The Bear, the Bricklayer, and the “AI Slop” Problem
A short exchange on LinkedIn compressed the entire debate about generative AI and creative work into twenty-odd comments. A man asked whether anyone had read his children's books; people who build AI for a living called them “ai slop”; he answered with an analogy about an architect who cannot lay bricks. The craft objection, the volume objection, the honesty objection — and where the architect analogy strains. The debate does not resolve. It clarifies.
A short exchange on LinkedIn last week did something most arguments about artificial intelligence fail to do: it stayed human. There was no manifesto, no thread of forty quote-tweets, no consultant promising to “demystify” anything. There was just a man asking, with a slightly desperate string of kisses, whether anyone had actually read his books — and the man, writing on his own LinkedIn feed, was Doug Scott.
Scott is a serial entrepreneur and angel investor. He founded Redbrain, a performance-marketing and shopping-ads business that drives well over a billion pounds in incremental sales for retailers each year, and built it without venture capital — alongside a string of other ventures including Climate VC, Potential.co and DeckSender, plus dozens of angel investments. These days he describes himself, with characteristic understatement, as someone who “ambles a lot.” The books are children’s stories about a bear, published online at theheld.ai. And the post — “Did anyone read my bear books:) please say yes:) xxxx”, which you can find on his profile at linkedin.com/in/dougsscott — was the kind of thing that usually scrolls past unremarked.
What made it worth stopping for was the fight that broke out underneath it. Because buried in twenty-odd comments was the entire debate about generative AI and creative work, compressed into a form small enough to actually think about.
“AI slop”
The phrase arrived in the bluntest possible form. Among the commenters were people who work in AI for a living, and more than one of them reached for the same two words: “ai slop.” It is, by now, a familiar verdict. “Slop” has become the term of art for content that is generated rather than made — text and images produced quickly, cheaply, and in volume, with no particular person standing behind any individual sentence or brushstroke. The word does a lot of work. It implies excess, low quality, and a kind of environmental pollution of the feed. It says: this was not worth the seconds it took me to scroll past it.
It is worth taking the criticism seriously rather than waving it away, because the people making it are not Luddites. Many of them build the technology for a living. Their objection is not that the tools exist. It is more specific, and it comes in roughly three parts.
The first is about craft. Writing a book is not merely the act of having a story exist at the end of the process. It is the slow work of choosing one word over another, of cutting the scene that does not earn its place, of discovering what you actually think by being forced to phrase it. A model that produces a finished draft from a prompt skips all of that. The critic’s worry is that what comes out the other side has the shape of a book without the substance of one — competent, plausible, and somehow empty.
The second is about volume. A single person can now produce in an afternoon what used to take a year. Multiply that across millions of people and the worry is not any individual book but the flood: a publishing ecosystem, a set of search results, a children’s section, silting up with material that nobody quite made and nobody quite needs. The cost of creating has collapsed, and the critics argue that the cost of attention has not — so something has to give, and what gives is the reader’s ability to find the good thing.
The third is about honesty. When the seams are invisible, the audience cannot tell what they are getting. One commenter’s sharper jab — that asking AI to write your book is no different from asking someone else to lay your bricks while you take the credit — is, underneath the needling, a point about authorship. Who, exactly, made this? And if the answer is “a model trained on millions of books by people who were not asked and not paid,” then the discomfort is not aesthetic at all. It is ethical.
These are real arguments, and they deserve real answers rather than a shrug. It is equally true that the critique can harden into a reflex — a way of dismissing anything with the wrong provenance without engaging with the thing itself. Both moves, the easy defence and the easy dismissal, are ways of not doing the work.
The architect’s reply
Scott’s response to the criticism was to offer an analogy rather than a rebuttal.
He cannot write books, he said. What he is good at is ideas — and the lifelong problem with being an ideas person is that interpreting those ideas, turning them into a finished thing, has always been the hard part. AI, in his telling, is the tool that finally closes that gap. He compared himself to an architect who cannot lay bricks. The architect designs the house; other hands, or other tools, build it. “At the end of the day he gets a house and I get a book.”
The analogy is not improvised. It is the spine of one of the books themselves: theheld.ai works through it directly — I am the architect, then what architecture is and what construction is — so the LinkedIn reply was less a defence improvised under fire than a compression of a position already set out at length. A reader who wants the full version of the architect argument, rather than the one-line version, can read it there.
Supporters of AI-assisted work find this persuasive because it is candid about its own terms. Scott is not claiming to be a writer or pretending the prose flowed from his pen; he is describing a division of labour. Critics, equally, can read the same analogy and find it evasive — a tidy story that names the comfortable part of the process, the having of the idea, and skips lightly over how much of the actual book was decided by the model rather than the man. Both readings are available, and which one a reader reaches for tends to depend on what they already believed before they arrived.
It is one way to defend AI-assisted work, and supporters would say it rests on a distinction the word “slop” tends to flatten. There is a difference, the argument runs, between using a model to flood a marketplace with generic content nobody cares about and using a model to externalise something specific a person has been carrying around for years. On this account the first is the thing the critics are right to despise; the second is closer to what every creative tool in history has done.
That second claim leans on a long lineage. The architect did not always have steel, or the elevator, or computer-aided design — and each of those tools was, in its moment, accused of cheapening the craft. Photography was going to kill painting. The synthesiser was not “real” music. The word processor would make writers lazy. In each case the tool changed the craft, sometimes drastically, and the change turned out to be a redistribution of effort rather than an abolition of it.
A critic, though, would not let that pattern stand unchallenged. Past tools, the counter-argument goes, still required the user to acquire a skill — a camera does not compose the photograph, and a synthesiser does not write the melody. A model that turns a prompt into a finished draft is different in kind, not just degree, because it can remove the skill entirely rather than merely re-routing it. Whether AI is the next entry in that lineage or a break from it is exactly the thing in dispute, and the analogy cannot settle it on its own.
Where the analogy strains
Still, an honest article does not let the better-sounding side win on rhetoric alone. The architect comparison has a weak joint, and it is worth naming.
An architect’s drawings are themselves a deep, trained craft. The bricklayer executes a plan that already contains thousands of considered decisions. The open question about AI-assisted books is whether “having the idea” and writing a prompt is genuinely analogous to architectural design — or whether it is more like sketching a house on a napkin and hoping the contractor figures out the rest. The difference between those two things is the difference between a collaborator and a vending machine. And from the outside, in a feed, you usually cannot tell which one produced what you are looking at.
There is also the training-data question, which the analogy quietly steps around. The architect’s steel was bought and paid for. The model’s fluency was assembled, in large part, from the work of writers who did not consent to the arrangement. That is a genuine unresolved problem, and no amount of elegant comparison makes it go away.
So the debate does not actually resolve. It clarifies. The useful question is not “is AI-assisted creativity good or bad” — that question is too big to answer and too vague to be worth asking. The useful questions are smaller and sharper. Did a person bring something specific and considered to this, or did they bring a vague wish? Are they honest about how it was made? Does the result reward the reader’s attention, or merely occupy it?
The bear books
Which returns us, finally, to the bear.
The bear books are on this publication, and a reader who has followed the argument this far is the right person to go and settle it: The Bear Inherited, The Bear Looked Up, The Bear Pitched, and The Bear Read the News. The longer-form bear writing — the picture book and the others — sits across the sister sites at thebearwasright.com, thebearloved.com, and theadhdbear.com.
One detail from the thread cuts in Scott’s favour: when a reader offered to “process” the books through an AI and read the summary, Scott objected, saying they were meant to be discovered slowly rather than compressed by a machine. A sympathetic reader sees someone who cares how the work is received, not only how it was made — not the posture of a person mass-producing filler. A sceptic can answer that caring about reception is not the same as the work being good, and that an author objecting to his book being summarised is hardly unusual or decisive. The exchange tells you something about Scott’s intent. It tells you nothing, by itself, about the books.
And that is the limit the whole debate keeps running into. Neither “AI slop” nor “the future of everything” can be confirmed from a distance. The label saves a reader the trouble of opening the book; the manifesto saves the author the trouble of having written a good one. Whether theheld.ai’s bear books are worth a child’s time is not a question this article — or the LinkedIn thread, or the phrase “ai slop” — can answer. It is a question only a reader with the book in front of them can settle.
The position this small LinkedIn thread accidentally staked out is harder than either slogan. It is that the tool is neither a saviour nor a pollutant in itself. It lowers the cost of turning an idea into a finished object, which means the world is about to fill up with finished objects — some empty, some the realisation of something a person genuinely wanted to make and could not until now, and many sitting somewhere in between. The technology does not sort them. People do.
Telling them apart will be the actual work of the next decade, and it will not be done by a label or by a press release. It will be done one book at a time, by readers willing to open them. That is true whether the verdict turns out to flatter the author or not — which is the only kind of impartial test worth having.