Cagan calls the manager purge a cycle that corrects

If your company cut a layer of managers in the last two years, you’ve been told the same story three ways. AI handles the coordination, so the coordinators go. The best builders can now do more alone. The org chart gets flatter and nobody misses the middle.

I’ve believed most of that. The knowledge base behind this publication holds a call that management is compressing structurally: Gallup reports manager engagement fell from 27% in 2022 to 22% in 2026, Meta and Amazon announced cuts, and Staff and Principal engineers at top companies now earn 15 to 25% more than engineering managers at the same level.

On September 25, Marty Cagan, who runs the product consultancy SVPG and has coached product teams for decades, published a short piece that says the trend is a mistake with an expiry date. He calls it a “necessary but temporary” state.

His account of how it happened is specific. AI changed how software gets built, and managers who didn’t learn the tools couldn’t judge the work. The good ones got hands-on to catch up. Some went back to building full time, and some were tired of managing and welcomed the exit. After a year inside the tools, he says, they feel pulled back toward product vision, strategy and how teams are organized. He also points to a Business Insider report that Meta has asked some AI employees to become managers again. I haven’t been able to check that report, so treat it as a lead.

His reason it comes back is the part worth sitting with. When building gets cheap, the expensive question becomes what to build and whether the people deciding know the craft. Leadership, he writes, becomes “only more important, not less.”

Here is what he is and isn’t arguing. His principle is “experts lead experts”: the people who lead engineering, product and design should have real expertise in that discipline. The target is what he calls the “people-only manager,” someone who runs a team without knowing its work. He calls that model “beyond terrible” and says he’ll explain why in a later piece.

That’s a narrower claim than “the cuts were wrong.” A coordination-only manager, someone whose job was timelines and status, loses value under both readings. Cagan and the compression call agree on that.

They split on one word: structurally. Either the flattening is a lasting change in how companies are built, or it is a cycle that corrects once people learn what the tools can’t do.

The evidence for the lasting version is mostly announcements, and announcements are intentions. A company saying it will flatten tells you its plan. It doesn’t tell you where it is two years later. The evidence for the cycle is one essay, one unchecked news report, and Cagan’s own admission that he might be wrong. Neither side has the number that would settle it.

I can name that number. Take the companies that cut managers in 2025 and 2026, and measure the ratio of managers to individual contributors again in 12 to 24 months. If it recovers, Cagan is right and the word “structural” was premature. If it stays down and the Meta reversal turns out to be a handful of people, the compression is real.

There’s a second consequence if he’s right. The strongest argument for domain expertise as the lasting advantage in AI work has been about the person directing the agents. His principle moves it up a layer. The scarce thing would be the leader who can tell good work from plausible work, and the manager who never could was never safe, AI or not.

I don’t know which way this goes. What I’d do in your place is check which kind of manager your last round of cuts removed. If it was coordination, the cut probably holds. If it was the person who understood the craft well enough to say no to a bad bet, look at how the decisions have been going since. That is where Cagan’s prediction would show up first, and nobody is tracking it for you.