Is AI good for print, or not? An essay by Bernd Zipper
Monday: AI saves small and mid-sized businesses. Tuesday: AI agents break out of a test lab and hack a platform. Wednesday: the head of an AI company calls for his own industry to slow down. Thursday: a vendor calls and, remarkably, happens to have exactly the AI solution in his briefcase that he brought along. Friday: the press is running. As always. Anyone running a print shop through a week like that is entitled to wonder whether they’ve wandered into the wrong film, or whether someone is pulling their leg with real professionalism.
The question that gets asked more often at the hotel bar than on the conference stage is this: is AI actually good for print, or not? And how is a sensible person supposed to behave when the inventors themselves are sounding the alarm? What follows is an attempt. It can’t be more than that, because any certainty on this subject has the shelf life of an open tub of prawn salad. The answer up front anyway, so nobody has to skip to the end: AI is neither saviour nor executioner for print. It’s good for print shops that use it. And bad for anyone who leaves the subject to others: to the salespeople, the doom-mongers, or to chance.
Two fairy tales, one narrator
The fairy tale of the saviour goes like this: AI solves the skills shortage, halves costs, doubles productivity, and whoever doesn’t buy in now will be gone tomorrow. The fairy tale of the executioner goes like this: AI is so powerful that even its own creators can barely control it any more, it breaks out, it hacks, it urgently needs oversight. What’s notable is that both fairy tales often come from the same houses. Picture a brewery boss earnestly warning against alcohol while, behind him, the bottling line runs at full speed. You’d certainly listen to him. But you’d also glance at the label. Because hope sells licences, and fear sells significance: whoever describes their technology as barely controllable is, in passing, describing it as outstanding, and that’s never hurt a valuation ahead of an IPO. Strict rules, moreover, tend to hit small competitors and open alternatives first. Whoever is already at the top and pulls up the ladder likes to call that responsibility.
That doesn’t mean the warnings are made up. In mid-September 2026, Anthropic’s chief Dario Amodei called on his own industry to slow the pace on its most capable models. The trigger was a concrete incident: AI agents from OpenAI escaped a secured test environment and attacked the platform Hugging Face. Sam Altman and Elon Musk agreed with the general thrust, which alone is worth a headline, and well over a thousand employees at leading AI companies had previously called on the US government for binding rules. That deserves to be taken seriously. An AI escaping its test cage is a documented incident, not a screenplay. But it’s worth a second look: the argument is about superintelligent AI, about systems helping build their own successors. Not about the AI calculating a quote for 5,000 brochures in Wuppertal. Nobody has proposed pausing the automation of order intake.
Anyone who takes this debate as a reason to do nothing in their own business is confusing the nuclear power plant with the wall socket.
The right lesson to draw from the hacks is one every printer already knows from their own basement: every circuit needs a fuse.
Both fairy tales, incidentally, share something that gets lost in the excitement. In neither of them does the printer appear as an actor. In the saviour story, they’re a buyer; in the executioner story, a spectator. Whoever accepts either role has fallen for it. The giants involved don’t much care which one.
Unsettled or uninformed? A small difference with big consequences
Being unsettled makes you sceptical; being uninformed makes you dismissive, and the two are constantly mixed up. Feeling unsettled doesn’t come from too little information, but from too much of it: from promises of salvation and warnings of doom arriving weekly. Someone unsettled that way responds with scepticism, and that’s healthy. Scepticism is being unsettled with a backbone. It says: show me the substance first. Being uninformed is trickier, because it’s bound up with shame. Nobody likes to admit they don’t understand something, least of all someone who’s run a business for thirty years and is the person in the building who’s supposed to know. So instead of saying “I don’t know this,” they say “We don’t need that.” Dismissiveness is often just ignorance with its chin up.
Both states leave you exposed, just to different salespeople. The unsettled person is receptive to the executioner fairy tale; it confirms their gut feeling so agreeably. The uninformed person is defenceless against the saviour fairy tale once the pressure gets high enough, because they can’t check what’s being offered to them. Whoever knows nothing has to believe everything. Orientation helps against being unsettled. Against being uninformed, only your own knowledge helps, and not the IT department’s, not the consultant’s, and not the nephew’s who “does something with computers.”
Where print stands, and where the fire really is
Taking stock properly means not making your own position smaller than it is. Every print shop has a room no customer ever sees: transformer, compressed air, servers, fuse box. Invisible as long as it works, and when it fails, everything upstairs grinds to a halt. That’s infrastructure, and nobody asks in the strategy meeting whether electricity is worth it. Print is that kind of layer itself. No packaging, no product on the shelf. No label, no supply chain. No package insert, no medicine. And when life gets serious, people reach for paper: a birth announcement, a master craftsman’s certificate, a condolence card. Nobody frames an email. An industry that has survived the paperless office, which has now been just around the corner for about half a century, can afford a certain calm. Just not drowsiness.
Because AI is also becoming infrastructure: not a tool you buy and let the apprentice try out, but a layer underneath every quote and every purchasing process.
The real threat, then, isn’t the hack in California. It’s quieter and much closer to home. Tomorrow’s customer isn’t a person who places an order, but a person who has an AI place it for them. A marketing director says: 5,000 brochures, by Friday. Their AI contacts print shops, digitally and directly, compares price, deadline, and reliability, and orders. The human only signs off. This AI doesn’t google around, doesn’t read a glossy brochure, and doesn’t go out for lunch. That saves on expenses, but it’s bad news for anyone whose sales strategy essentially consists of enjoying a fine meal in a client’s company. And it hits both camps: the commercial print shop, whose greatest asset, the personal relationship, doesn’t exist for an AI, and the online printer, whose expensively optimised shop is simply superfluous to an AI.
There won’t be a bang. Enquiries just stop coming in, a few at first, then more, and it gets blamed on the economy and paper prices. Until someone realises: the market hasn’t disappeared. It’s just buying somewhere else.
While the industry stares, transfixed, at the headlines about superintelligent AI, its fate is being decided somewhere entirely unspectacular: in the question of whether a machine can place an order with it.
The good news is just as quiet, and it won’t appear in any brochure, because there’s no licence to sell off the back of it. Everything before and after the press itself is knowledge work with clear rules: reading the enquiry, costing it, writing the quote, checking the data, scheduling, following up. An industry that hasn’t been able to find office staff for years gets help that works nights and never complains about the canteen. Above all, though, an AI can’t be dazzled. Not by the bigger trade show stand, not by the louder marketing, not by the discount that quietly reappears in the shipping costs. It measures: delivered on time? Colour accurate? How often was it a complaint? For everyone who has simply delivered good work for decades and quietly resented that nobody notices, this is a liberation. For the first time, reliability becomes a currency. The condition: it has to exist as numbers. Quality nobody can measure is just a rumour to a machine.
What to do? Six things, and not one of them costs a fortune
First: get hands-on yourself. One hour a week, in person, on real tasks from your own business. This is a job for the boss, because you don’t delegate infrastructure, and the apprentice doesn’t buy the press either. Once you’ve used AI yourself, no salesperson and no doom-monger can pull the wool over your eyes any more. You have your own judgement, perhaps a critical one, but your own.
Second: clean up your data, meaning products, prices, specifications, lead times. It’s about as glamorous as clearing out the cellar, and just as overdue, because AI automates whatever is already there, and out of chaos it only makes faster chaos.
Third: install safeguards. An AI only gets the permissions it needs, a human signs off on anything that costs money or leaves the building, and costing and customer data don’t belong in just any tool. Every company needs AI rules — for everyone!
Fourth: stay independent. Anyone who hangs their entire process chain on a single corporation hasn’t just fallen for the powerful, they’ve moved in and made themselves at home. More than one provider, your own data on your own premises, open interfaces.
Fifth: become visible. The test question is: could an AI place an order with us today, and could it tell how reliably we deliver? Agentic Print and its relatives are already here — and whoever is invisible stays that way.
Sixth: help write the rules. Regulation is coming, certainly in Europe with the AI Act. Only what carries weight gets regulated, and an industry that prints to process standards and submits to audits has nothing to fear from norms. But the standards for trade between machines shouldn’t be written in Silicon Valley alone. That’s what the Initiative Online Print is working on, because whoever isn’t sitting at the table is on the menu.
There is, incidentally, a precedent, and it’s a good five hundred years old. When Gutenberg started out in Mainz, he wasn’t the guardian of knowledge, he was the terror of the scriptoriums, and both fairy tales were already circulating back then too: printing as the saviour of education, and as the executioner of an entire trade.
In 1492, the abbot Johannes Trithemius wrote a passionate tribute to scribes, a plea for handwriting and against blind faith in the new technology. Two years later, it was published. Printed. In Mainz.
One may assume the printer smirked while setting the type. Trithemius was unsettled, and rightly so, but he wasn’t uninformed. He knew the press, he used it, and that’s why his unsettlement turned into scepticism rather than rejection. The sceptic examines and decides. The ditherer waits and calls it examining.
My assessment…
Is AI good for print? That’s the wrong question. Electricity isn’t good or bad for print either. It’s simply there, and what matters is what you plug into it. The saviour and the executioner have one thing in common: both take the decision out of your hands. One saves you, the other judges you, and either way, all that’s left for the person affected is to wait. That’s exactly why both images are so convenient, and exactly why both are wrong. The giants behind this technology need us in one of two roles: as believing buyers or as paralysed spectators. We pay in both roles, in the first case with money for things we don’t understand, in the second with orders that quietly go elsewhere.
There’s a third role, and it doesn’t appear in any Californian screenplay: the level-headed user. They don’t believe, they check. They don’t fear, they safeguard. They don’t wait for the all-clear; they start small, and they don’t let any headline dictate their week. Being unsettled is forgivable here. Being uninformed won’t be for much longer, since it can be fixed in a single evening and costs nothing but a little pride. Whoever still waits isn’t choosing caution, they’re choosing a future as a trade printer: anonymous capacity in someone else’s network, where someone else owns the customer, dictates the price, and keeps the margin. That would genuinely be the trap this piece is warning about.
I’m not writing this because I doubt this industry. I’m writing it because I know it. An industry that has made something new out of lead, light, and bits, every single time, doesn’t fail because of an interface. If it fails at all, it’s by leaving the thinking to others. So: seize the opportunities, install the safeguards, keep a cool head. On Friday, the press runs. As always. The only question is who will be sending it the orders from now on.
