Most "AI layoffs" aren't AI layoffs. Four levers explain why.

TL;DR

The news cycle is full of AI-driven layoff stories. Most of them are not what they seem. A framework for decoding them: Lever 1 is labor efficiency (the smallest lever, only confirmed where work is finite, like customer support). Lever 2 is reskilling (the actual driver behind most announcements; companies are firing AND hiring in the same quarter). Lever 3 is org-shape changes (flatter orgs, fewer managers, project allocation over reporting hierarchy). Lever 4 is AI washing, where layoffs from 2021 overhire corrections are framed as AI-driven to satisfy a board narrative. Five of six major tech companies had net headcount growth across 2023-2025 despite layoff headlines. The headline is Gross Layoffs. The truth is a skill-mix shift.

The number that tells the wrong story

In 2025, 783 tech-company layoff events affected 245,953 workers. Challenger, Gray & Christmas attributed 48,414 US cuts directly to AI. Those are the numbers that make the headline.

Here are the numbers that don’t:

Five of six grew. The layoff number tells the wrong story about the workforce trajectory. A reader with only the layoff number cannot distinguish between genuine contraction, skill-mix swap, and performance management dressed as restructuring. A reader with gross layoffs + gross hires + net delta can.

Lever 1: Labor efficiency (the smallest lever)

This is what people think is happening everywhere. It’s only happening in one place: customer support.

Salesforce cut support from 9,000 to 5,000. Benioff said it directly: “I need less heads.” Klarna’s AI assistant handled 2.3 million conversations in its first month, equivalent to 700 full-time agents, with resolution time dropping from 11 minutes to under 2. IBM displaced approximately 200 HR roles with an internal AI agent that automates 94% of routine HR transactions. Shopify’s CEO told managers they must prove a task cannot be done with AI before requesting headcount.

The mechanic is straightforward: ticket resolution is unit-priced, volume is bounded, and an AI agent can resolve a known issue faster than a human can type the response.

But even the cleanest case isn’t as clean as the headline suggests. Klarna’s CEO reversed course 15 months later, publicly stating the company would “always have a human if you want.” The purely automated model produced lower-quality outputs. Gartner found that only 20% of customer-service leaders have actually reduced staffing due to AI, and predicts that half of the firms that cut CS staff will rehire by 2027.

Labor efficiency is real in support because the work is finite. For every other function, the work is not.

Lever 2: Reskilling (the real driver)

Mark Zuckerberg, January 2025: “We typically manage out people who aren’t meeting expectations over the course of a year, but now we’re going to do more extensive performance-based cuts during this cycle, with the intention of backfilling these roles in 2025.“

Backfill was the explicit plan. The story narrative was “Meta cuts 5%.”

Meta grew its headcount 6.5% that year. Microsoft laid off ~15,000 and grew R&D headcount 12.5%, with AI/ML/NLP/vision roles comprising 22%+ of new R&D hires in the last 18 months. Salesforce cut 4,000 support roles, hired 2,000 Agentforce sellers, and ended the year up 9%. The motion is not for fewer people. It’s different people.

Satya Nadella, November 2025: “We will grow our headcount, but that headcount we grow will grow with a lot more leverage than the headcount we had pre-AI.”

The reskilling lever is the largest. Most AI-era headcount announcements are a skill-mix swap with a net positive delta. The press release says “layoff.” The 10-K says “growth.”

Lever 3: Org-shape changes

Every AI-era org is flattening. Amazon CEO Andy Jassy announced a plan to increase the ratio of individual contributors to managers by at least 15% by the end of Q1 2025, with the explicit goal of “fewer managers, flatter organizations.” Morgan Stanley estimated this could eliminate approximately 13,834 manager roles. Citi’s Project Bora Bora cut from 13 management layers to 8 and eliminated 1,500 managerial roles, 13% of global leadership. Meta’s Zuckerberg went after “managers managing managers” explicitly, pushing the average direct-report ratio from 3-4 to a target of 7-8.

This lever is distinct from the other two. Nobody is being replaced by AI. The management layer is compressing because flatter orgs move faster, and the AI moment rewards speed. Middle management compresses, doers gain influence, and the hierarchy simplifies.

The planning implication: headcount is flat, but the composition shifts from managers to ICs. If your headcount system only tracks total headcount and not the skill-mix composition, you can’t see the reshape happening.

Lever 4: AI washing

The unspoken one. Some portion of 2024-2025 layoffs framed as “AI-driven” are actually 2021 overhire corrections that companies deferred, repackaged, and shipped under a board-friendly AI narrative.

Block is the clearest case. Jack Dorsey’s February 2026 earnings call announced a 40% RIF alongside “We expect to hire more senior AI engineering talent.” The company reported a 40% lift in production code shipped per engineer since September 2025, and announced a broad hiring freeze through 2026 with carve-outs for senior AI hires, sales, and marketing. Block had gone from ~13,000 to ~6,000 across 2024-2026. The AI framing made the cut legible to investors. The underlying motion was cost normalization from the 2021 surge.

AI washing isn’t malicious. It’s narrative management. But it means a CHRO reading the headlines cannot distinguish between “this company has a real AI-labor-substitution thesis” and “this company is correcting a hiring mistake from three years ago.” Both look the same from the press release.

The planning implication

If you’re a CEO, CFO, or CHRO trying to plan your own org’s posture, you need to read your own motions correctly before you can plan correctly.

Are you pulling Lever 1 (genuine labor efficiency, bound to support)? Lever 2 (reskilling, which means you’re hiring as fast as you’re cutting)? Lever 3 (flattening, which means the total stays flat but the composition shifts)? Or Lever 4 (correcting a prior overhire and framing it as AI)?

Each lever requires different infrastructure. Lever 2 requires scenario planning and redeployment tracking, running simultaneously with hiring. Lever 3 requires project-allocation visibility that your HRIS doesn’t provide. All four require the data to distinguish them, which means tracking gross layoffs, gross hires, and net delta by skill category in real time.

The org that can read its own levers correctly is the org that can plan correctly.

FAQ

Is AI actually replacing workers or not?

In customer support: partially yes, though even Klarna reversed course and Gartner predicts half of firms that cut CS staff will rehire by 2027. In engineering: no. Engineering backlog is infinite; productivity gains accelerate clearance but don’t reduce team size. Every major tech company’s engineering team is flat or growing through 2024-2026. In management: yes, but via org-shape change (Amazon’s 15% IC-to-manager ratio increase, Citi’s 13-to-8 layers), not AI substitution directly.

How do I know which lever my company is actually pulling?

Track three numbers for every restructure: gross layoffs, gross hires in the same period, and net headcount delta by function. If gross hires exceed gross layoffs, you’re reskilling. If the total is flat but the composition shifts, you’re flattening. If gross layoffs exceed hires and the company was overcapitalized in 2021, it’s probably a correction.

Why does this matter for headcount planning?

Because each lever produces a different planning need. Reskilling means your hiring plan and your separation plan run in parallel. Flattening means your org-design plan changes without your total headcount changing. If your system only tracks total headcount and not composition, you can’t distinguish the levers, which means you can’t plan for them.

Where did the net-headcount-delta data come from?

Public 10-K filings and Macrotrends headcount data for each company, reconciled against earnings-call transcripts and press releases. All sourced from 2023-2025 fiscal-year disclosures.