Tame Your CEO Before Is Too Late

Your CEO doesn't have an AI strategy. Your CEO has whatever the last Anthropic or OpenAI keynote told them to have.

This isn't an insult. It's a structural observation. The same way an influential account on X shifts opinion for thousands of followers who never read the original source, a single Zuckerberg memo or Jassy directive rewrites roadmaps across entire industries overnight. Not because the memo contains superior analysis. Because it contains permission. Permission to do what everyone was already considering but nobody wanted to go first.

The Playbook That Writes Itself

In May 2026, Meta laid off approximately 8,000 employees, roughly 10% of its workforce. That same year, Meta's capital expenditure guidance for AI infrastructure reached $125-145 billion. That's a 270% increase over 2024's $39.2 billion. Zuckerberg warned employees that in the AI race, "success isn't a given."

Amazon cut 30,000 corporate workers between October 2025 and January 2026, its largest layoff in company history, during a quarter when revenue grew 21%. AI infrastructure spending for 2026: $200 billion.

Microsoft cut over 15,000 employees in 2025. CEO Satya Nadella called it the "enigma of success": record revenues alongside thousands of layoffs. AI infrastructure spend that fiscal year: $88.7 billion, on track for $190 billion in calendar year 2026.

Google continued steady headcount trimming across 2024-2026 while planning $175-185 billion in AI capital expenditure, and raised $80 billion in equity specifically to fund it.

The pattern:

These aren't independent strategic decisions. They're synchronized responses to the same pitch deck.

The Copycats

When Meta fires 10% and triples its AI spend, it doesn't stay a Meta story. It becomes "industry best practice." And the cascade is already underway.

Bernstein analyst Mark Shmulik warned clients that if Meta succeeds in "redrawing the blueprint for an AI-enabled organization," others will rush to replicate it, triggering "a cascade of hurried pivots, half-formed strategies, and reactive restructuring across the ecosystem."

Jack Dorsey cut approximately 40% of Block's 4,000-person workforce and bragged that AI-powered smaller teams enabled "a new way of working." His prediction to investors: "Within a year, most companies would reach the same conclusion."

Shopify CEO Tobi Lutke posted an internal memo requiring employees to "prove why they cannot get what they want done using AI before asking for more headcount." Eight months later, the rest of the industry adopted the same metric.

Marc Benioff announced Salesforce would hire no more software engineers in 2025, citing AI productivity gains.

The contagion spread beyond tech. Nike cut 1,400 roles. TCS announced plans to cut approximately 12,000 jobs targeting middle and senior management. In the first five months of 2026 alone, 115,430 people from 152 tech companies were laid off, nearly matching all of 2025 in half the time.

And here is the part that should concern you: SHRM published analysis asking whether midmarket CEOs are "redesigning for the future or just following a fashion that Fortune 50 companies can afford to botch and survive." Their data shows fewer than one in sixteen jobs is truly automatable in both technical and practical terms.

The CEO who fires 30% of their team because Zuckerberg fired 30% of his, without having Zuckerberg's revenue to absorb the risk, Zuckerberg's engineering bench to fall back on, or Zuckerberg's infrastructure to leverage, isn't making a strategic decision. They're following a trend. The same way someone on X retweets a take they haven't thought through because the account that posted it has a blue checkmark and 500K followers.

Box founder Aaron Levie diagnosed it precisely: "CEOs are uniquely prone to AI psychosis because they're sufficiently distant from the last mile of work that still has to happen to generate most value with AI."

MIT Professor Paul Osterman went further: "AI is a perfect excuse to justify big layoffs. It makes it seem as if it's not our decision, our fault. It's the technology."

Even Sam Altman acknowledged that some companies are "AI washing," blaming unrelated layoffs on the technology.

The numbers confirm the disconnect. PwC surveyed 4,454 CEOs across 95 countries: 56% report getting nothing from their AI investments. McKinsey found 88% adoption but only 6% qualifying as "high performers." Gartner found that 80% of large enterprises reduced headcount after automation projects, with zero correlation between layoffs and ROI. Companies reporting significant returns laid off workers at the same pace as those with negative returns.

They're not optimizing. They're imitating.

The Door That Closed

The AI narrative didn't just affect headcount decisions. It collapsed the first rung of the career ladder.

Stanford's Digital Economy Lab published the most rigorous study on the subject in August 2025, using high-frequency payroll records from ADP. The findings: employment for workers aged 22-25 in AI-exposed occupations fell 6% between late 2022 and July 2025. Meanwhile, employment for workers 30 and older grew between 6% and 13% in the same job categories. When controlling for other factors, a 16% relative decline in early-career employment, starting specifically in 2024, not earlier.

Indeed's Hiring Lab confirmed the shape of the collapse: junior tech titles down 34% versus senior titles down 19%. The share of postings requiring five or more years of experience rose from 37% to 42%.

Revelio Labs found entry-level U.S. postings down 35% since January 2023. But the most revealing data point is the gap between appearance and reality: entry-level job postings increased by around 47% between late 2023 and late 2024, while actual hiring into those levels fell by an estimated 73%. Companies advertise junior-sounding roles and fill them with experienced engineers who cost more but produce faster with AI tools.

The math is simple. A senior developer with an AI copilot is now more productive than a senior developer plus a junior developer. The tasks that juniors used to own (fixing bugs, writing tests, generating boilerplate) are precisely the tasks AI handles well. And when a survey of hiring managers finds that 70% believe AI can do the jobs of interns and 57% trust AI's output more than the work of recent graduates, the economic conclusion writes itself.

Computer science graduates now face 6.1% unemployment versus 3.6% overall, nearly double the general rate. For a degree that was supposed to be the safest bet in the economy.

The pipeline problem is the part that nobody wants to calculate. 58% of enterprises already worry that cutting entry-level roles will cause senior talent shortages within five years. They know the ladder is broken. They're breaking it anyway, because the quarterly incentive to cut costs now outweighs the five-year risk of having nobody qualified to promote.

Gartner predicts that half of companies that cut customer service staff for AI will need to rehire by 2027. The correction is already baked in. But by then, the people who would have been gaining experience in those roles will have gone somewhere else. Or nowhere at all.

Knowledge Debt

Technical debt is a concept every engineer understands: shortcuts taken today that compound into expensive problems tomorrow. Knowledge debt is the same mechanism applied to humans. It's the institutional understanding that walks out the door with every layoff and never comes back.

When a company lays off the person who knows why that system was built the way it was, they don't lose a line item on a spreadsheet. They lose context. The kind of context that no documentation captures because nobody thought to write it down. It was just something Sarah knew because she was there when the decision was made in 2019, and she remembered the three alternatives they tried first and why they failed.

Unlike technical debt, knowledge debt compounds silently. You don't know what you lost until you need it. And by then, the person who had it is working somewhere else or has left the industry entirely.

The replacement is a token quota. Instead of a person who understands the history, the constraints, the political dynamics, and the unwritten rules, you have an LLM that can generate plausible-sounding answers about any of those things. Answers that are statistically likely but contextually wrong. The AI doesn't know that the client hates email follow-ups because of an incident in 2021. The AI doesn't know that the legacy API has an undocumented rate limit that only triggers on the third Tuesday of months with 31 days. The AI gives you the general case. Institutional knowledge is the specific case.

We already have case studies. Boeing outsourced 70% of its design and engineering work and lost what insiders called "tribal knowledge that can never be recovered." The result was fuselage defects, shimming problems, and safety failures that cost billions and killed people. Klarna replaced 700 customer service workers with AI in 2023. By mid-2025, customer satisfaction had dropped, engineers were pulled to work phones, and CEO Sebastian Siemiatkowski admitted: "We focused too much on efficiency and cost."

The pattern is so consistent it has a name. Forrester Research calls it the "AI Boomerang": 55% of employers now regret AI-driven workforce reductions. Two-thirds are rehiring. 73% failed to come out financially ahead. One-third spent more on restaffing than they saved on the original cuts.

And here's the compounding spiral: fire the juniors -> the seniors who remain now have no one to mentor, no one to transfer knowledge to -> when those seniors leave (and they will; burnout after layoffs is well-documented), the knowledge leaves with them -> now you have neither junior talent nor institutional knowledge -> your AI dependency becomes not a choice but a structural necessity.

2025 research confirms the fundamental limitation: AI serves as an "epistemic partner" that augments human interpretation, but cannot replace tacit judgment. Rules-based AI is "constrained by its inability to adapt to changing situations." Knowledge, for humans, is situated and provisional. For AI, it's universal and static. The difference is precisely the thing that makes institutional knowledge valuable.

Companies are trading deep human understanding for shallow AI inference. And the irony is recursive: the more you depend on AI, the less institutional knowledge you retain, which means the less capable you become of evaluating whether AI is giving you the right answer. You lose the ability to know what you don't know.

That's not efficiency. That's organizational amnesia sold as innovation.

The Bill Nobody Reads

Every AI query has a physical cost that the narrative conveniently omits.

Sam Altman said in June 2025 that a single ChatGPT query consumes about 0.34 watt-hours, "what an oven would use in a little over one second." He got defensive. He said "it takes a lot of energy to train a human." What he didn't contextualize is scale. Epoch AI estimates that a GPT-4o query consumes 0.42 Wh, Claude 3.7 Sonnet consumes 0.84 Wh, and GPT-5 in extended reasoning mode can consume up to 40 Wh, over 130 times a standard Google search. At 2.5 billion messages per day, ChatGPT alone consumes approximately 850 MWh daily.

The water footprint is equally striking. AI data centers consumed 264 billion gallons of water in 2025, equivalent to the annual usage of 1.8 million Americans. Google's single Iowa data center: 1 billion gallons in 2024. One Meta data center consumes 10% of its county's water supply. A peer-reviewed study at NeurIPS 2024 estimated that using GPT-4 to write a single email costs approximately 3 liters of fresh water.

The grid can't keep up. PJM Interconnection, which manages the electrical grid for 65 million Americans, projects a 6-gigawatt shortfall by 2027 driven by data center demand. Capacity market prices jumped 10x in a single year, from $28.92 to $329.17 per megawatt. The IEA estimates 20% of planned data center projects are at risk of delays due to power constraints.

Now scale that to every company in the world routing Slack messages, email drafts, code reviews, meeting summaries, and performance evaluations through LLM endpoints. The combined AI infrastructure spending of Meta, Amazon, Google, and Microsoft is approaching $700 billion for 2026 alone. That's not R&D. That's operational cost (electricity, water, cooling, silicon) consumed to generate outputs that include, among other things, an AI-written Slack message that says "sounds good!"

The sustainability question the C-suite doesn't ask is the one the AI company narrative doesn't include: what is the environmental cost per unit of actual value created? Not per query. Per unit of value. Because most queries don't create value. They create convenience. And MIT researchers note that many AI queries "can be answered just as well through a search engine, a book, email or phone call," at a fraction of the energy cost. Convenience at planetary scale has a price that shows up on someone else's bill: the electrical grid, the water table, the communities downstream from data centers competing for the same resources.

None of this appears in the pitch deck your CEO reviewed before approving the "AI transformation initiative." The narrative is about productivity gains, competitive advantage, and being left behind. The bill arrives later, addressed to nobody in particular.

When Fear Became Content

There was a window, somewhere around 2023-2024, when "AI is going to take my job" was said with genuine anxiety. People were scared. The fear was reasonable and warranted. It still is.

But something shifted. The fear became a format. "A $20/month subscription replaced me" became a TikTok template. "I lost my job to a chatbot" became a punchline. Gen Z repurposed "clanker", a Star Wars term, as shorthand for AI systems displacing workers, filming themselves yelling it at delivery bots. The hashtag gathered millions of views in weeks. "Prompt engineer" exists simultaneously as a $300K job title and a meme that evolved into "vibe coding": building applications through natural language without understanding what the code does. Junior developers interviewed in 2025 could build anything with AI but couldn't explain a for loop. The displacement didn't stop. The emotional register changed.

The numbers show the normalization working. The share of Americans fearing AI will reduce their jobs dropped from 48% in 2024 to 31% in 2025. Not because the threat diminished (the layoffs accelerated). Because the dominant meme format, "AI will take our jobs / meanwhile AI: [absurd failure]," functions as both coping mechanism and normalizer. Package the fear as content, and the fear becomes entertainment.

This is what normalization looks like. Not acceptance. Normalization. The structural violence of replacing human labor with token generation gets processed through the same content machine that processes everything else: package it, add a hook, make it shareable. The algorithm doesn't distinguish between someone documenting their job loss and someone performing their job loss for engagement. Both get views. Both get served.

The danger is precise: when a society jokes about its own displacement, the pressure for policy action evaporates. Fear generates political energy. Memes generate clicks. You can't build a labor movement on a format that requires a punchline. You can't organize collective bargaining when the dominant narrative frame is ironic detachment.

Meanwhile, the lobbying machine fills the vacuum that humor creates. Over 450 organizations lobbied on AI policy in 2025, up from six in 2016. Meta launched a Super PAC. Andreessen Horowitz and OpenAI invested $100 million in another. Tech firms lobbied for a 10-year ban on state AI regulation. The EU pushed back AI Act compliance deadlines. While workers were turning displacement into content, the companies doing the displacing were turning that distraction into legislative cover.

And the connection comes full circle. The same platforms that amplify CEO herd mentality, where one Zuckerberg post sets the tone for a thousand LinkedIn hot takes about "AI-first organizations," also convert worker fear into engagement metrics. Meta earned $164.5 billion in AI-targeted ad revenue in 2024. Research shows Facebook's algorithm promoted misinformation at rates 64% higher than factual content. The same companies building AI that displaces workers profit from the engagement generated by fear of that displacement.

The people selling the narrative and the people being harmed by it are producing content for the same feed. That's not a metaphor. That's the business model.

What Follows

I don't have a solution. If I did, I'd be selling it on a platform that proves the problem.

But I have a diagnosis. The AI narrative, the one that flows from OpenAI keynotes to Anthropic blog posts to Zuckerberg memos to your CEO's quarterly strategy deck, is not a strategy. It's a supply chain. The AI companies produce the conviction. Big tech validates it with billion-dollar bets. Mid-market CEOs consume it and execute it without the infrastructure, the cash reserves, or the institutional knowledge to absorb the consequences.

The result is cascading knowledge debt across entire industries. Not because AI is inherently destructive, but because the decision-makers adopting it aren't forming independent opinions about it. They're following the narrative the way people follow opinion shifts on X: reactively, competitively, without reading the original source.

Dario Amodei himself said it at Dreamforce: "I must be honest: real replacement will start to occur in two to five years." And then, in the same breath, he suggested that taxing AI companies like his own "may need to happen eventually." The company selling the replacement acknowledges the damage and proposes taxing itself later. That's not a plan. That's a deferral.

The window for course correction isn't closing because AI is unstoppable. It's closing because the herd is already running. And once every mid-market CEO has restructured their org around the same pitch deck, the cost of admitting it was premature becomes higher than the cost of continuing. That's how narratives become infrastructure. That's how temporary conviction becomes permanent dependency.

The question for the person reading this, probably a mid-level manager who sees the pattern but doesn't control the budget, is not whether you can stop it. You can't. The question is whether you can make your CEO read the bill before they sign it. The full bill. Not just the line item that says "productivity." The one that includes the knowledge that walked out the door, the juniors who never got hired, the pipeline that will be empty in five years, and the kilowatt-hours burning in a data center so an AI can draft an email that someone will rewrite anyway.

Tame your CEO before it's too late. Or at least make sure someone in the room has read something other than the pitch deck.

Want to discuss where this is heading? Let's talk.