AI投資で組織図が変わる

Field Report — Tech LaborThe GreatFlattening.Capital that once funded headcount is now funding GPUs. The org chart underneath is changing shape — and the titles printed on business cards may be the last thing to go.81,747Tech workers laid off · Q1 2026$725B2026 AI capex, top 4 hyperscalers · +77% YoY275,000AI-related job postings open in the U.S.~50%Of Q1 cuts attributed to AI & automationIn the first quarter of 2026, roughly eighty thousand tech workers lost their jobs — the highest quarterly total the industry has posted in at least two years. In the same quarter, the four largest hyperscalers committed to a combined seven hundred and twenty-five billion dollars of AI infrastructure spend, a seventy-seven percent jump year over year. The numbers are not coincidental. The capital that funded engineering hires in 2021 is funding GPU clusters in 2026, and the executives signing the checks have stopped offering euphemisms about it.This is not a replay of the 2022 correction. That round was about reversing pandemic-era over-hiring under rising rates. This round is structural. Salesforce's CEO told investors he needs less heads. Meta framed its ten-percent cut as redirecting compensation budget toward AI research. Oracle eliminated up to thirty thousand legacy roles to fund data center expansion. The language has shifted from discipline to substitution.Underneath the headlines, the org chart is being redrawn. The traditional pyramid — CEO at the apex, three to four management layers descending to engineers in teams of six to eight — is being compressed from both ends. Middle management is the most visible casualty, but the cascading effect on titles, responsibilities, and career paths is more interesting than the headcount math suggests.01The Roles in the CrosshairsRisk is not evenly distributed. Five clusters are absorbing the bulk of the cuts, and the cluster you fall into matters far more than which company you work for.Customer support & CX92%Salesforce: 4,000 cut. The default tier-1 conversation is now an LLM; humans handle escalations only.QA & content moderation84%Pivoting toward AI eval engineering — but with substantial role compression.Middle management78%Gartner: one in five companies will eliminate over half of middle managers by year-end.Recruiting & HR ops72%Meta absorbed 35–40% cuts in these functions specifically.Entry-level engineers & generalist IT68%Stanford research shows entry coding roles already affected. IBM is the rare exception, tripling junior hires.Legacy infrastructure (on-prem DBA, sysadmin)65%Oracle eliminated up to 30,000 such roles to redirect funding to data centers.Senior ICs in AI/ML, infra, security+56%Wage premium attached to high-demand AI roles; 92% increase in postings YoY.The category at the bottom of that chart matters. Bloomberg estimates roughly half of AI-attributed layoffs result in the same role being rehired offshore or at lower bands within twelve months. Some of what looks like AI substitution is really geographic and salary repricing under an AI banner. The net effect is the same for the worker who lost the job, but it changes the prediction about whether the role itself disappears.02The Org Chart, Before and AfterTwo diagrams below. On the left, the shape most large tech orgs had in 2022: a five-layer pyramid where a director ran three managers, each managing seven engineers. On the right, the shape Amazon, Meta, and Intel are explicitly building toward — a compressed three-layer structure where one technical lead oversees fifteen-to-fifty contributors augmented by AI agents.Before · 2022The PyramidFive layers · 1:7 span · 32% of headcount in managementCEOVPVPVPDIRDIRDIRDIRDIRDIRMGRMGRMGRMGRMGRMGRMGRMGRMGRMGRMGRMGRL1L2L3L4L5×1×3×6×12×84Span of control · 1 : 7 Leadership / IC Management layerAfter · 2026 →The PlateauThree layers · 1:25 span · ICs paired with AI agentsCEOLEADLEADLEADLEADL1L2L3AI×1×4×100×∞“Three layers between intern and CEO.”— Mark Zuckerberg, Meta internal memo, 2025Span of control · 1 : 25 Human IC AI agent Lead / staff ICThe structural change is not subtle. Two full management layers — director-of-managers and many traditional engineering managers — get compressed into a single "lead" role that is part technical contributor, part coordinator, part talent buffer. Below the lead, the IC tier expands and pairs with persistent AI agents that handle code generation, ticket triage, scheduling, status reporting, and increasingly, autonomous task execution.Three things follow from this shape. Decisions move faster because there are fewer escalation steps. Coordination overhead lands on the lead who now runs twenty-five people instead of seven. And the rungs of the career ladder are missing — the traditional path of IC → senior IC → manager → director → VP has lost its middle steps.03Span of Control · The Hidden MetricThe clearest quantitative signal of the flattening is direct reports per manager. Gallup has tracked this annually since 2013. The trajectory is one-directional and accelerating.8.120139.4201810.9202412.12025~162026E~252028EAVG. DIRECT REPORTS PER U.S. MANAGER · GALLUP & AUTHOR PROJECTIONS · SOLID = ACTUAL · HATCHED = PROJECTEDManagers managing managers, managing managers — that whole thing is over. Either contribute directly or step aside.— The new orthodoxy, paraphrased from Meta & Amazon internal communications04Titles Are Changing Faster Than the WorkThe most underreported part of this story is in the titles themselves. The work has not vanished — most of it has been redistributed and renamed. The grid below maps roles that existed in volume in 2022 to what they are becoming in 2026.Engineering Manager (8 reports)Tech Lead / Player-Coach (15+)Codes 30–50% of the time. Coaching shifts from synchronous 1:1s to asynchronous review. Career development conversations become quarterly, not weekly.Senior DirectorFunctional Lead / GMTwo layers of reports collapse into one. The role becomes more strategic and less people-managerial — closer to a small business unit owner.Senior RecruiterTalent Partner / Pipeline EngineerSourcing, screening, and scheduling are AI-driven. The human role narrows to closing senior hires and managing the AI's prompt and policy layer.Customer Support RepCX Escalation SpecialistTier-1 vanishes. The remaining humans handle only what the bot escalates — meaning every call is harder and the cognitive load is higher per ticket.QA EngineerAI Eval EngineerLess manual test writing, more building and maintaining eval pipelines that probe AI outputs for regressions, hallucinations, and policy violations.Junior DeveloperApprentice / Often: EliminatedThe industry is bifurcating. Most companies are cutting the entry tier. A small minority — IBM most visibly — is tripling down on it as a future leadership pipeline.Project ManagerProgram Lead / Agent OrchestratorCoordinating humans and AI agents in mixed workflows. Less status-chasing, more designing the "operating system" of the team.Data AnalystInsight EngineerDashboards and recurring reports automate. The remaining work is hypothesis design, root-cause investigation, and instrumenting new questions for AI to answer.DBA / SysAdmin (on-prem)Platform / Reliability EngineerLegacy stack roles are being eliminated wholesale. The path forward is migration to managed cloud + AI-driven incident response.05The 2028 ProblemEvery flattening company is making the same trade — short-term margin for long-term capability. The savings are immediate and measurable. The cost has a delivery date that most boards are not yet asking about.Korn Ferry's 2025 research found that thirty-seven percent of employees at flattened organizations report feeling directionless. Deloitte found that only six percent of Gen Z aspires to senior leadership. The companies eliminating their middle managers in 2026 are simultaneously dismantling the apparatus that produces senior leaders in 2028. The bench gets thin in two directions at once: the people who would have grown into the role are gone, and the people who might have stepped into it have decided the role is not worth wanting.The most aggressive flatteners are betting they can hire externally when they need depth back. That bet rests on the assumption that the labor market will reproduce senior leaders without a development pipeline. It will not. The market for senior tech leaders in 2028 will be the empirical test of the choices being made right now.What replaces the lost development pipeline is the open question of the cycle. Some companies are experimenting with structured staff-IC tracks that absorb the leadership-development function without requiring formal management roles. Others are relying on rotational programs and external coaching. Many are doing nothing — and trusting that AI will continue to compress the work fast enough that the bench problem never quite arrives.That is the strongest version of the bullish case for the flattening: that AI tooling continues to absorb so much coordination, prioritization, and translation work that the leadership functions currently in crisis simply cease to be necessary in their current form. The weakest version is the one most companies are quietly betting on — that the problem is someone else's quarter.

※ 著作権に配慮し、引用は冒頭3段落までです。続きは元記事をご覧ください。