I Watched 3 Companies Lay Off Their Managers. All 3 Hit the Same Wall.
TL;DR
Management isn’t one job — it’s three bundles: routing, sensemaking, and accountability — Nate argues companies are flattening org charts as if all management is just information logistics, when the harder human parts are actually signal extraction and coaching.
AI can already eat routing, but it still struggles with judgment-heavy sensemaking and long-horizon accountability — he calls routing basically “solved,” while saying even 10x better models would still be partners rather than reliable replacements for managers making hard calls.
Kimmy shows the upside and cost of radical flatness — Moonshot AI’s 300-person, $16 billion company runs with no titles, no OKRs, no KPIs, and five co-founders doing sensemaking for roughly 50 people each, producing extraordinary speed alongside anxiety, drift, and employees crying in meetings.
Block is making the cleanest structural bet: split the manager role on purpose — Jack Dorsey’s model uses an AI “world model” for routing, rotating DRIs for problem-specific sensemaking, and “player coaches” for human feedback and growth, instead of asking one middle manager to do everything.
Meta’s approach is the opposite: compress management rather than decompose it — Mark Zuckerberg widened spans, cut layers, and intensified performance pressure, with 25-30 senior leaders reporting to him and the bottom 5% cut, trading fewer managers for sharper accountability and likely more burnout.
The real AI adoption challenge is a management design problem, not just a tooling problem — Nate’s core claim is that leaders need to reimagine what managers actually do before cutting them, because the best predictor of whether people thrive at work is still whether they have a good manager.
The Breakdown
The layoff trend everyone can feel
Nate opens with a blunt observation: nearly half of U.S. companies have removed management layers in the last year, usually under the banner of getting “flatter, leaner, faster” with AI. His setup is that a lot of companies removed something load-bearing, then expected everyone else to improvise.
The three jobs managers were doing all along
He breaks management into three functions. First is routing — plain information logistics, the Roman-centurion-to-railway-manager job of getting the right update to the right person at the right time. Second is sensemaking, where good managers filter noise, surface signal, and explain what actually matters; third is accountability and feedback, the very human work of ownership, coaching, and performance.
What AI can take, and what still looks stubbornly human
Routing, in his view, is basically done: AI is good at synthesizing updates upward and distributing decisions downward. Sensemaking is different because it depends on business history, product context, and the kind of “peel back the onion” conversations agents still don’t handle well; accountability is harder still because ownership isn’t just task completion, it’s the long-running feeling that something is your baby and you are on the hook for it.
Kimmy: radical flatness, radical speed, real psychological cost
The first case is Moonshot AI, maker of Kimmy K2, where a Renwu reporter spent about 100 hours embedded with the company. At this 300-person, roughly $16 billion startup, there are no formal departments, no titles, no OKRs, no KPIs; agents route information, five co-founders do the company’s sensemaking, and accountability is mostly self-imposed. Nate’s most vivid example is a PM launching three agents to process 3,000 user feedback items, multilingual sentiment, and competitor monitoring, getting a requirements doc and 70% of implementation by 11:30 a.m. — but the tradeoff is “weightlessness,” anxiety, isolation, and employees crying when projects miss.
Block: Jack Dorsey tries to split the bundle cleanly
Block’s model starts from the same historical diagnosis but takes a different route. Jack Dorsey and Roelof Botha propose an AI “world model” for routing, rotating DRIs — directly responsible individuals — to own cross-functional problems like merchant churn for 90 days, and “player coaches” who still build but also handle human growth and feedback. Nate calls this the sharpest structural innovation of the three because it explicitly preserves a human accountability loop while preventing sensemaking from calcifying into permanent middle management.
Meta: fewer layers, wider spans, harsher consequences
Meta is the biggest and least ideological example: not decomposing management so much as compressing it. Zuckerberg now directly oversees 25-30 senior leaders, 15 product divisions sit under just three executives, and managers either take much wider spans or move into IC roles, while accountability gets sharper with bottom-5% cuts. Nate says the stock-market verdict has been strong — roughly 3x growth since the efficiency push — but the internal anecdote is relentless pressure and open questions about burnout and retention.
The real shift is shorter loops, not just fewer bosses
His broader point is that AI changes management because it compresses the loop from market signal to shipped product. In a world where AI-native teams can go from customer feedback to production code in hours, old approval chains stop making sense, and the real question becomes how much autonomy people under a manager can handle when agents amplify their reach.
The practical advice for managers, leaders, and ICs
Nate closes by turning the framework into a playbook. If your role is mostly routing, show your value in sensemaking, coaching, and IC work; if you’re a leader, decompose manager work before you cut it; and if you’re automating, start with routing first. For ICs, his ending is simple and surprisingly human: the single biggest predictor of thriving at work is your relationship with your manager, so finding a good one is still worth its weight in gold.