Everyone Is an AI Company Now — So the Label Tells You Nothing
When demand is boundless and supply is free, you do not get innovation; you get a gold rush.
The renaming
A vendor sends through its renewal deck. Set it beside last year’s and the exercise takes about ninety seconds. The product is the same product — the same screens, the same data model, the same three features shipped since the last cycle. What has changed is the vocabulary. What the 2022 deck called “advanced analytics” the 2023 deck calls “AI-powered insight.” The “rules engine” is now “intelligent automation.” A line that used to sit in eight-point grey at the foot of a slide — “powered by machine learning” — has migrated to the title, in bold, and been promoted to “Generative AI.” Nothing under the hood has moved. The find-and-replace has.
This is the defining commercial event of the year in enterprise software, and it deserves to be named plainly: over the course of 2023, every software company became an AI company. Not by building anything, in most cases, but by saying so.
Why the whole market moved at once
The trigger is not mysterious. A chatbot arrived last winter and did something no enterprise product had managed in a decade — it put the technology directly into ordinary hands, no procurement cycle, no implementation partner, an answer in the time it takes to type a question. The board saw it. The board’s teenagers saw it. By spring, “what is our AI strategy?” had become the question every executive was being asked and none could afford to answer with silence.
That is the mechanism, and it is worth stating precisely, because it explains the uniformity. Demand for the word became infinite and instantaneous. The supply of the word costs nothing — a slide edit, a press release, a repriced SKU. When demand is boundless and supply is free, you do not get innovation; you get a gold rush. And this one is a rush on vocabulary, not on gold.
The distinction that survives the noise
So here is the single distinction worth carrying into every vendor meeting between now and the end of next year. When a supplier says “AI,” it is selling one of two quite different things, and they are not the same purchase.
- A capability — the product now does something it demonstrably could not do a year ago, something the buyer could not easily do for themselves, and that new doing is what the money is for.
- A label — the product does more or less what it always did, and the word has been painted onto the outside of it, exactly as “cloud” was painted onto hosted software a decade ago and “digital” onto everything the decade before that.
The rush is overwhelmingly the second kind. This is not cynicism; it is arithmetic. Building a capability takes scarce talent, data you may not have, and quarters you cannot spare. Applying a label takes an afternoon with the marketing team. Faced with an executive demanding an AI story by the next board meeting, most product organisations will, entirely rationally, ship the afternoon’s work.
A worked example, and the question that cuts through it
Take a mid-market service-desk tool — a composite, but a familiar one. Last year it sold “smart routing”: a rules engine that read a ticket and dropped it into the right queue. This year it sells an “AI agent.” Put the only question that matters: what can it do this November that it could not do last November?
In the honest version, the answer is real and narrow. It now drafts a suggested first reply by drawing on the knowledge base, and a human still reads and sends it. That is a genuine capability. Say it trims average first-response time from four hours to fifty minutes on the third of tickets it can safely touch — worth paying for, and worth measuring. In the dishonest version, the answer is: the same routing as last year, relabelled, at a list price twenty per cent higher because the adjective now commands a premium. Same code. New coat. Fifteen points of margin conjured out of a word.
The diagnostic that separates the two costs nothing and never fails: ask the vendor what the system gets wrong.
A supplier selling a real capability can describe its failure modes in specifics — where the generation invents things, what share of drafts an agent rejects before sending, which categories of query it refuses. A supplier selling a label cannot, because there is nothing new underneath to fail. A candid failure story is the surest evidence that something was actually built.
Enthusiasm has no failure modes. Engineering does. The presence of a precise, slightly embarrassing account of where the thing breaks tells you more than any capability slide, because no marketing department volunteers weaknesses for a product that is only a rename.
The objection, which is fair
The honest counter to all of this is that reflexive dismissal is its own error, symmetrical to the hype. Some of what is being relabelled is real, and some of what looks like a fresh coat is the leading edge of a genuine shift that will compound over years. The buyer who priced “cloud” at zero in 2010 because the brochures were overheated went on to miss a decade. The sceptic’s failure is the mirror of the enthusiast’s: one overpays for the word, the other underpays for the thing.
Both errors are real, and the answer to both is the same. It is not to disbelieve the category — the underlying capability is arriving, unevenly and faster in some corners than the slides admit. It is to refuse to buy the word, and to insist on paying only for capability that has been shown to work on your own material.
The posture for the next eighteen months
Neither the executive’s panic nor the contrarian’s sneer, then. Something duller and more effective:
- Price the label at zero. Assume “AI-powered” adds nothing to a product’s worth until it is demonstrated, and pay no premium for the adjective alone.
- Buy the delta, not the noun. Make the vendor state, in one sentence, what the product does now that it could not do twelve months ago — and pay for that sentence only if it is both true and useful to you.
- Make them show the break. A capability has failure modes it can describe on your data; a rename does not.
- Re-run the test at renewal. The label will still be on next year’s deck. The capability may by then have genuinely arrived — or may still be a slide. Check, every cycle, rather than deciding once.
The word “AI” is doing an enormous amount of work in this market right now, and almost none of it inside the products. That gap will close, but not evenly and not on the marketing calendar. The suppliers who merely renamed their software this year will spend next year explaining why the capability everyone was promised has not shown up; the few who quietly built something will let the results make the argument for them. Our job, from the buyer’s chair, is the least glamorous one in the room: to stop reading the label, and to start reading the delta.