The Licensing Document, Named

Read this sentence: “We are implementing AI-driven workflow modernization, resulting in 15% efficiency improvements across the organization.”

Now read this one: “We are firing 15% of the company.”

Same announcement. Different words, different intent, under a different garb. The first is investor-ready, press-friendly, LinkedIn-shareable. The second is what actually happens. Somewhere between those two sentences, a translation occurs. The translation does not invent new facts. It simply selects the framing that requires the least external defense.

We are calling this pattern, throughout this piece, the Licensing Document. The practice of announcing decisions in a language that functions, in public, as a permission slip for the decision. Not a lie. Not exactly honesty either. A specific third thing. The corporate-communications layer of the AI era, where the actual decision gets dressed up in whatever wardrobe the current cultural moment is willing to reward.

This is the third piece in a series about a condition we’ve been calling The Great Blanding. The visible cultural and commercial beige-ification produced when AI is being used, at scale, as a cover for decisions people were already making. In the first piece, we named what the cost looks like at 11 PM in a specific kitchen. In the second piece, we prosecuted the specific organizational cowardice that produces those decisions. In this piece, we are doing the quieter, more uncomfortable work. We are naming the mechanism. The reason the cowardice works. The reason the cost compounds. The reason an executive who would have been fired ten years ago for making the same decision, in different words, is now being promoted. Glorified. Looked up to.

When an executive tells you a decision is AI-driven, listen carefully to which of the words got deleted. The real reason is underneath. The AI frame is the licensing document that makes the real reason quotable.

The rest of this piece is about how to read one.

What the Licensing Document Replaces

Here is a small catalogue of phrases that AI has, over the last two years, quietly displaced in corporate communications.

“Workforce reduction” has become “AI-driven workflow optimization.”

“Cost-cutting” has become “AI-enabled operational efficiency.”

“We don’t trust our middle management” has become “Flattening the organization to reflect AI-native workflows.”

“We want to pay less for the same output” has become “Leveraging AI to reduce friction in our value chain.”

“We have no strategy and we’re bet-hedging in public” has become “Investing across the AI stack to maintain optionality.”

In every case, a specific, publicly defensible story has replaced a specific, publicly indefensible one. The indefensible one is still underneath. It did not go anywhere. The executive making the decision still knows what the decision is. The employees affected by the decision still know what the decision is. The customers on the receiving end still know what the decision is. Only the announcement has changed. Only the audience the circus is performed for has been given a version of the story it is allowed to applaud for.

This is not a new observation. The New York Times named the phenomenon explicitly in early 2026, using the term AI-washing: the practice of citing AI as the reason for layoffs that, on inspection, have nothing structural to do with AI. A Forrester report from January 2026 put it plainly: “Many companies announcing A.I.-related layoffs do not have mature, vetted A.I.” They are not actually operating at a level of AI deployment that would justify the stated headcount reductions. The deployment is aspirational. The layoffs are real.

The numbers make the gap visible. Of the 1.2 million job cuts U.S. companies announced in 2025, AI was cited as a reason for just 55,000. About 4.5%. That is the stated attribution. The cultural narrative, meanwhile, the LinkedIn discourse, the CNBC panels, the McKinsey reports, the investor decks, treats AI as the dominant story of the year’s workforce decisions. The narrative is roughly an order of magnitude larger than the thing the narrative is describing. The AI frame is bigger than the AI itself.

Here is a diagnostic you can use for the rest of your career: the first test of whether a decision is AI-driven is to ask what the announcement would have said if AI had never been invented. If the answer is “we wouldn’t have made this announcement,” or “we would have said something more honest and less celebrated,” the AI frame is doing licensing work. You can see the seam between the decision and its costume. The seam is the story.

One more honest observation is worth pausing on. Research from the Oxford Internet Institute has suggested that companies are scapegoating AI to cover for business decisions that would otherwise reflect poorly on leadership. Previously, the institute noted, there was stigma attached to AI adoption. Now companies position themselves at the technology frontier to appear innovative and competitive. The cultural polarity has flipped. The same decision that would have been criticized in 2019 as “can’t we do better than just firing people to hit the quarter?” is, in 2026, praised as “AI-native operational discipline.” The decision didn’t change. The clapping audience changed.

A Story That Shows the Mechanism

Let us walk through one example, fully anonymized, close enough to the ground that you can feel the cracks.

At a technology company, the CEO decided that AI adoption would become a condition of continued employment. People who didn’t demonstrate sufficient enthusiasm about AI were either managed out, or quietly released. The stated logic was simple: the company needed to be AI-forward, and anyone who couldn’t or wouldn’t adopt was a drag on that mission. The actual logic was different. The actual logic was that the CEO wanted a smaller, cheaper team. He couldn’t say that. Not in 2025. Not in a market where layoffs-for-cost-reasons required apology, severance theater, and a round of defensive press coverage. So he reframed the decision. Adopt AI or you’re fired was not a performance standard. It was a license to cull.

The CEO’s stated conviction was that a single person, armed with AI tools, could replace a team of several people. The conviction was baseless. There was no pilot. No evidence. No honest test of the claim. But the conviction didn’t need evidence. The conviction needed an audience, and it had one. A board, a set of advisors, an investor group, a LinkedIn feed, all of whom were currently rewarding AI-forward posturing regardless of operational reality. The CEO’s conviction was, in 2025, socially endearing. It didn’t need to be true. It just needed to be stateable.

The firings happened. The survivors did not produce better work. They produced identical, dark-blue, ugly-as-fuck templates from tools that promised a design revolution and delivered wallpaper. The marketing department’s output increased, but the quality remained the same or in most cases, measurably worse. The executive got the cost savings. The customers got the worse product. The fired got the unemployment clock.

And here is the part that matters for this piece. No layoff tracker will ever record this company’s cuts as AI-related. The public story was that people weren’t a fit for the evolving organization. The private reality was that AI was the lever used to make the cuts possible without paying the reputational cost of admitting them. That is the licensing document, operating internally. The same mechanism we see in public announcements, turned inward, used to manage people rather than messaging.

We wrote in a previous piece that executives love to hide behind the phrase “you can’t measure brand,” because the un-measurability is strategically useful. It protects the budget cuts they already wanted to make. The licensing document is the same move, on a different domain. The un-auditability of AI adoption is strategically useful. It protects the workforce cuts executives already wanted to make. Same rhetoric. A different shroud for the cowardice. And, as we’ve argued elsewhere, the hardest decisions in any company are always the decisions about what you actually know versus what you’re willing to claim to know. The licensing document is how executives avoid that question. They don’t answer it. They make it unaskable.

The Zuckerberg Translation

Now the public version of the same move, at a larger scale.

In January 2025, Mark Zuckerberg announced that Meta would be laying off approximately 3,600 employees. 5% of its workforce. The stated reason, in an internal memo to staff: “I’ve decided to raise the bar on performance management and move out low performers faster.” The phrasing is worth reading twice.

Let us translate the sentence into the three possible framings Zuckerberg could have used, and understand why he chose the one he did.

Framing one: performance management. “We have identified 3,600 employees who are demonstrably failing to meet performance standards and are releasing them.” This framing is legally defensible. It is also, in practice, operationally expensive. It requires documented performance issues for 3,600 individuals, contemporaneous reviews, calibrated rankings, HR-approved processes, a justifiable distribution of underperformance across teams and levels. It is slow. It is defensible in court. It is reputationally acceptable. It is also, critically, falsifiable. A journalist or an aggrieved ex-employee can check whether the performance documentation was actually there.

Framing two: cost-cutting. “We need to reduce headcount by 3,600 for financial reasons in a year of intense competition.” This is the most honest framing. It is also the most reputationally catastrophic in a year where Meta reported approximately $47 billion in quarterly revenue and was, by any conventional measure, highly profitable. “We are cutting 3,600 people because we want the cost savings” is the kind of sentence that starts a bad news cycle. It is true. It is also publicly indigestible. An executive with functioning comms instincts does not say this.

Framing three: low performers. “I’ve decided to raise the bar on performance management and move out low performers faster.” This is the framing Zuckerberg chose. It has three properties the other two don’t. It is unfalsifiable. Nobody can check whether the 3,600 were actually the lowest-performing 5% in any rigorous sense, because the performance evaluation was conducted on an accelerated timeline specifically to enable the cuts. It is culturally congruent. The “high standards, no room for mediocrity” language plays beautifully in the current discourse. And it is personally elevating for the person saying it. Zuckerberg is positioned not as a cost-cutter but as a demanding leader making hard choices.

The 3,600 employees are equally unemployed in all three framings. The framing is not for them. The framing is for the market, the press, the other executives watching, and the board. Its job is to make the decision quotable in a form nobody in that audience can punish.

The cracks become visible almost immediately. The same Fortune piece reported in February 2025 that laid-off Meta employees were publicly sharing that they had received positive performance reviews before being cut under the “low performer” framing. The internal reality and the external announcement did not match. The announcement was a licensing document. It licensed a decision made for reasons the language did not describe.

Zuckerberg’s sentence is not literally false. Any mass layoff of 3,600 people will, statistically, contain some underperformers. The question is not whether the sentence is technically accurate. The question is whether the sentence is the story. It is not. The story is the decision. The sentence is the license.

A Moment to Remember What This Costs

We are going to pause the analysis for one paragraph, deliberately, because the licensing-document frame makes it easy to forget what is actually underneath.

Every one of these decisions, the CEO at the technology company using adopt-AI-or-you’re-fired as a culling lever, Zuckerberg’s performance-standards frame, every licensing document we’ve named so far and every one we’re about to, produces exactly the same downstream outcome. A specific person. In a specific kitchen. At a specific hour of the night. Doing math about rent and runway and whether the next round takes them. We wrote about that person in the first piece of this series. The analysis in this piece is about the mechanism. The cost is what the mechanism produces. We do not want the analytical distance to obscure the ground truth. The ground truth is a kitchen.

With that grounding restored, back to the mechanism.

Klarna, in Reverse

There is one more public case worth walking through, and it is the most instructive of all. Because it shows the licensing document operating in both directions. First to license the decision. Then, later, to license the reversal.

In 2023 and 2024, Klarna’s CEO Sebastian Siemiatkowski built a public narrative around the company’s aggressive AI adoption. The company’s OpenAI-powered chatbot, Siemiatkowski claimed, was doing the work of 700 customer service agents. The company froze hiring. The workforce shrank from 5,000 to 3,800 through what Siemiatkowski described as “natural attrition.” On X, in January 2025: “AI can already do all our jobs.” The posture was loud, confident, consistent. The narrative held straight through the company’s pre-IPO publicity cycle.

Then the admissions began. By May 2025, the same FinTech Weekly piece reported that Siemiatkowski had told Bloomberg something different. The AI-first customer service strategy had produced “lower quality” output. Klarna would begin recruiting human agents again. By October 2025, after the successful $19.65 billion US IPO, Siemiatkowski stated the quiet part plainly: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.” Rehiring. Flexible remote workforce model. Students, rural workers, loyal customers.

Look at the shape of the admissions. The 2023–24 posture was a licensing document for aggressive cost-cutting: “we are replacing 700 agents with AI because AI works.” The 2025 admissions are a licensing document for the reversal: “we went too far, efficiency over-indexed, quality suffered, we are now correcting.” Neither statement is a lie. Both statements are selective. Both statements are optimized for the audience they were performed for at the moment they were performed.

The most telling detail is the timing. The pre-IPO admissions were careful. The full admission, “we went too far,” came after the IPO, when the licensing-document function was no longer needed for the previous narrative and had to pivot to licensing the correction. Licensing documents are context-sensitive. They change shape when the incentives change shape. The same executive who used AI as a license to cut 700 people, used “lower quality” as a license to admit the cuts had been wrong. The mechanism doesn’t disappear when the decision reverses. The mechanism just finds new words.

Klarna is not an outlier in the AI-era admissions cycle. An IBM study of 2,000 CEOs published in May 2025 found that only 25% of AI initiatives had delivered expected ROI over recent years, and only 16% had scaled enterprise-wide. Most of the decisions being announced in licensing-document language are not, by the executives’ own admission, producing the results the language is claiming. Which means the language is serving a different function than describing results. That function is licensing.

The costume rotates. The body underneath doesn’t.

Why the Licensing Document Works in 2026

Why now? Why has AI specifically become the perfect licensing frame in this particular two-year window? Four conditions are doing the work, and all four have to be true at once.

Condition one: the audience actually wants to believe. Boards, analysts, the press, the LinkedIn algorithm, the consulting class. They want an AI story. They were told, across 2022 and 2023, that AI was the defining technology of the decade. By 2024, the narrative had hardened into a quasi-religious consensus: companies that are AI-forward win, companies that are not, lose. In that environment, when a company announces an AI-driven decision, the audience’s default assumption is that the company is doing something strategic. The licensing document works, in other words, because its primary audience is actively looking for reasons to accept it. The audience wants the circus. The executives are just performing that circus.

Condition two: the language is technically unfalsifiable. “AI-driven workflow optimization” doesn’t mean anything specific. Which means it cannot be disproven. The executive can define the terms post-hoc to match whatever the decision turned out to be. If the layoffs produce productivity gains, the AI-driven frame is validated. If the layoffs produce quality collapse, as at Klarna, the frame retreats into “we focused too much on efficiency.” If the layoffs produce neither, the frame stays in place indefinitely, because nobody is going to run a retrospective audit on a phrase that never made a testable claim. This is not an accident. This is the language’s primary design feature.

Condition three: the comparison set is currently broken. In a market where every competitor is announcing similar-sounding AI-driven decisions, nobody gets punished for making their own. Each company’s licensing document is validated by every other company’s licensing document. The collective effect is that the market has lost the ability to distinguish AI-driven strategic clarity from AI-framed organizational cowardice. Everyone sounds the same. This, incidentally, is The Great Blanding operating at the strategic-communications layer. The same homogenization that has flattened advertising, journalism, and visual design has now flattened earnings-call language. Same mechanism. Same cost.

Condition four: the alternative is reputational suicide. Consider what happens to the executive who announces, honestly, “we are cutting 15% of staff for cost reasons in a profitable year.” In 2026, this sentence is career-limiting. It invites media scrutiny, shareholder lawsuits, employee rage, and, most damningly, competitive disadvantage. Because every other executive in the market is framing their own identical decisions in AI-forward language, the honest executive looks uniquely exploitative by comparison. The incentive structure is actively hostile to honest framings. Executives are not stupid for choosing the licensed version. They are rational. The system has trained them.

There is one instructive counter-example. When ASML cut 1,700 jobs in January 2026, the CFO did not invoke AI at all. He described the cuts as reducing layers and letting engineers do their work. An analyst observing the announcement noted that this kind of honesty is rare, and that it “stands in sharp contrast to the firms wrapping their layoffs in AI branding.” That observation is the whole argument of this piece, in one sentence. Honesty, in 2026, is rare specifically because the licensing-document economy makes it expensive. ASML could afford it because it is the most critical chip-equipment company in the world. Its strategic position is so strong that the CFO can say what he wants. Most companies cannot. Most companies need the costume, and the circus.

When all four conditions hold, AI becomes the perfect licensing document. Wanted by the audience. Unfalsifiable in its language. Validated by peer-group consensus. And protected by the career-limiting cost of choosing any more honest alternative.

One thing we want to be explicit about, because a piece like this gets misread otherwise. We are not anti-AI. Real AI is doing real work in the world. AlphaFold is folding proteins. ML systems are helping radiologists catch cancers earlier. Engineers are shipping better code faster because of Copilot-class tools. Writers are thinking more clearly with Claude in the loop. The medicine is good. The science is good. The carefully-considered use of AI by people who know what good looks like, that is good too.

What this piece is prosecuting is the licensing layer. The corporate-communications surface where AI has been reduced to a phrase that makes cowardly decisions quotable. The technology is not the problem. The technology is the uniform. The problem is the body underneath.

What the Licensing Document Costs

The licensing document is not free. It compounds three costs, all of which are currently being paid by people who did not make the decisions.

The trust cost. Every licensing document erodes the credibility of every legitimate AI-driven decision. When the word AI means both “genuine transformation of how we operate” and “cover story for a political firing,” the word starts meaning nothing. The executives who are doing real AI work, and some are, find their announcements drowned in the noise of executives who are using AI language to launder unrelated decisions. The signal-to-noise ratio collapses. Trust is a depleting resource. The licensing economy is depleting it faster than any single company can replenish it.

The aggregate-labor cost. Decisions announced in licensing-document language rarely get audited for honesty. The people who pay the price, the 3,600 at Meta, the 700 at Klarna, the marketing team at the technology company we described, the executive’s overruled designers, are dispersed, unorganized, and individually powerless. The aggregate cost is borne by an aggregate that doesn’t have a voice. The same Quartz analysis that covered the ASML announcement also reported on Mercer’s Global Talent Trends 2026 findings: employee concerns about AI-related job loss jumped from 28% in 2024 to 40% in 2026, with 62% of employees feeling their leaders underestimate the emotional and psychological impact of AI on the workforce. The leaders are not underestimating. The leaders are licensing. The gap between the workforce’s anxiety and the executive suite’s language is the gap the licensing document is specifically designed to maintain.

The strategic-clarity cost. This is the worst of the three, because it damages the very organizations making the decisions. Companies that hide cost-cutting inside AI-restructuring language lose the internal clarity about what they are actually doing. The board doesn’t know. The remaining employees don’t know. Increasingly, the executives themselves don’t know, because they have been using the licensing language with themselves for so long that they have started to believe the licensing language is the strategy. A company that cannot describe itself honestly to itself cannot make the next decision honestly either. The Great Blanding is, at root, an epistemic crisis. A generation of executives is forgetting the difference between the decisions they are making and the stories they are telling about those decisions. That forgetting is the cost that compounds longest.

How to Read a Licensing Document

We want to leave you with a diagnostic you can use for the rest of your career. Four questions. Ask them of any AI-driven announcement you encounter.

One. What decision is being announced? Ignore the frame. Describe the outcome in plain language. Who was fired? What was cut? What was changed? Strip the AI vocabulary and translate the sentence into the version your grandmother would understand.

Two. Would this announcement have been made if AI didn’t exist? If the answer is no, if the same decision would have been announced in different language, or, more damningly, not announced at all, the AI frame is doing licensing work. The announcement exists because AI is culturally protected. The decision exists for other reasons.

Three. Who is the announcement for? If the primary audience is the market, the press, or the investor class, not the customers, not the remaining employees, not the people affected, the announcement is a licensing document. Announcements that serve their subjects rarely need licensing. Announcements that serve the announcer always do.

Four. What word got deleted? The real reason is underneath. The AI frame is the replacement. Name the replacement. Recover the original. Practice the translation in your head until it becomes automatic. “AI-driven workflow optimization” is “layoffs.” “Leveraging automation to focus on our highest-value work” is “we couldn’t afford the team we had.” “Reallocating toward automation-ready functions” is “we have no idea what we’re doing, so we’re telling the market we’re being strategic while we figure it out.” Every licensing document has a translation. Your job is to learn to read it.

These questions don’t make you cynical. They make you literate. The AI era is producing a lot of announcements. Most of them are licensing documents. A small number are real. The difference matters, and you are now equipped to tell them apart.

In the next piece in this series, we prosecute the grift class that profits from the licensing-document economy. The influencers, the X-is-dead merchant monkeys, the consultants who monetize the anxiety the licensing documents produce. And then, in the final piece, we sit down with you. Not above you. Not beside you. With you. And talk about what is actually worth defending in the middle of all of this.

Sources and References

On AI-washing and the gap between stated and actual AI-driven layoffs. The New York Times’s February 2026 reporting and Forrester’s January 2026 impact forecast, both linked inline. TechCrunch, February 2026 — “AI layoffs or ‘AI-washing’?” Fortune’s February 2026 reporting on the gap between announced AI-driven layoffs and actual AI deployment, linked inline. Quartz’s coverage of the ASML counter-example and the Mercer 2026 Global Talent Trends findings, linked inline. Oxford Internet Institute’s analysis on AI-as-cover-for-layoffs via industry summary, linked inline.

On the Meta layoffs and the Zuckerberg performance-management framing. Fortune’s February 2025 reporting on the layoffs and the glowing-reviews discrepancy, linked inline. Fortune’s March 2026 retrospective on the broader arc of Meta’s cuts since 2022, linked inline.

On the Klarna AI reversal and the quality admission. FinTech Weekly’s May 2025 reporting on the Bloomberg admission, linked inline. Mind The Product’s May 2025 piece on the IBM CEO study showing 25% AI-ROI realization, linked inline.

From the Methodborne archive. The Violence of Hype and the Slow Invisibling, part one of this series, linked inline. The Great Industrial Cowardice, part two of this series, linked inline. Brand Is Measurable. And It Shows Up In Your Price Tag, linked inline. Why Most Early-Stage Startups Get Brand Strategy Wrong, linked inline.