The memo says AI. The headcount says the rest.

On its first-quarter earnings call on 8 May, Airbnb told investors that artificial intelligence now writes about 60% of the new code its engineers produce, roughly twice what the company estimates the industry manages. Twelve days later, Meta told 8,000 employees that artificial intelligence is the reason they are leaving. Both companies released a number with the word AI attached to it, and both numbers traveled well. But they are not the same kind of document. One is a figure a company is proud to volunteer; the other is a figure a company is required to disclose. When a corporation chooses what to announce, the choice is the story. Airbnb told you about the 60%. Meta had to tell you about the 8,000. The difference between those two is where the strategy actually lives.

The number Airbnb wanted you to hear

Sixty percent is a clean figure, and clean figures should be the first thing you distrust. Brian Chesky offered it on the call as evidence of leverage: "where you might have needed a team of 20 engineers before, an engineer can now spin up agents to do a lot of work under supervision." That sentence is built to be quoted, and it was. What it does not contain is the only quantity that would make it a strategy rather than a slogan: a headcount.

Decode the phrasing carefully. The claim is not that AI does 60% of Airbnb's engineering work. It is that AI wrote 60% of the code that engineers produced — "AI-coauthored," in the more careful versions. A senior engineer who accepts an autocomplete suggestion, then rewrites half of it, has produced AI-written code by this definition. The metric measures keystrokes, not judgment, and keystrokes were never the expensive part of software. The expensive part is deciding what to build and being accountable when it breaks. Airbnb has quietly told you as much: Chesky also conceded that "I do not think anyone has figured out AI for travel or e-commerce yet," and described the company's own AI assistant as still clumsy. The same call carried both claims. The 60% is the one that made the headline.

The 60% measures keystrokes, not judgment. Keystrokes were never the expensive part of software.

Now follow the action instead of the adjective. If AI had genuinely collapsed Airbnb's need for engineers by anything like the ratio Chesky's "20 versus 1" implies, the tell would be in the hiring. It is not. At the time of the call Airbnb was running well over 200 open roles, and an outside analysis of those postings found that 44% referenced AI or machine learning. A company that had automated away its engineering need does not advertise hundreds of jobs, nearly half of them in the very capability that supposedly made the jobs redundant. It hires harder. Airbnb also installed Ahmad Al-Dahle, who built Meta's Llama models, as its chief technology officer. You do not pay for that résumé to wind down an engineering organization. You pay for it to expand one.

So what is the 60% actually conceding? That Airbnb wants to be read as an AI-native company by a market that now prices the label, and that it has found a number flattering enough to earn the label without committing to anything the number would normally imply — no margin guidance tied to it, no engineering headcount cut to prove it. It is a strategic disclosure with the strategy left out. That is not a lie. It is positioning, and positioning is cheap precisely because it is free.

The number Meta had to disclose

Meta's announcement reads differently because it had to. On 23 April the company told staff it would cut roughly 8,000 jobs — about 10% of the workforce — with the reductions beginning 20 May and more flagged for the second half of the year. Alongside the cuts it canceled about 6,000 open roles, a combined contraction of some 14,000 positions, and reorganized survivors into AI "pods" with new titles: AI builder, AI pod lead, AI org lead. The internal memo, from HR chief Janelle Gale, supplied the official translation: the company is acting "to run the company more efficiently and to allow us to offset the other investments we're making."

Read that last clause as the honest part of an otherwise managed sentence. "Offset the other investments we're making" is a layoff explaining what it is paying for. The other investments are Meta's 2026 capital-expenditure plan for AI infrastructure: a stated $115 billion to $135 billion, up from $72 billion the year before. A layoff that funds a data-center build is not primarily a story about productivity. It is a story about reallocation. The eight thousand people are not being replaced by the AI they were reorganized around; they are being converted into the budget that buys the chips. That is the trade the memo describes once you stop reading it as a vision statement and start reading it as an accounting entry.

A layoff that funds a data-center build is not a productivity story. It is a reallocation, and the people are the line item being moved.

Meta is not alone, which is how you know it is structural rather than situational. Tech layoffs crossed roughly 142,000 in the first five months of 2026, a third higher than the same stretch of 2025, and the companies cutting hardest are not the ailing ones. They are the profitable ones — Meta, Amazon, Microsoft, Alphabet and Oracle among the largest, the same names that have collectively committed something on the order of $700 billion in 2026 capital spending. Snap trimmed about 1,000 roles, its chief executive citing AI's effect on "repetitive work." Cloudflare cut more than 1,100. Oracle's reductions ran to tens of thousands. The pattern is consistent enough to read as a sentence: profitable company posts record quarter, profitable company cuts workforce, profitable company points at AI. The first clause and the second clause used to be incompatible. They have been reconciled by a third, and the third is doing the same work "synergies" did a decade ago.

What "AI-first" is a euphemism for

"AI-first" is the phrase of the year, and like all good corporate phrases it is built to be heard one way and to mean another. Heard, it promises a company leaning into the future. Translated, it means: the AI build-out is now expensive enough that something else has to be cut to fund it, and the something else is people. That is not necessarily the wrong decision — concentrating resources behind a real bet can be exactly right. But it is a different decision than the announcement describes. The announcement describes ambition. The headcount describes the price the ambition required, and the company would much rather you read only the first document.

There is a quieter document still, the one nobody pins to the top of the memo: what got discontinued while everyone watched the launch. At Meta, the AI reorganization sits on top of more than a year of attrition in Reality Labs — the metaverse division that was, not long ago, the entire stated reason the company changed its name. The VR game studios were shut, hundreds of roles across the division removed, the moonshot demoted from destiny to cost center without a single press release announcing the demotion. The metaverse did not get canceled. It got quietly defunded so the next narrative could be capitalized. That is the most reliable tell in corporate strategy: the previous all-in bet does not get a farewell. It just stops appearing in the slides.

Put Airbnb and Meta side by side and the symmetry is the point. Airbnb volunteered a flattering AI number and kept hiring. Meta disclosed a painful one and called it AI. Same vocabulary, opposite obligations. The market currently rewards both moves, which is exactly why a company will make whichever one its situation requires and reach for the same three letters to explain it. The word is load-bearing for the share price and weightless as a fact.

Reading the two documents against each other

If you want to know what a company actually believes about AI, do not read the part it wrote to be quoted. Read the part it was obligated to file, and check whether the two agree. Where they disagree, the obligation is under oath and the boast is not.

• Follow the headcount, not the percentage. A real automation gain shows up as fewer engineers, not a bigger number in a slide. Airbnb's open roles are the rebuttal to Airbnb's own 60%.
• Read the layoff memo as the truest document the company published this quarter. "Offset the other investments we're making" tells you precisely what the cut is paying for.
• Match the bullish words against the bearish spend. A company that is genuinely confident in AI's near-term leverage does not need to fire 8,000 people to afford the leverage.
• Watch what stops getting mentioned. Reality Labs was the future until it was the cost center. The bet that disappears without a eulogy is the one the company has actually abandoned.

None of this requires assuming bad faith. AI is genuinely rewriting how software gets written and how large organizations are staffed; the leverage is real, even if it is smaller and lumpier than the round numbers suggest. The point is narrower and harder to spin: a company's announcement tells you what it wants to be true, and its disclosures tell you what it is willing to pay for. Airbnb wants to be an AI company and is paying to hire one. Meta is funding an AI build-out and is paying for it with payroll. The memos say AI in both cases. The headcount says which kind of AI company each one actually is — and the headcount, unlike the memo, had no choice but to be honest.