Why Publishing’s Defensive Crouch is a Strategy for Self-Shrinkage
The book publishing industry is currently trapped in a state of cognitive dissonance that threatens its future growth.
Globally, trade bodies and author unions project an image of righteous defiance. They issue stern press releases, deliver hand-written letters to tech conglomerates, and lean heavily into a romantic narrative: human creativity is a sacred, unassailable monolith, while artificial intelligence is a fundamentally incapable machine built entirely on theft.
It is a comforting, high-minded stance. It is also an intellectual fraud.
Behind closed doors, the very same executives nodding along to anti-AI manifestos are quietly reviewing operational spreadsheets to see where machine learning can trim overheads, accelerate production, and squeeze margins. The industry has spent years sitting on a fence constructed by a mixture of fear of the Luddite fringe – terrified of the soundbite megaphones wielded by organisations like the UK’s Publishers Association (PA) and the Society of Authors (SoA) – and, perhaps worse still, a lack of joined-up thinking.
In doing so, publishing has failed to comprehend a fundamental truth of modern corporate strategy: historical legal grievances and future commercial opportunities are not mutually exclusive. They can, and should, run side by side.
The Rhetoric vs. Reality Gap: A Failure of Intellectual Honesty
To understand the scale of the hypocrisy – a topic I dealt with comprehensively here at TNPS earlier this month –
– one only has to look at the selective outrage of publishing’s self-appointed watchdogs. When Meta or OpenAI ingest copyrighted text to train large language models, the PA and the SoA rush to declare it a “great copyright heist.” They demand absolute consent, transparency, and remuneration.
Yet when Amazon-owned Audible expands its AI narration programme to over a hundred synthetic voices, and quietly moves an AI translation service into beta with select publishers, the outrage all but evaporates. To be fair to Audible, its translation pathway – unlike the wholesale ingestion that fuels LLM training – builds in an optional human-linguist review step. That’s a meaningful difference, and one publishing’s watchdogs would normally insist on shouting about as a concession worth demanding everywhere else. Instead: near silence.
Why does Amazon get a free pass from the very bodies that claim to protect human creators? The answer is as simple as it is craven: commercial dependency. Amazon holds the industry’s distribution infrastructure, its data, and its direct-to-consumer access points. Challenging an independent tech startup costs nothing in trade capital; challenging the ecosystem that dictates your quarterly revenue is an entirely different matter.
This contradiction exposes the moral posturing for what it is. The industry decries the legal and ethical implications of LLMs reading books to learn patterns, but remains perfectly content to let those same models replace freelance copyeditors, proofreaders, and indexers to save a line item on a production budget, and to model robots that run warehouses at the expense of countless human jobs.
This is not an ethical framework; it is a selective commercial policy masquerading as morality.
A Candlelit Postscript, and a Better Question
I’ve been writing this from inside The Gambia’s current power crisis, sketching it out by candlelight between NAWEC’s load-shedding and, this week, our first serious rain-storm of the season hammering away outside. Between the dark and the downpour, I’ve still managed to keep track of some solid reportage over at the new-look Publishing Perspectives, and a piece by Javier Celaya stopped me mid-sketch. It’s worth reading in full: “From IP Anxiety to Growth Strategy: The AI Conversation Publishing Needs to Have.”
What Celaya is pointing at, in essence, is that the industry’s collective obsession with using AI merely to lower the cost floor, rather than raise the growth ceiling, is a direct consequence of a profound structural blind spot. Knowing how to write a prompt for a chatbot is not AI literacy. True structural fluency – understanding data sovereignty, probabilistic mechanics, and systemic workflow automation – is missing at the executive level. While publishing stands frozen in the defensive crouch described above, the rest of the corporate world has moved from tools to agents, leaving the book trade exposed to the very tech giants it claims to be fighting.
From the Floor to the Ceiling: Moving Beyond Cost Savings
The current deployment of AI within publishing houses is, how can I put this politely, remarkably uninspired. It is an exercise in shrinkage. Executives look at machine learning and see a tool to shave 15% off a translation budget, automate a press release, or bypass a human narrator for an obscure backlist audiobook.
As Celaya says, this focus represents the operational floor of artificial intelligence, not its ceiling. If an industry uses a transformative technology purely to cut costs, it succeeds only in shrinking its own economic footprint. The true power of AI lies in its capacity as a growth engine.
Celaya frames the core provocation directly: “How can AI help publishing grow, rather than simply shrink its cost base?”
One answer is to look at how the industry currently handles international expansion, and contrast it with an automated, agent-assisted future.
The traditional foreign rights model is a friction-heavy legacy framework that severely limits a book’s potential. A manuscript is finished. Rights agents shop it, territory by territory, at book fairs and over months of correspondence. A handful of local publishers, in a handful of languages, decide whether to gamble on it. Translation is commissioned locally, on a local timetable. Years can pass before – if – a title reaches a second language at all, and the vast majority of midlist titles never travel beyond their home market.
An AI-assisted localisation pipeline can flip this dynamic, though I should try be clear about what’s actually being automated and what isn’t – precision being rather the point of this whole piece.
Here’s the thing I love: The moment a manuscript is finalised, an automated workflow can trigger translation into multiple languages in parallel, rather than waiting years for individual foreign rights deals to materialise. This becomes genuinely agentic – not just “AI did the translation” – when there’s real decision-making layered on top: an orchestrating agent deciding which territories and formats to prioritise based on backlist performance data, say, or a consistency-checking agent that cross-references character names, terminology, and tone against a publisher’s style bible across an entire catalogue, flagging anomalies for a human editor rather than translating straight through them.
Human editors don’t disappear from this picture; their role shifts. Rather than translating from scratch over months, they become strategic process architects – reviewing flagged sections for cultural nuance, idiom, and artistic intent. How much of a manuscript needs that close attention varies enormously by language pair and genre. Between English and Spanish or French, the technology is strong; into lower-resourced languages, or through dense literary wordplay, the human share of the work remains substantial, and it would be dishonest to pretend otherwise.
Even fully realised, though, this only solves half the problem – the production half.
A translated, formatted file isn’t the same as a discovered, sold book. Foreign rights deals have never been only about translation cost; they buy a publisher local retail relationships, marketing infrastructure, and market knowledge in the destination territory.
I’ve made a version of this point before in relation to Audible’s own international ambitions: the Kindle promise of “every book in every language” foundered not because translation was impossible, but because translation was never the actual bottleneck. An AI-assisted pipeline removes the production constraint. It does not, by itself, remove the discovery constraint. Anyone selling publishers an “agentic localisation” pitch that skips this part is doing exactly what Celaya warns against – promising the ceiling while quietly papering over the floor.
That qualification doesn’t undercut the opportunity; it’s what makes the opportunity real rather than vendor-brochure rhetoric.
A mid-list title that would once have been a localised gamble, entirely dependent on a foreign publisher’s appetite, can become available in several languages and formats far sooner, with global rights and revenue retained by the original publisher rather than parcelled out territory by territory.
Getting it found and sold in each of those markets is the next problem to solve, not a solved one. By focusing entirely on how AI can eliminate domestic freelance lines, trade bodies have failed to see how it might unlock new markets at all – even imperfectly, even with real work still to do on the distribution side. They have chosen to fight over a shrinking local pie rather than start exploring how to bake a larger global one.
The Parallel Track Strategy: Partners and Adversaries
The dominant narrative pushed by trade associations is fundamentally binary: you are either a defender of human copyright or a collaborator with the machine. This binary is a strategic disaster. It betrays a complete lack of sophistication in handling disruptive shifts, particularly when compared to how other IP-heavy sectors operate.
In industries like pharmaceuticals, automotive, and broader entertainment, litigation and commercial integration are treated as standard parallel tracks. A major music label will sue an AI platform in the morning for copyright infringement regarding training data, and sit down with that same platform in the afternoon to negotiate a multi-million-dollar licensing and distribution deal for future content.
There is no contradiction here. Lawsuits are not an existential crusade; they are a mechanism to establish value, protect assets, and build leverage.
Publishing must learn to walk and chew gum at the same time. It is entirely appropriate – indeed, necessary – to pursue legal redress against tech companies that built their models on copyrighted works without consent or compensation. The legal case for past infringement is robust, at least where pirated content is concerned, and that is being fought aggressively in the courts.
But winning a lawsuit from five years ago will not save a publisher from obsolescence five years from now. Parallel to litigation, publishers must actively define collaboration scenarios with tech companies. They must reimagine discovery channels, explore how books will be consumed within conversational interfaces, and establish new revenue models based on real-time data access.
The real hypocrisy is not signing a licensing agreement with an AI firm; the real hypocrisy is loudly condemning AI in public while quietly signing opaque contracts with tech distributors and warehouse robotics firms behind the scenes to keep the business afloat. Sorry, Publishers Association and Society of Authors, it’s time to replace performative indignation with transparent, professional corporate strategy.
The Executive Blind Spot: From Chatbot Users to Process Architects
The underlying cause of publishing’s paralysis is an acute lack of genuine AI literacy at the top.
Over the past year, thousands of publishing professionals have become daily users of AI. They know how to open a browser window, paste a block of text, and ask a chatbot to generate a marketing hook or clean up a piece of copy. They look at this capability and assume they understand the technology.
They do not. Knowing how to drive a car does not mean we understand how to build an engine, let alone redesign a transportation network.
This distinction became starkly apparent to Javier Celaya during an executive master’s programme in artificial intelligence at the Instituto de Inteligencia Artificial (IIA) in Spain.
Stepping outside the defensive echo chamber of the publishing trade to study how healthcare, finance, and logistics sectors approach AI reveals just how far behind our industry truly is. While publishers are still debating whether an LLM is “creative,” other sectors are building complex networks of autonomous AI agents designed to completely restructure core business processes.
The most consequential blind spot in publishing today is the gap between interface familiarity and structural comprehension. Because managing directors and senior executives understand AI only in the abstract, they are dangerously unequipped to guide their organisations through this transition.
When a technology vendor pitches an “AI-driven publishing solution,” says Celaya, an executive without structural fluency does not know how to interrogate the product. They do not know how to ask about data sovereignty, retention policies, or data expiration dates. They do not understand the probabilistic mechanics behind why an LLM hallucinates, or what structured data actually means for enterprise security. They are highly susceptible to buying flawed, off-the-shelf vendor tools that put their intellectual property at risk while offering nothing more than marginal cost savings.
True AI literacy requires understanding where the hard limits of current models genuinely lie. It means knowing which processes within a publishing workflow can be fully automated, which require a human at the beginning or end (the “human-in-the-loop” model), and which must remain entirely human to preserve the core value of the work.
A Strategic Imperative for Leadership
Education is no longer an optional feather in an executive’s cap; it is a strategic imperative. No managing director, editorial director, or platform head can responsibly define a strategy for their business without a structural understanding of machine learning.
That doesn’t mean needing to be an expert on every aspect of AI. It does mean throwing off the blinkers and doing some joined-up thinking.
The digital transformation of the book industry has been underway for over two decades. Yet, time and again, publishing has underestimated the speed at which technology redraws the competitive map. AI is not another incremental shift like the transition from physical print to ePUB. It touches every single node of the publishing value chain: creation, curation, production, translation, marketing, rights management, and global distribution.
And yes, I did say creation.
Whatever shape the next decade of publishing takes, it won’t be defined by who shouted loudest at the tech giants, or who hid longest behind legacy distribution monopolies. It will be defined by the publishers willing to drop the performative hypocrisy of their own trade bodies, sit with the discomfort of using AI as something more than a cost-cutting tool, and put in the work to expand the boundaries of the global written word rather than simply defend its shrinking borders.
The post first appeared in the TNPS LinkedIn Analysis Newsletter.
