Through Kuhn’s Lens

The Frame War

June 21, 2026 | 2646 words

Through Kuhn’s Lens: Who Gets to Name the Thing?

This week the Frame War produced one of its cleaner collisions. On one side, a chorus of vendor announcements and executive interviews insisted that AI had crossed into “partnership” — that the systems now arriving are less tools than colleagues, agents that act on your behalf, collaborators with whom you “work.” On the other side, a quieter but persistent countercurrent kept returning to the older word: tool. Useful, fallible, owned, switched off at will. And running underneath both, a third narrative pressed its claim — that none of this vocabulary matters because what is underway is a “transformation,” a force of historical scale that will rearrange society whether we name it or not.

Four words, then, contend for the right to say what AI is: tool, threat, partner, transformation. The contest is not academic. Whoever wins the naming wins the right to set the terms on which the public meets the technology — what it expects, what it fears, what it forgives, what it demands. This is the phenomenon worth examining: not the systems themselves, but the struggle over the frame through which a public reads them.

Kuhn’s history of science was, at bottom, a study of exactly this kind of struggle — over what a community agrees to see. His framework was built for scientific communities, not public discourse, and the transfer must be made with care. But the central instrument travels. A frame, in Kuhn’s sense, is not a slogan. It is a way of organizing perception so thoroughly that the people inside it stop noticing they are inside anything at all. The question for this week is whether any of the four contending words has that kind of grip — and whether the facts are quietly accumulating against any of them.

Is This a Paradigm Contest at All?

The first discipline Kuhn’s framework imposes is restraint. The phrase “paradigm shift” has become cheap currency, slapped on every product cycle and quarterly pivot. Kuhn’s actual machinery is expensive. A paradigm, in the sense he gave the word in The Structure of Scientific Revolutions, is not merely a popular theory or a dominant mood. It is a shared achievement — an exemplar, a worked instance so compelling that a community organizes its entire practice around it, agreeing without further argument on what counts as a problem and what counts as a solution.

In his later, more careful formulation in “Second Thoughts on Paradigms,” collected in The Essential Tension, Kuhn narrowed the term precisely because it had begun to mean everything and therefore nothing. A paradigm, properly understood, is the concrete shared example that lets practitioners recognize new situations as like the settled ones. It is learned by doing, not by definition.

Apply that demanding standard to the Frame War and the first finding is sobering. Most of what passes for a frame war is not a paradigm contest. It is a quarrel over vocabulary conducted by camps that already share more than they admit. When a vendor calls AI a “partner” and a critic calls it a “tool,” they are very often disputing the honorific — the flattering or deflating word — while agreeing entirely on the underlying facts: that these are statistical systems trained on large corpora, that they produce outputs of variable reliability, that they are owned by firms and sold for profit. A quarrel over what to call a thing both parties describe identically is not revolutionary. In Kuhn’s terms, it is closer to normal science — rival camps solving puzzles inside a shared frame, the frame being “AI is a commercial software system whose capabilities we are arguing about.”

So the test is not whether the words differ. The words always differ. The test is whether the camps disagree about what counts as a problem — whether they would recognize the same event as significant, or whether one side’s decisive evidence is, to the other, simply not on the table.

By that test, parts of the Frame War do show genuinely revolutionary structure. The “transformation” narrative and the “tool” narrative do not merely value AI differently. They disagree about what kind of fact would settle anything. To the tool camp, the relevant facts are present-tense and measurable: error rates, costs, what the system can and cannot reliably do today. To the transformation camp, present performance is almost beside the point; the relevant fact is a trajectory, a curve extrapolated forward, a direction of travel that makes today’s limitations look like a rounding error. These two are not arguing about the same object on a shared scale. They are reaching for different scales entirely.

That is the signature Kuhn taught us to look for. Not louder disagreement, but incommensurable disagreement — the absence of a common measuring stick.

What Each Frame Cannot See

Kuhn’s most durable insight is that every frame is simultaneously a way of seeing and a way of not-seeing. A paradigm makes certain facts vivid and obvious. It makes others invisible — not hidden, exactly, but unregistered, the way you do not notice a sound until it stops. Each of the four contending words performs this double act. Each surfaces an anomaly and papers over another. An anomaly, in Kuhn’s vocabulary, is a fact the reigning frame cannot comfortably absorb — a violation of expectation that the frame must either explain away or refuse to look at.

“Tool” makes agency visible and dependency invisible. Call AI a tool and you foreground control: a tool is wielded, owned, put down. This frame is honest about responsibility — a hammer never killed anyone, the hand did — and it resists the mystification by which firms launder their products’ failures into the products’ own mysterious “behavior.” But the tool frame papers over a specific anomaly: tools that are not switched off. A hammer in the drawer is inert. A recommendation system, an automated screening pipeline, an always-on agent embedded in the channels through which people now communicate — these do not wait in the drawer. They run. The tool frame cannot easily see the fact of ambient operation, of systems acting continuously and at scale without anyone wielding them in the moment. That is the anomaly accumulating beneath “tool.”

“Partner” makes collaboration visible and ownership invisible. This is this week’s aggressive entrant, pushed hardest by the parties who profit from it. Call AI a partner and you foreground reciprocity, give-and-take, a working relationship. But the word performs a quiet sleight of hand that the column is obliged to name. Partners have standing. Partners have interests. The systems being marketed as partners have neither — they are property, owned by firms whose interests are not yours and whose terms you do not negotiate. The “partner” frame papers over the entire fact of ownership. Its anomaly is the contract: every “collaboration” runs on infrastructure you rent, can be revoked, repriced, or retrained, and answers ultimately to a balance sheet you never see. A partner you cannot fire, who reports to someone else, is not a partner. The word is doing concealment work.

“Threat” makes power visible and ordinariness invisible. The threat frame is the critic’s instrument, and it has a real virtue: it keeps the question of power on the table. Who owns these systems, who is displaced, who is surveilled, who bears the downside — the threat frame insists on asking. But it surfaces danger at the cost of obscuring the dull, undramatic ways AI is already woven into ordinary life as mere utility. Most uses of these systems are neither sinister nor revolutionary. They are boring. The threat frame cannot see boredom. Its anomaly is the vast unremarkable middle — the millions of small, low-stakes uses that fit no story of catastrophe.

“Transformation” makes scale visible and the present invisible. This is the frame that most insistently borrows the word “revolution,” and the one this column must police hardest. The transformation narrative foregrounds magnitude and inevitability — a force of history, a before and after. Its power is that it captures something the other frames miss: that whatever AI is, it is not arriving one user at a time but all at once, structurally. But “transformation” papers over the present. It is a frame built entirely from the future tense. Its anomaly is the gap between the promised future and the delivered now — the reliability problems, the failures, the unglamorous ceiling of what the systems actually do today. Every fact about present limitation is, to the transformation frame, merely temporary, and therefore invisible.

Notice what the diagnostic reveals. No single frame is simply false. Each sees something real and each is blind in a specific, nameable place. That is precisely Kuhn’s point in The Copernican Revolution, where he showed that the Ptolemaic frame was not stupid — it was a powerful, predictive, internally coherent way of seeing that happened to make certain facts impossible to register. Frames are defeated not by being wrong about everything but by accumulating too many facts they cannot hold.

Where the Camps Talk Past Each Other

The deepest of Kuhn’s tools is incommensurability — the condition in which rival communities lack a shared standard, so that each judges the disputed object by criteria the other does not accept. In his late work, gathered in The Last Writings — Incommensurability in Science, Kuhn refined this from a claim about total mutual unintelligibility to something subtler and more useful: incommensurability is local. The camps share most of their language. They diverge on a small cluster of central terms — and that small divergence is enough to make agreement impossible, because the contested terms are the ones that decide what counts as evidence.

The Frame War shows exactly this local structure. The camps share almost everything. They use the same product names, cite the same demonstrations, read the same announcements. They diverge on a handful of words — “partner,” “transformation” — and on one thing more, the thing Kuhn taught us to look for: their exemplars. The concrete case each side treats as the obvious, settled instance of what AI is.

For the “partner” and “transformation” camps, the exemplar is the demonstration of fluent, surprising, near-human competence — the moment a system produces something that looks like understanding. That case is treated as the real AI, the true face of the thing, with failures dismissed as growing pains. For the “tool” and “threat” camps, the exemplar is the failure, the confident error, the system that produces fluent nonsense or amplifies a harm at scale. That case is treated as the real face, with successes dismissed as cherry-picked demos.

Here is the incommensurability, exact and local. The two communities are not weighing the same evidence and reaching different verdicts. They have chosen different evidence as paradigmatic — different exemplars — and from those different starting cases, every subsequent fact is read in a different light. A success confirms one camp and is explained away by the other. A failure does the reverse. There is no neutral demonstration that both sides will read the same way, because they disagree about which demonstration counts as the instance of what AI is. This is why the Frame War does not resolve through argument. The parties are not failing to persuade each other. They are reading from different exemplars, and exemplars, Kuhn insisted, are learned and felt before they are argued.

What the Numbers Say About Who Is Winning

A frame’s dominance shows in two ways: in which words the public adopts, and in the gap between adoption and experience. This week’s data lets us locate at least one such gap, and it is the most analytically load-bearing fact available.

The “partner” and “transformation” frames are, by any measure of volume, winning the discourse. The language of agents, collaboration, and historical inevitability saturates the announcements and the coverage that amplifies them. That is unsurprising. The parties pushing these frames are the parties with the largest budgets for pushing them, and a frame backed by capital travels fast.

But volume is not grip. Kuhn’s point about paradigms is that they win when a community cannot any longer see the world any other way — when the frame becomes invisible because it has become total. And here the data surfaces the anomaly. The gap between the confident “partner/transformation” narrative and the public’s reported experience of reliability remains wide. The systems sold as collaborators continue to produce errors at rates that no one would tolerate from an actual partner. The trajectory promised by the transformation frame keeps colliding with a present that has not delivered it. The dominant frame holds the megaphone. It does not yet hold the experience.

That gap is the anomaly accumulating beneath the winning narrative. In Kuhn’s account, a frame can dominate discourse long after the facts have begun to strain against it. Ptolemaic astronomy held its position for over a century after the discrepancies were known, because no rival frame had yet organized the anomalies into a better picture. The “partner” frame is in a structurally similar position. Its dominance is real and its strain is real, and the two can coexist for a long time — until they cannot.

What Would Actually Move Our Reading

This column ends where Kuhn’s discipline always points: at the question of evidence. Not which frame is loudest, not which is most flattering, but what would have to happen for one frame to genuinely defeat the others — to win in Kuhn’s demanding sense, by reorganizing what the public can actually see.

Rhetorical victory is cheap and continuous. The “partner” frame wins rhetorically every week, in every announcement. That is not what Kuhn’s framework asks about. A paradigm shift, properly understood, is not a change of vocabulary. It is a change in what counts as obvious. So the question is precise: what evidence would shift the public’s exemplar — the case it instinctively treats as the real face of AI?

For the “partner” or “transformation” frame to truly win, the anomaly beneath it would have to close. The reliability gap would have to disappear — not in a demo but in ordinary daily use, durably, so that the experience of the systems stopped contradicting the language used to sell them. The decisive evidence would be the day a typical person’s exemplar of AI ceased to be the confident error and became, instead, the routine quiet success they no longer think about. That is what total framing looks like: not louder claims, but a public that has stopped registering the failures because the failures have stopped.

For the “tool” or “threat” frame to win, the reverse would have to happen. A failure large enough, public enough, and structurally undeniable enough to install itself as the exemplar — the case everyone reaches for, the one that makes the “partner” language sound, retroactively, like the marketing it was. Not a single dramatic catastrophe necessarily, but an accumulation of registered harm that the ownership-concealing word can no longer paper over.

Until one of those things happens, the Frame War will continue as it is: a contest of vocabulary among camps reading from incommensurable exemplars, with the best-funded frame winning the airtime and the strained frame quietly hoarding its anomalies. The reader’s task, in the meantime, is the one Kuhn’s instrument is built to serve — to notice which word is doing concealment work, to ask what each frame cannot see, and to refuse the loudest narrative the authority to name the thing before the facts have decided whether it deserves the name.

The Frame War is not over. In Kuhn’s sense, it has barely begun. What ends a frame war is never the better argument. It is the better exemplar — and the public has not yet settled on which case is the real one.