Org Charts Flatten. Dependencies Don’t.

I keep seeing predictions that middle management won’t survive AI. The hierarchy is flattening, and we don’t need managers anymore. I understand why the argument is appealing. It follows a certain logical flow: AI increases leverage, so teams ship faster, and the org chart naturally flattens. If managers were mostly overhead before, why wouldn’t they be the first layer to disappear?

Many of the coordination tasks people associate with management like status updates, meeting notes, and progress tracking are exactly the kinds of work AI should help reduce. I would even argue it’s the type of work managers should have never really been doing, but expectation-setting across teams and across levels of leadership isn’t mechanical coordination. It’s interpretive work. My thinking is that the assumption that this coordination work just disappears is where the argument breaks down the most.

I’ve seen what this looks like when the translation layer disappears, even temporarily. Both times I went on parental leave my teams were aligned and empowered before I left. Both times someone helped cover pieces of my role. But neither time was there a full-time manager doing the translation work I normally handled.

Nothing broke. There wasn’t any particular crisis that happened. But coordination slowed. Expectations got fuzzier. Leaders above the team got pulled closer to execution details. Engineers on the team absorbed alignment work that normally wasn’t theirs. When I returned from leave, the consistent feedback from both above and below my role was the same: they hadn’t realized how much invisible coordination I carried as engineering manager until I wasn’t there.

Let’s put a pin in the debate around whether AI actually helps engineers ship faster and assume for the sake of this post that it does. Faster execution changes the shape of coordination. When teams can move quickly, the cost of misalignment rises just as quickly. Decisions that used to unfold over weeks now unfold over days. Dependency surfaces shift sooner than roadmaps do. The result isn’t less need for coordination. It’s less time to do it well. Unless someone is paying attention to how work is connecting across teams, AI is removing the shallow coordination layer and exposing the deeper one.

In practice, that translation work often looks less like making decisions and more like making expectations understandable. Leadership says “invest in AI acceleration,” but engineers hear “adopt tools.” Engineers don’t typically like being told which IDE to use, let alone a top-down mandate to change how they work. What counts as meaningful adoption? Where is experimentation encouraged, and where does reliability still dominate? How do engineers balance speed gains against production risk in systems that other teams depend on? Those questions don’t answer themselves just because new tools exist.

If anything, AI has introduced a certain amount of havoc on delivery expectation alignment. Leaders assume teams should be shipping faster. Engineers are unsure how to estimate their own velocity. Teams lack historical data to anchor expectations because everything is still so new. Someone still has to help teams interpret what actually changes.

Our most senior engineers are the ones feeling the world turn upside down. AI is changing not only how they’re expected to work, but what being good at that work even looks like. I’ve had to coach senior engineers through the shift from expecting to author every line of code themselves to being responsible for shaping what gets produced with AI assistance. That’s not just a tooling change. It’s a change in how they understand where their value comes from.

I’ve also had to help more gung-ho engineers find the balance between experimentation and responsible engineering because the introduction of AI has not and will not reduce the quality standards we expect in code shipped out the door. That transition doesn’t manage itself.

Managers form an important translation layer in many organizations, but that doesn’t mean managers are the only people capable of doing this work. Senior individual contributors may carry parts of it. Organizations may distribute it differently as they flatten. But expectation-setting across teams and levels of leadership doesn’t disappear because responsibilities move to a different title.

Without that translation layer, adoption rarely fails outright. But you will see how it’s rolled out across the organization fragment.

Standards drift between teams working on the same platform. Some engineers experiment aggressively while others hesitate. Expectations become inconsistent. Instead of unified acceleration, organizations end up with pockets of progress that are harder to pull together into a single coherent engineering strategy.

That’s the real risk: not elimination, but fragmentation across the engineering org that’s invisible until things stop working as they should.

AI doesn’t remove the negotiation of ownership, or shape realistic delivery expectations across platform surfaces, or prevent roadmap optimism from turning into roadmap drift.

If management roles were primarily overhead, organizations would have flattened them long before AI entered the picture.

It would be naive of me not to acknowledge that AI will probably eventually assist with parts of this coordination work too. Maybe significantly. But right now, the need to interpret expectations across teams and levels of leadership is increasing faster than it’s being automated.

AI may very well change the number of managers organizations need. It may even change who does this work and what we call the role. It doesn’t eliminate the coordination work those roles evolved to handle.

Dependencies don’t disappear when you flatten an org chart. Someone still has to manage them.