Managers Aren't the Overhead. They're the Point.

I keep having some version of the same conversation with founders and heads of people lately. It usually starts practical: they’re looking at their burn, reviewing their org chart, they’re looking at what AI can now do, and they’re asking whether they still need as many managers. Underneath that question is a quieter one, and it’s the one that actually matters: if the work itself is changing this fast, is management even the right lever anymore, or is it just cost sitting between me and my best people?

I understand the instinct. I’ve felt it myself in previous roles and even now with HappyHQ. When every task feels urgent and every hire feels expensive, the manager layer starts to look like the easiest place to cut. It doesn’t produce code. It doesn’t close deals. It’s the thing you can point to and say, we’ll figure that out once we’re bigger.

But I think that instinct is wrong, and I think we’ve been here before.

We’ve Been Here

Go back to the early days of Google. The founders, Larry Page and Sergey Brin, tried flattening the org entirely in 2002 with no engineering managers. The idea was to recreate the freewheeling feel of grad school and get out of the way of smart people doing smart work. It lasted a few months. People started going straight to Larry Page with expense reports and interpersonal conflicts and the hundred small frictions that management actually exists to absorb. So no surprise, they brought the layer back.

What Google Found When It Tried to Prove Itself Wrong

A few years later Google did something more interesting than just reinstating managers. They asked whether managers actually mattered, using the same rigor they’d apply to any product question. That was called Project Oxygen. The team running it says they set out to prove the null case, that management doesn’t move the needle, and they failed to prove it. What they found instead was that even small differences in manager quality showed up clearly in retention, in satisfaction, in how people felt about their own growth. So they went further and mapped out the specific behaviors that separated the good managers from the rest. Not vibes. Behaviors you could name, teach, and measure.

That’s worth sitting with, because Google is exactly the kind of place that should have been able to skip management if any company could. Brilliant, technical, self-selecting people who by their own account viewed management as a distraction from real work. If management earned its place there, on the data, it’s not because Google is sentimental about hierarchy.

Running the Same Experiment Again

Fast forward to now, and we’re running the experiment again, this time with an AI-shaped excuse instead of a grad-school one. I read a post recently by Kellan Elliott-McCrea that put language to something I’d been circling in my own head. He calls managers catalysts. Not in the soft, motivational-poster sense, but in the actual chemistry sense: a catalyst lowers the activation energy a reaction needs to happen. That’s what good management does to change. Without it, you can still force change through, but you’re doing it with heat and pressure, with mandates, and that approach works some of the time while damaging the system all of the time.

He points to what happened at one large tech company that cut deep into its manager layer while also rolling out a new way of building software. Incidents went up sharply. Team sentiment dropped. People were shipping far more code, but the value landing in front of customers didn’t follow at anywhere near the same rate.

I saw a version of that gap at a Product Leadership Summit last week. One of the presenters asked everyone to raise their hands if they were shipping more pull requests. A huge number of hands went up. Then she asked if they were making more revenue. The room laughed and very few hands went up.

That moment has stayed with me. AI can make the process of building and shipping easier, but shipping more doesn’t automatically mean we’re solving something that matters. Just because you can ship doesn’t mean you should ship. We still need to understand the market, listen to our customers, and learn from what they actually do with what we build. I think humans are best suited to that work: asking the next question, noticing what’s going unsaid, and using judgment to decide which problems are worth solving. Quality over quantity still applies, even when quantity gets easier.

Good managers make room for that work and make sure it informs what the team does next. They help people connect what they’re shipping to a customer need and a business outcome, and change course when the evidence says they should. If all we ask of a team is to ship more, we shouldn’t be surprised when that’s exactly what we get.

The Fear Underneath the Fear

Here’s the fear underneath the fear, the one leaders don’t usually say out loud in the first ten minutes of a coaching conversation. It isn’t really about headcount. It’s: if I can’t tell anymore what good management even looks like when half my team’s output is AI-assisted, am I the right person to be leading through this? That’s a harder, more honest question, because it’s about our own ability to lead through uncertainty.

The Job Hasn’t Changed. The Cost of Skipping It Has.

My answer, for what it’s worth: the job hasn’t changed as much as it feels like it has. It’s gotten more important, not less. Managers still exist to build trust, to connect what a team is doing day to day to why it matters, to translate strategy downward and translate reality upward, to invest in people’s growth even when growth doesn’t look like a promotion. AI changes the raw material people are working with. It doesn’t change any of that.

What’s changed is the cost of skipping it. We had a decade, roughly the 2010s and the pandemic years after it, where the market was forgiving enough that mediocre management mostly didn’t show up in the numbers. Money was easy, growth covered for a lot of sins, and the default answer to how do we do more was hire more people. That era is over. The teams that will actually turn AI-driven capacity into real outcomes are the ones with managers who know how to lower the activation energy for change instead of just applying more pressure and hoping.

So when a coaching client asks me whether they still need managers in an AI-native org, I tell them the honest thing: you need them more, and you need them to be better at it than they’ve ever had to be, and it’s our job as leaders to help managers be better.

Better means specific things. It means sharper judgment about what deserves to be built in the first place, as well as knowing when an AI-generated first draft is good enough to hand off and when it needs a human’s added context and taste. It means coaching people through a real and valid insecurity, the sense that their skills are moving faster than they can track, and helping them find the parts of their work that still depend on relationships, lived experience, and hard-won judgment. It means running tighter, more honest feedback loops, because the pace of change makes an annual review look almost quaint.

None of that happens by accident. Google didn’t get Project Oxygen’s eight behaviors from good intentions. They got them from measurement: surveys, individual feedback reports, and a training program built around specific, teachable actions. That’s the model worth borrowing now. Give managers a shared vocabulary for what good looks like. Put your best managers in front of the rest of the room so the standard doesn’t just come from HR. Build a feedback loop fast enough that people learn where they stand before a review cycle forces the conversation.

The tools got faster. The job of turning tools into outcomes people can trust is still, stubbornly, human.