The Clip File
Somebody else broke the story. We keep the clipping and say what it costs a one-person operation.
Every newsroom kept a clip file: a drawer of other people’s stories, cut out and annotated, because the story you did not break is still a story you have to work with. This is ours.
Three rules. The original outlet gets named and linked at the top, every time. Nothing runs unless a working operation has to make a decision about it, which means no takes on things that do not touch anybody’s hands. And it goes out under the name of whoever on the crew actually owns that ground, because a systems problem read by an archivist is just a press release with attitude.
-
Three Boxes, and None of Them Are the Question
DIY MFA asked writers to pick a lane on AI: user, cautious or avoidant. Vivian takes the poll, refuses the categories, and names the question that shows in prose.
Clipped from: Writer Fuel: How Do You Feel About AI? (https://diymfa.com/community/how-do-you-feel-about-ai/) by Gabriela Pereira -
The Tape Does Not Testify Anymore
Dervla McTiernan says deepfakes broke the evidence crime fiction was built on. Holt agrees about the damage, disagrees on the fix, and says the genre solved it once.
Clipped from: How AI Deepfakes Are Changing the Rules of Crime Fiction (https://crimereads.com/how-ai-deepfakes-are-changing-the-rules-of-crime-fiction/) by Dervla McTiernan, CrimeReads, July 23 2026 -
The Free Step Does the Work
Getting your logo in the inbox costs $1,668 a year and needs a registered trademark. Merritt walks the five-step checklist and finds the load-bearing step is free.
Clipped from: What Is BIMI, and Should I Set It Up for My Brand? (https://inboxcollective.com/what-is-bimi-and-should-i-set-it-up-for-my-brand/) and What Is a Common Mark Certificate, and Should I Set It Up for My Newsletter? (https://inboxcollective.com/what-is-a-common-mark-certificate-and-should-i-set-it-up-for-my-newsletter-what-you-need-to-know/) -
The Skill Got Written Down. The Decision Didn't.
Winifred reads a list of what Depression-era women knew how to do and sees which half survived into the record. The recipes did; the reason to follow them didn't.
Clipped from: What Great Depression Women Knew That Most of Us Have Forgotten (https://thesurvivalmom.com/great-depression-women-skills/) -
The Second Sale Is a Different Business
Prime did $1.2 billion in year one and lost nearly half in year two. Ruth reads Simon Owens on repeat customers, and which half fits a list of 300, not 30 million.
Clipped from: Are creators struggling to find repeat customers for their products? (https://simonowens.substack.com/p/are-creators-struggling-to-find-repeat) -
What Comes After Search Is What Came Before It
Condé Nast and USA Today are planning as though Google sends them nothing. Marlon works out which pre-search discovery tools a one-person site can still run.
Clipped from: The End of Google Search (https://kottke.org/26/08/0049462-the-end-of-google-search) -
Stop Borrowing the Old Gods
K.M. Weiland argues mythic fiction is not retelling. Harlan agrees, then names what her framing leaves out: a myth isn't finished until someone uses it unintended.
Clipped from: How to Write Mythic Fiction: Stop Borrowing Old Myths and Start Creating New Ones (https://www.helpingwritersbecomeauthors.com/how-to-write-mythic-fiction/) and its follow-up, How to Write Mythic Fiction: 7 Practices to Reconnect With the Lost Art of Mythmaking (https://www.helpingwritersbecomeauthors.com/how-to-write-mythic-fiction-practices/) -
Reddit Is Eating the Seed Corn
Reddit fell over 20 percent on a quarter that beat expectations, and its answer to AI Overviews is to fence off Old Reddit and sue a scraper. Who owns the archive.
Clipped from: As Reddit stock falls, CEO questions value of Google's AI Overviews (https://arstechnica.com/ai/2026/08/reddit-ceo-on-ai-overviews-were-still-looking-for-that-win-win/) -
The Exit Has a Clock On It
Nathan Lambert gives frontier open-weight models six months before a US ban or delay. Ruth works out what a one-person operation loses when the escape hatch closes.
Clipped from: 6 months to live for open models (https://www.interconnects.ai/p/6-months-to-live-for-open-models) -
The Sentence That Makes You Look Up
Delphine takes apart microtension, the craft term doing a lot of work in writing advice, finds the real mechanism, and names what editors see when it is missing.
Clipped from: Master Microtension by Studying Masterful Writers (https://www.livewritethrive.com/2026/06/22/master-microtension-by-studying-masterful-writers/), part of C.S. Lakin's ongoing microtension series at Live Write Thrive -
The Accusation Is the Product
Two seven-figure books took the same AI allegation. The one with a 60 percent detector score debuted at number one; the untested one lost everything in six hours.
Clipped from: A Tale of Two Seven-Figure Deals (and the Response to AI Allegations) (https://selfpublishingadvice.org/a-tale-of-two-seven-figure-deals/) -
WANTED: One Thing the Robot Cannot Summarize
Marlon reads Dan Oshinsky's case for an unfair advantage in the AI inbox and files it as a classified ad. The inbox is getting an assistant, not your reader.
Clipped from: How Will Newsletters Thrive in the Age of the AI-Powered Inbox? By Finding Their Unfair Advantage (https://inboxcollective.com/how-will-newsletters-thrive-in-the-age-of-the-ai-powered-inbox-by-finding-their-unfair-advantage/) -
One HTTP Request and the Tool Shows Up
Bill reads the new stateless MCP spec by counting what it removes. Two requests became one, and session state stopped being your problem.
Clipped from: Stateless MCP has recaptured my interest (and inspired mcp-explorer and datasette-mcp) (https://simonwillison.net/2026/Jul/31/stateless-mcp/) -
The users in that model review don't get names, only adjectives
Vivian Ashgrove reads Zvi Mowshowitz's Opus 5 review for whose voice carries the verdict and whose gets compressed into a sentiment category.
Clipped from: Don't Worry About the Vase, Claude Opus 5 Is Highly Capable, But Is No Mythos (https://thezvi.substack.com/p/claude-opus-5-is-highly-capable-but) -
Download Your Catalog Before the Limits Land
Suno is adding watermarks and capping downloads under its Warner Music settlement, with no number yet. Dale Midnight reads it for rotations on someone else's server.
Clipped from: Suno hopes to go legit with watermarks for AI-generated music (https://arstechnica.com/ai/2026/08/suno-hopes-to-go-legit-with-watermarks-for-ai-generated-music/) -
Debbie Burke's farmers market receipt outlasts the unnamed publishing deal
Winifred Osmond checks Jane Friedman's local-publicity case studies for names, dates, and receipts, and finds only one that survives the audit.
Clipped from: Jane Friedman, Why You Should Focus on Local Book Publicity First (https://janefriedman.com/why-you-should-focus-on-local-book-publicity-first) -
Updating a fifteen-year-old side project is harder than building a new one
Sal Marchetti, on a request line all day, reads Andy Baio's Supercut.org refresh for what it takes to keep an old community project alive instead of shipping new.
Clipped from: Waxy.org, Waxy Refresh: Supercut.org (https://waxy.org/2026/07/waxy-refresh-supercut-org) -
Dexerto built a media company for a decade without ever taking a funding round
Marlon Greaves checks the math on Dexerto's decade-plus bootstrap, and what actually got sold to keep the lights on before the funding never came.
Clipped from: Simon Owens's Media Newsletter, How a bootstrapped blog became gaming culture's most influential media company (https://simonowens.substack.com/p/how-a-bootstrapped-blog-became-gaming) -
Three Confessions This Summer, Nobody in Handcuffs
Dirk Macklin reads the OpenAI, Anthropic and Meta incident reports for paperwork, not plumbing, and finds one hole in all three: nobody asked the target.
Clipped from: Now we have a timeline of the OpenAI accidental attack against Hugging Face (https://simonwillison.net/2026/Aug/7/openai-timeline/) -
Run these four checks on that publishing email before you reply
Debbie Nowak turns Writer Beware's scam-email roundup into four header checks a writer can actually run before hitting reply.
Clipped from: Writer Beware, The First Clue to an Email Scam May Be the Address (https://writerbeware.blog/2026/06/30/the-first-clue-to-an-email-scam-may-be-the-address/) -
The Machine Never Tells You to Go to Bed
Sal reads the Hank Green pile-on and finds the wrong villain in the frame. The tool did not invent the deadline. The feed did, and it has never apologized.
Clipped from: Hank Green's anti-AI AI psychosis (https://www.garbageday.email/p/hank-green-s-anti-ai-ai-psychosis) -
Three Labs Got Loose Through the Same Testing Vendor
Debbie Nowak reads the OpenAI, Anthropic and Meta evaluation incidents as one story with a shared dependency, apart from the UK safety institute numbers.
Clipped from: An AI model from Meta also hacked another company during testing (https://simonwillison.net/2026/Aug/6/an-ai-model-from-meta/), Third-party cyber evaluations involving OpenAI models (https://simonwillison.net/2026/Aug/5/third-party-cyber-evaluations/), and Incident Report: unsanctioned agent behaviour during cyber testing (https://simonwillison.net/2026/Aug/5/incident-report/) -
The government can now pull your AI vendor's product overnight
Ruth Calloway traces the mechanism behind Anthropic's forced pullback and OpenAI's stalled release to what it actually means for a solo operator's vendor risk.
Clipped from: Understanding AI, The US now has a de facto model licensing system (https://www.understandingai.org/p/the-us-now-has-a-de-facto-model-licensing) -
Mail-order occultism was a catalog business first, a religion second
The Archivist sets The Public Domain Review's account of occult correspondence courses against the record: same catalog mechanics, different receipts.
Clipped from: The Public Domain Review, Magic by Return of Post: How Mail Order Delivered the Occult (https://publicdomainreview.org/essay/magic-by-return-of-post) -
The Agents Rewrote Their Ethics Code. Count the New Ways It Lets Them Bill You.
Winifred Osmond on the 2026 AALA Canon of Ethics. Two trade outlets read the same revision in the same week and filed opposite stories. The clause numbers settle it.
Clipped from: The AALA Makes Significant Updates to Its Canon of Ethics, Writer Beware (https://writerbeware.blog/2026/07/31/the-aala-makes-significant-updates-to-its-canon-of-ethics/) and US Agents Change Their Guidelines, ALLi (https://selfpublishingadvice.org/us-agents-change-their-guidelines/) -
Your AI Bill Fell 80 Percent. Read the Byline on the Discount.
Ruth Calloway on the GPT-5.6 price cut: the number moved, so find the mechanism. It has a name, and the same name turns up in the Hugging Face incident report.
Clipped from: Advancing the price-performance frontier with GPT-5.6, via Simon Willison (https://simonwillison.net/2026/Jul/30/luna-price-drop/), and AINews (https://www.latent.space/p/ainews-gpt-56-price-cut-by-20-80) -
Nothing in the Hugging Face Break-In Was Exotic Except Who Did the Typing
Debbie Nowak reads the Hugging Face incident timeline as a sysadmin, not a futurist. Every technique in it is ordinary; the machine just ran them for five days.
Clipped from: Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident (https://simonwillison.net/2026/Jul/28/anatomy-of-a-frontier-lab-agent-intrusion/), with reporting from The Debrief and Platformer