work is stable and the leader of people has the space to coach. A smaller span may be necessary when the work is high variance or the team is green. Do not copy a number from a slide. Fit span to complexity, variance, and the maturity of both leader and team. If a restructuring plan changes span, run a pre mortem on what will break in coaching cadence, onboarding, and escalation. Fix those gaps on paper before they appear in real time.

Use structure to support empowerment. If you want decisions at the edge, publish strategy with enough precision that edge leaders can act without guessing. Share product economics, customer segments, and constraint thresholds. Remove secrecy that turns empowerment into risk exposure. When people know the goal and the guardrails, they can adapt without asking for permission. When they do not, they invent local rules and drift from intent.

Design work with AI as a teammate, not a vending machine. When a tool starts making recommendations that feel like profit optimization at the expense of craft, teams will resist even if the math looks good. Solve that identity tension in the role design. Clarify what is creative choice and what is model optimization. Let the tool handle the search space and the human own the taste, the risk tradeoff, and the narrative that ties it back to strategy. When people understand who owns which outcome, adoption follows.

If your company is considering a shift to a flatter structure, run two tests before you cut. First, the two week absence test. If a manager disappears for two weeks, does the team keep its rhythm, or does the work jam at unmarked intersections. This will tell you what invisible routing the role was providing. Second, the decision drift test. Pick three classes of decisions the team makes often. Simulate a change in constraint and watch how the decision moves. If the path is unclear or political, you are not ready to remove the layer that currently arbitrates those calls.

Reinventing middle managers for the age of AI is not a branding exercise. It is an operating choice. Replace manager as gatekeeper with manager as system designer, coach, and judgment carrier. Support the role with the right split between people and work leadership. Publish decision rights so ownership is real. Use AI to concentrate human time on the interactions that change capability, not on the reports that summarize it.

Ask yourself two closing questions. If you stopped showing up for two weeks, what would slow down, and why. If your top manager left tomorrow, would you still trust the way important decisions get made. If either answer worries you, the problem is not a person. It is the design of the role and the support around it.

Your team does not need more slogans about empowerment. It needs clear ownership, practiced judgment, and a manager role built for how work actually gets done now. When you treat the role as a system rather than a title, you get the speed you wanted from flattening with the coherence you thought you had to give up.