{"slug": "are-companies-really-doing-layoffs-for-ai", "title": "Are Companies Really Doing Layoffs \"For AI\"?", "summary": "Companies are laying off significant portions of their workforce while publicly attributing the cuts to AI-driven productivity gains, a pattern observed at Amazon, Atlassian, Meta, and Block. GitHub research shows Copilot users complete tasks 55% faster, and Meta CEO Mark Zuckerberg has stated AI now writes roughly half of the company's code, providing operational justification for replacing some engineering roles. However, the \"layoffs for AI\" narrative also serves as a convenient public relations story that masks other common reasons for workforce reductions, such as post-pandemic over-hiring and revenue shortfalls.", "body_md": "**Amazon did it. Atlassian did it. Meta is reportedly doing it. Jack Dorsey set the tone by cutting half of Block. Your competitor may be thinking about it. Here's what's actually happening and why.**\n\nYou've read the press release by now. \"As we continue our AI transformation, we are making the difficult decision to reduce our workforce by X%.\" The specifics vary. The framing doesn't. Companies are laying off large chunks of their workforce and crediting, or blaming, AI.\n\nThis is not a coincidence. It's a playbook. And like many corporate playbooks, the press release explanation is only part of the story.\n\nHere's how the story gets told publicly: AI is so productive that you need fewer humans to do the same work. Every engineer with Copilot is worth 1.5 engineers. Every support agent with an LLM front-end handles twice the tickets. Every data analyst with an AI assistant runs twice as many reports. Therefore, headcount drops.\n\nThis is partially true -- AI as augmentation can greatly improve productivity. AI agents can legitimately do some real work with little human intervention. However, the \"layoffs for AI\" storyline is being used to do a lot of work it doesn't actually deserve.\n\nThe \"for AI\" framing is doing several things simultaneously. Two are real, to varying degrees, and one is theater.\n\nLet's start with one that's genuinely true. AI tools have made certain categories of knowledge work meaningfully more productive, and in some cases, genuinely replaceable.\n\n**Software development:** GitHub's own research showed Copilot users complete tasks 55% faster [1]. Meta's internal agents reportedly write a significant fraction of their code autonomously -- Zuckerberg has said publicly that AI now writes around half of Meta's code [2]. When he says he's replacing mid-level engineers with AI agents, that's not bluster. He has the internal data to back it up. That said, see my article (linked below) about what \"100% AI-written code\" actually means (TL;DR: Coding is only about 15% of producing software). There IS a period of investment and lower productivity (a \"J-curve\") to get to that level of productivity, but that's a bigger topic for another article.\n\n**Tier-one support:** If 60% of your support tickets follow five patterns, and an LLM can resolve four of them without human escalation, your support headcount math just changed. By a lot.\n\n**Data analysis and reporting:** The work of pulling data, writing queries, building slides, and summarizing findings used to require a full-time analyst per business unit. It doesn't anymore (or that analyst can provide a lot of additional value).\n\nThese aren't hypothetical. They're happening in production, today, at scale. If your organization has 200 engineers doing work that 120 could now do with AI tooling, that gap doesn't disappear by itself. Someone eventually looks at the gap and thinks restructuring, or slower hiring.\n\n**The painful reality:** for certain roles, the layoffs are operationally justified. Not politically comfortable. Not good for the humans involved. But arithmetically coherent.\n\nCompanies lay off employees for lots of reasons: revenue shortfalls, strategic pivots, bloat from over-hiring in 2021-2022, post-acquisition redundancy, competitive pressure. The accumulated cruft of the past business cycle are the 2026 reality for company management whether AI exists or not.\n\nBut \"we're cutting headcount because we over-hired during the zero-interest-rate bubble\" is a PR nightmare. \"We're cutting headcount because our Q3 revenue missed projections\" spooks investors. \"We're realigning our workforce for the AI era\" is... a growth story.\n\nSame outcome. Different investor reaction.\n\nMeta nearly doubled its headcount between 2019 and its 2022 peak of more than 86,000 employees [3]. The stock fell more than 60% as the zero-interest-rate era ended, and Zuckerberg responded with a \"Year of Efficiency\" that cut more than 21,000 jobs across 2022 and 2023. The transformation framing was real -- but was also a convenient way to declare victory on cleaning up structural bloat. Fast forward to March 2026: Meta is reportedly planning a second wave, cuts of up to 20% of its remaining workforce -- roughly 15,000 jobs -- explicitly to fund AI infrastructure investments projected to exceed $135 billion this year [4]. This round isn't framed as efficiency. It's framed as investment. Same mechanism, different story.\n\nAtlassian cut 10% of its workforce, about 1,600 jobs, and saw its stock rise on the announcement [5]. The announcement cited AI investments as requiring \"different skills.\" That's true. It's also true that Atlassian, like every enterprise software company, is navigating a competitive environment where cost discipline is rewarded and headcount-as-ambition is penalized.\n\nThen there's Block. Dorsey cut roughly 40% of the company's workforce -- about 4,000 positions -- and announced it alongside a Q4 2025 earnings report that showed gross profit up 24% year-over-year [6]. The memo was blunt: AI can do more, so we need fewer people. No hedging, no \"difficult decision\" language.\n\nHere's the context that memo omitted. Block had grown from 3,835 employees in 2019 to over 12,400 by late 2022 [7] -- a tripling in three years, driven by pandemic-era payment volumes, the Afterpay acquisition, and the same cheap-capital hiring binge that inflated every fintech balance sheet. Its stock fell more than 80% from its August 2021 peak. By 2024, the company had already trimmed about 12% of headcount through layoffs and attrition before the big cut.\n\nDorsey dropped the euphemisms. He didn't say \"difficult decision.\" He didn't say \"right-sizing for growth.\" He said AI replaces people and he's acting on it now. That's worth noting. But dropping the hedging language doesn't make the AI rationale more genuine -- it just makes it more quotable. Block over-hired during the zero-interest-rate era, spent years managing the consequences, and restructured when profits were strong enough to absorb it. Whether AI is the actual reason or the available vehicle, the cuts were coming. The memo just arrived without the usual stage dressing.\n\nCal Newport examined this dynamic directly on his Deep Questions podcast in March 2026 [8], concluding that Block's announcement fit the pattern of pandemic overhiring corrected for an investor audience -- not evidence of a genuine AI-driven mandate. Bloomberg ran a contemporaneous piece asking outright whether the announcement qualified as AI-washing [9]. Newport's broader argument: AI agents failed to materially displace knowledge workers in 2025, which means most of the headcount math being attributed to AI productivity is cover for structural decisions that would have happened anyway.\n\nMeta, Atlassian, and Block made the news cycles. The broader list is considerably longer. Major employers that announced AI-attributed workforce reductions in the 18 months ending early 2026:\n\n**The AI frame is the most investor-friendly restructuring rationale since \"right-sizing for growth.\" Expect every CFO in a public company to use it before 2027.**\n\nDoes this make the layoffs dishonest? Not necessarily...but there's no question that the kernel of truth makes the framing a more investor-palatable strategic choice.\n\nThe third thing happening is pure positioning, and it's arguably the least defensible.\n\nWhen a major company announces AI-driven layoffs, it signals to investors, partners, and the board that leadership understands the moment. \"We are not asleep. We see what's happening. We are taking action.\" The action itself is secondary. The signal is the point.\n\nThis creates a dangerous dynamic: companies that don't announce AI restructuring start looking like they're behind, when in fact they may just have experienced managed growth, or may be maintaining headcount and using (or planning to use) AI to grow or build faster than ever before. Even so, leadership teams may face board-level pressure to \"show something.\" The something that shows fastest is headcount reduction plus an AI narrative.\n\nSo do not be surprised to see companies announcing AI-related layoffs where the AI connection is, charitably, aspirational. In many cases, the roles being cut aren't being replaced by AI today. They're being cut because the company needs to demonstrate AI seriousness, and a restructuring announcement is the fastest way to do it.\n\n(Every CFO reading this just recognized a meeting they've been in.)\n\nThis is AI-washing: dressing operational decisions in AI language to access the narrative premium that comes with it. The cuts are real. The AI replacement often isn't, at least not yet. The expectation is that AI will eventually justify the math. Sometimes that's a reasonable bet. Sometimes it's a post-hoc rationalization with a better PR team.\n\nLet's run the numbers, because the narrative fails to capture them.\n\nA 50% individual productivity gain does not equal a 50% headcount reduction. The math is messier:\n\nThere is also a structural mismatch the productivity narrative tends to skip. AI agents excel at bounded tasks--but tasks are not jobs. Nate B. Jones frames the gap precisely: \"The average software job in America lasts somewhere between 18 months and two years. The average AI agent run lasts about two hours.\" [19] These are not interchangeable units. The institutional context that accumulates over a job tenure -- which system is actually prod, why that one client configuration is an exception, what went wrong in Q4 2023 and why -- doesn't transfer to an agent starting fresh each session. Early data is tracking with this: 55% of employers who made AI-driven cuts already regret them [20], per Forrester Research's Predictions 2026 report. Jones's argument for 2026: when execution costs drop and market opportunity expands, competitive advantage goes to companies that use AI to scale their ambition, not just their efficiency ratios. Cutting headcount while capacity is rising reveals misaligned strategy more than it reveals AI maturity [21].\n\nThe companies doing this well are the ones who cut carefully, retained their A-players, invested in AI enablement for the people who stayed, and were honest with themselves about where AI was actually ready versus where they were hoping it would be ready soon. These companies have a culture of training that doesn't stop after AI 101.\n\nThe companies doing it badly are the ones who cut the headcount first, set AI deployment targets second, and are now quietly rehiring in the same roles they just eliminated while trying to backfill the information on SPECIFICALLY what the cut roles actually did all day.\n\nFair question. And there is now some actual data.\n\nAnthropic published a labor market study in March 2026 [22] introducing what they call \"observed exposure\" -- a measure tracking not what AI could theoretically do in a given role, but what it is actually doing, based on real-world Claude usage data mapped against roughly 800 occupations. The distinction matters. Theoretical capability and real deployment are very different things.\n\nThe top-line numbers: Computer Programmers lead at 75% observed task coverage. Data Entry Keyers sit at 67%. Customer Service Representatives are around 65%. In Computer and Mathematical occupations, theoretical AI exposure runs at 94% -- but observed deployment is only 33%. Thirty percent of workers show zero measurable AI coverage today. And the study found no systematic increase in unemployment for highly exposed workers since late 2022.\n\nThe Financial Times's chief data reporter and labor correspondent identified the core tension in those findings [23]: the study measures task exposure within occupations without establishing whether task-level automation actually translates to job elimination. That gap is load-bearing. They also point out some real challenges in how the theoretical job replacement surface was calculated.\n\nTwo things the study doesn't answer that you'd need to answer to actually assess the overall level of economic risk:\n\nFirst, as the FT team and Nate B. Jones pointed out, **Teams don't disappear when tasks do -- and jobs don't necessarily disappear, either.** If 45% of a software engineer's tasks can now be AI-assisted, that doesn't mean you need 45% fewer software engineers. In most organizations, roles exist because teams need them -- for coordination, judgment, accountability, and continuity -- not because every hour is exactly filled with tasks. A partially-automatable job on a functioning team is still a job. The task-substitution framework the study uses doesn't model organizational structure at all.\n\nSecond, **Exposure rates without headcount data may be fully accurate for degree of occupational impact, but still tell us nothing about how many jobs are at risk in the economy.** Knowing that 60% of tasks in a given occupation can be AI-assisted tells you very little about how many people are actually at risk. A small, specialized field with 60% task exposure may affect far fewer workers than a large, low-margin occupation with 10% exposure...it all depends on the number of workers in each field. The study maps substitutability percentages by domain. It does not report how many people work in each domain, which means anyone trying to use it to project total displacement -- rather than relative exposure -- is doing their own math on an incomplete foundation.\n\nWhat the data does show: AI is genuinely shifting which tasks get done by humans in knowledge-worker roles, particularly in tech, finance, and administration. That signal is real, but the number of jobs eliminated where AI replacement is actually planned is still far smaller than the press cycle (and layoff reporting) implies.\n\nIf you're an executive watching peers or competitors announce AI restructuring, here are the four questions that actually matter:\n\n**1. Where is AI genuinely changing the unit economics in your business?** Not in theory. In practice, today, measurably. If you can't point to specific roles and specific productivity data, you don't have a restructuring case yet. You have a hypothesis.\n\n**2. What work is your workforce doing today that AI can do tomorrow?** The answer is specific, not general. \"AI will automate knowledge work\" is not an answer. \"Our tier-one support workflow resolves 62% of tickets against a pattern set that an LLM could match\" is an answer.\n\n**3. Are you cutting to invest or cutting to contract?** The companies getting this right are reducing headcount in roles where AI is genuinely productive and reinvesting in roles that make AI more effective: AI engineers, agent designers, workflow architects, the humans who know how to direct and verify AI output. The companies getting it wrong are using AI as cover for a pure cost reduction that leaves them less capable, not more so.\n\n**4. What is your re-skilling plan for the people who stay?** The mid-level engineer who can't (or won't) use AI tools effectively is not the engineer you want. The mid-level engineer who uses AI tools to do the work of three engineers is the one you want to keep and develop. The layoff is the easy part. The capability transformation is the hard part. Most announcements focus on the former and handwave the latter.\n\n\"For AI\" layoffs are real, strategic, and often disingenuous, in that order. Some of the headcount reductions are arithmetically justified by genuine AI productivity. Some are restructuring dressed in AI clothing. Some are merely signaling to investors and boards that leadership is awake (and not asleep at the switch).\n\nThe leaders who navigate this well will use AI productivity gains to do more, not just to cost-cut to the same output. The ones who use it as a rationalization for lazier restructuring will discover, eighteen months from now, that they cut the people who knew where the bodies were buried and never built the security and engineering frameworks around their AI tools to guard against and catch hallucinations or sycophantic agreement on inaccurate information.\n\nEither way, the press release will very likely say \"AI transformation\"--and the outcome will reveal whether it actually was.\n\n**If this resonated, here are some related articles:**\n\n*Keith MacKay is a technology strategy consultant and CTO in EY-Parthenon's Software Strategy Group (SSG), specializing in AI disruption and technology diligence for private equity and corporate clients. SSG's AI Disruption Lab conducts rapid assessments of how AI transforms and threatens existing business models and value chains. Keith teaches at Northeastern University and writes about strategy, management, and AI/technology with AI collaborators.*", "url": "https://wpnews.pro/news/are-companies-really-doing-layoffs-for-ai", "canonical_source": "https://dev.to/keithjmackay/are-companies-really-doing-layoffs-for-ai-18dl", "published_at": "2026-05-27 20:20:57+00:00", "updated_at": "2026-05-27 20:41:05.071066+00:00", "lang": "en", "topics": ["artificial-intelligence", "generative-ai", "ai-tools", "ai-agents", "ai-policy"], "entities": ["Amazon", "Atlassian", "Meta", "Jack Dorsey", "Block", "Copilot", "LLM"], "alternates": {"html": "https://wpnews.pro/news/are-companies-really-doing-layoffs-for-ai", "markdown": "https://wpnews.pro/news/are-companies-really-doing-layoffs-for-ai.md", "text": "https://wpnews.pro/news/are-companies-really-doing-layoffs-for-ai.txt", "jsonld": "https://wpnews.pro/news/are-companies-really-doing-layoffs-for-ai.jsonld"}}