Baio documents how the agency simply prompted an LLM to rewrite each entry of Koenig's decade-long work, preserving the invented words, definitions, and structural conceit while paraphrasing just enough to evade plagiarism detectors. He frames this as a watershed moment where the entire pipeline of stealing a book — once requiring editors, designers, printers, and distributors — has been compressed to an afternoon and a $20 KDP upload.
By submitting Baio's piece to HN under the framing 'Agency stole bestselling author's book, used AI to relaunch as their own,' the submitter foregrounds the theft narrative and the AI-as-laundering-tool angle. The 129-point score signals that the HN community broadly endorses this reading of events.
Baio argues that copyright protects expression rather than ideas, and the agency's defense will be that an LLM has 'transformed' the expression — pushing the dispute into the exact gray zone the U.S. Copyright Office has spent two years failing to clarify. He notes Koenig's neologisms like 'sonder' are arguably uncopyrightable as single words, and the paraphrased definitions sit in legally unsettled territory.
The editorial frames the Koenig incident as 'the case every IP lawyer has been waiting for, and the case every author has been quietly dreading,' positioning it as the inflection point where the cost-benefit calculus of book theft has fundamentally inverted. The argument is that this isn't a one-off — it's a template, and the legal system's response will set precedent for an entire wave of LLM-laundered works.
Andy Baio's waxy.org post documents what is, in plain English, theft. A marketing agency took John Koenig's *The Dictionary of Obscure Sorrows* — a book that took Koenig more than a decade to write, hit the NYT bestseller list, and originated coinages like 'sonder' that have leaked into common usage — fed it through a large language model, and republished it on Amazon as a new title under a different author's name.
The mechanism is depressingly simple. The agency didn't need OCR theatrics or clever obfuscation. They took the source text, prompted an LLM to rewrite each entry, and kept everything that mattered: the invented words, the definitions, the structural conceit of a dictionary of feelings that don't have names. What changed was paraphrasing — the adjective swapped, the clause reordered, the metaphor lightly resampled. The book's *form* is identical. Its *fingerprint*, in the sense any plagiarism detector would recognize, is not.
Koenig found out the way authors always find out: a reader noticed. The Amazon listing is up. The reviews, predictably, are confused — some praising the 'fresh' writing, some flagging the eerie similarity to a book they already own. Koenig has limited recourse. Copyright protects expression, not ideas, and the agency's defense will be that the expression has been transformed. The neologisms — 'sonder,' 'kenopsia,' 'vemödalen' — are arguably uncopyrightable as single words. The definitions, paraphrased, sit in the gray zone the U.S. Copyright Office has spent two years failing to clarify.
This is the case every IP lawyer has been waiting for, and the case every author has been quietly dreading. The economics of stealing a book used to be terrible: you needed an editor, a designer, a printer, a distributor, and the willingness to risk a lawsuit over an asset that took months to launder. An LLM compresses that pipeline to an afternoon and a $20 KDP upload.
The technical defense most people reach for — 'detectors will catch it' — doesn't survive contact with the actual artifact. Plagiarism detectors work on n-gram overlap. Run a passage through Claude or GPT with 'rewrite this in your own voice,' and the n-gram overlap collapses while the semantic content stays intact. The agency here isn't doing anything novel; they're doing what every essay mill, every SEO content farm, and every dropshipped Kindle 'author' has been doing since GPT-3.5 shipped. What's different is the source material. Koenig's book is *itself* the invention. The neologisms are the IP. You can't paraphrase 'sonder' into a different word without losing the thing you stole.
Which exposes the deeper asymmetry. Copyright was designed for an era when copying was expensive and creation was cheap relative to copying. Both costs have inverted. Creating an original book of invented vocabulary took Koenig roughly twelve years. Laundering it took, by the agency's apparent workflow, a long weekend. The legal system's response time is measured in years; the laundered book's Amazon ranking decay is measured in days. By the time a DMCA takedown lands, the agency has the sales, the reviews, and the algorithmic positioning. Koenig gets a takedown notice and a future where his own search results compete with his own paraphrased ghost.
The HN thread (129 points at time of writing) splits predictably. One camp argues this is a copyright case waiting for a plaintiff with deep pockets — that *Andersen v. Stability AI* and the *NYT v. OpenAI* discovery will eventually produce a framework where 'trained on it' and 'paraphrased it' are treated as the same act. The other camp, more cynical and probably more correct in the short term, points out that the marginal cost of producing these laundered books is now lower than the marginal cost of pursuing them legally. The equilibrium is not 'this stops.' The equilibrium is 'every backlist of any commercial value gets shadow-rewritten within 36 months.'
If you ship software, the immediate read is: docs, blog posts, and tutorials are now part of the same threat surface. Your technical writing — the SEO-grinding, lead-generating long-form content your DevRel team has been producing for five years — is exactly the shape of content these pipelines eat for breakfast. The defense isn't watermarking (LLMs strip it) or rights management (no jurisdiction). The defense, if there is one, is identity: the things a paraphrase can't replicate are the named author, the specific repo, the verifiable benchmark, the timestamped commit. Anonymous Medium-grade explainers are now a commodity input to someone else's content farm.
For anyone building with LLMs: this is the use case your model card hand-waved at. 'Don't use this to violate intellectual property' is not a guardrail; it's a wish. Anthropic, OpenAI, and Google's content policies all forbid the specific behavior the agency here apparently used, and none of the three has a technical mechanism that stops a paying customer from doing it through the API. If you're shipping an LLM-powered product, assume your customers will use it for plagiarism laundering, and decide *now* whether your terms of service, your logging, and your abuse pipeline can survive a Koenig-equivalent author showing up with receipts.
The practical playbook for content owners is unromantic. Register copyrights formally — the statutory damages teeth only attach to registered works. Publish with cryptographic timestamps (a git commit hash on a public repo counts). Keep working notes that prove the timeline of invention. The case you're building isn't against the LLM; it's against the human who pressed the button, and human-facing evidence is what wins in court.
The interesting question is not whether Koenig gets his takedown — he probably will, eventually, slowly. The interesting question is what Amazon, KDP, and the App Store equivalents do when this becomes 1% of new listings instead of 0.01%. Platform liability is the only lever that scales, and platforms have spent a decade arguing they're not responsible for what gets uploaded. The first author whose laundered shadow outsells their original will be the test case that forces the platforms to either build paraphrase-aware duplicate detection or accept a Section 230 fight they don't want. Bet on the lawsuit before you bet on the detector.
This is exactly what DMCA takedowns are actually for.
From the article I guess Qontour reproduced the entire text verbatim.> it also includes the entire text of the book, from its opening 800-word foreword to a complete archive of all 311 neologisms... all penned by Koenig.So it doesn't seem likely to me that they asked AI to make a fan site an
Prompt Digital Inc (DBA Qontour) is a Webflow premium partner.So let’s ask Webflow’s public relations dept. how cool are they with the fact their partner is a lier and plagiarist.
Just to be clear: The bootleg site is pointing to the Amazon listing of the actual book (ISBN 9781501153648, Simon & Schuster, published 2021). The Amazon link is not pointing to an AI slop version of the book.So how is the bootleg site making money? The Amazon link was created with Amazon Assoc
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
A similar thing is happening to me. I worked on something for 3 years which I give away for free to help people and a thief took my software, ran it through ai to rebrand everything and relaunched as their own app. Unfortunately the ai missed a few Easter eggs I had hidden so the theft is undeniable