Everybody can complain now
The Guardian reports that MPs' offices, councils, courts, schools and employers in the UK are swamped by complaints, appeals and requests written with AI. "Half the emails we receive are some form of AI," says Hollie Wickens, a caseworker in a Labour MP's office. Council complaints used to average about two pages. Most are now ten times longer, says the head of Lawyers in Local Government, and the worst one she knows of runs to about 200 pages. Researchers call it "agentic flooding." And many of these letters, one expert says, forget to mention what the person actually wants.
This is just fun. Finally everybody can complain effectively about everything. Inaccurate AI requests still have to be answered by humans, though, so somebody is reading those 200 pages.
The backlash continues to grow
In the US midterms, 38 of 44 ballot measures in 10 states that sought to restrict data centers had passed so far, according to Redbud Consulting, as Ars Technica reports. In an NBC News poll on Election Day, 76 percent of voters said they opposed data centers in their communities. The tech industry's explanation, per Ars: the backlash is "anti-Trump," not "anti-tech". Yeah, sure. The backlash continues to grow, as I wrote in No. 4.
https://www.theverge.com/policy/1026365/data-centers-midterms-results
The same elections also saw more than $187 million spent on nearly 300 ads generated or enhanced with AI, according to the Wesleyan Media Project, and most carried no disclaimer. In Australia, Victoria's major parties are trading blows over AI-generated videos of each other's leaders. Unsurprising development. This must be the glorious future we have been promised.
AI, speeding up AI
METR, an independent evaluator, spent almost three months inside Anthropic with deep access. Its conclusion: Anthropic is likely improving its models on well-specified tasks at about twice the rate of March 2024 to October 2025, and METR "would not be surprised" by anything from 1.3x to 3x. It guesses that about 80% of today's speedup comes from AI itself. METR also says AI is "not yet able to make highly significant and fully autonomous contributions" to the most important parts of model training, though researchers described "signs of life." And Anthropic redacted "a majority of the evidence we found most informative."
https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/
Big story. To me it seems like they reached recursive self-improvement, AI that speeds up its own development. But Anthropic is trying not to make everybody freak out. Ajeya Cotra's essay on the report is worth reading next to it.
https://www.planned-obsolescence.org/p/ai-is-dramatically-accelerating-ai
Apple is concerned too, which is probably also unsurprising. In its annual filing it warns that AI agents "may behave in unintended or unpredictable ways, including by taking actions that were not directed, instructed or authorized," and that its ability to control third-party agents "may be limited." No shit, sherlock.
At the same time, Bloomberg's Mark Gurman reports (via The Verge) that CEO John Ternus is pressing Craig Federighi to turn Siri into a real agent that can compete with OpenAI's Dots and Meta's Muse. Control over Apple's own apps is expected with iOS 27.3, third-party apps likely in the spring. So Apple is doing its usual thing, copying promising software and integrating it directly.
https://www.theverge.com/ai-artificial-intelligence/1029552/apple-siri-ai-agentic
Meta added Spaces to its Muse agent. In Mark Zuckerberg's words: "tell Muse what you need and it builds it." The examples: a BBQ game, an event tracker and a home control dashboard. Simple software is losing worth by the day.
Designers need a say
Nielsen Norman Group's Caleb Sponheim explains the "agent harness," the software loop between an AI model and its tools. Its context, permissions and limits decide what users can do, so the article tells UX people to specify that behavior alongside the screen and to work with engineers on defaults and permission dialogs. Yes, we need to have a say here.
https://www.nngroup.com/articles/agent-harness/
Noman calls skills "the apps of the agentic era," with two users: the person who installs one and the agent that runs it. Words like "clean" or "minimal" didn't get far with the agent. "You're designing an app with nothing but words." Important and underrated, imho.
https://iamnoman.com/writing/designing-a-skill
On UX Collective, Dolphia imagines an AI-designed success state with a small burst of confetti. It's flawless, and it appears both for someone who finally launched their business and for an administrator updating the forty-seventh record of the morning. Michał Krztoń asks who designs the stop button, and suggests marking every step of a workflow as deterministic, fuzzy or irreversible. Interesting thoughts here.
https://uxdesign.cc/ai-can-generate-the-interface-what-makes-it-worth-caring-about-a5f3a3e81986
https://uxdesign.cc/your-ai-is-non-deterministic-but-your-product-cant-be-3ec2a0cf870c
Mathieu Poli argues that a polished look now costs one prompt, so native components matter more: they keep an app from looking fake, and they move with the system. A year after iOS 26, Poli finds, AI models still mostly draw the old iOS. He maybe onto something here.
https://uxdesign.cc/native-is-a-floor-not-a-ceiling-09bc4f27ddce
Patrick Neeman goes back to Marcia Bates's 1989 paper on "berrypicking": real searches wander, change shape and collect value in pieces. The old systems assumed one query and one best result. "The chat box makes the same assumption, only faster." This is valuable insight.
Figma's agent now makes vector illustrations and icons with grouped layers and strokes instead of fills, so you can change the stroke weight without redrawing. Pen.dev has been able to do that for months, so nice that Figma finally managed to have something here too.
https://www.figma.com/release-notes/
Keeping your head
Sergey Kuhne describes what running many coding agents at once does: 20 threads, all those contexts in your "brain RAM." The fix is a queue of tickets in git, an agent that builds only what's tagged as safe, and one "owner inbox" with everything that needs a human decision. One item, one decision, then close it. "It is still exhausting. But I don't feel distracted or lost." Highly valuable insight into something I personally struggle with, the toll I wrote about in No. 3.
https://sergeykuhne.com/blog/fighting-ai-brain-fry
Murat Demirbas on writing as a way of thinking: "You can now produce six pages with no thinking at all, and the six pages still look like evidence that the thinking happened." Which is true.
https://muratbuffalo.blogspot.com/2026/11/writing-as-thinking-tool.html
Also worth a look
Tokenizer Mono
George Mandis built a font based on IBM Plex that draws a colored box around each AI token as you type, with one style for each of 28 tokenizers. The font does it all by itself, no scripting. Love this.
https://george.mand.is/2026/11/the-tokenizer-font/
Making AI chat accessible
Eric Bailey on Smashing Magazine, with concrete patterns for screen readers: announce a reply only once it has finished streaming, and put a hidden heading like "You said:" before each message. Helpful.
https://www.smashingmagazine.com/2026/11/making-immersive-llm-chat-experiences-more-accessible/
Wisp
"A little AI fairy who lives on your desktop." It floats beside your cursor on Mac or Windows, sees your screen, opens apps, arranges windows and replies to email. Sounds interesting. I haven't tried it yet, and the launch page says nothing about what happens to what it sees.
https://www.producthunt.com/products/wisp-9
A lawsuit against ChatGPT
A North Carolina man is suing OpenAI and Sam Altman, alleging that ChatGPT gave his wife detailed instructions for poisoning him with oleander, and that OpenAI didn't stop the chats even after they became "obviously suspicious." She was indicted in October on charges including attempted first-degree murder. OpenAI says it banned the user and gave information to law enforcement. Unhappy wife, unhappy life.
What would you finally complain about, now that it costs nothing? Leave a comment here or reply on Bluesky.