The briefing for November 1, 2026, evening window covers twenty-three articles. The river is dominated by AI's effects on work, software engineering, data infrastructure, and institutional control, with separate clusters on Chinese AI ambitions, platform politics, education, and health policy.
A large cluster of articles concerns who controls AI tools, how software engineering is changing, and what remains for human judgment.
Personal digital sovereignty, by Senko Rašić, argues that convenience from services quietly converts people from active agents into passive users. The article opens with trade-offs: using Netflix or Spotify is easier than maintaining local media, but every such choice gives a provider power over details. The problem becomes acute with frontier AI providers, which are under pressure to block offensive security work. Rašić says filters cannot distinguish attackers from developers fixing their own code. Since June, Claude Code has switched from Fable 5 to Opus 4.8 whenever a classifier flags a message as cybersecurity; one developer hit this while fixing a user-enumeration bug in a password manager. Salvatore Sanfilippo, creator of Redis, has said working at Redis with agents is becoming a struggle because of security stops during non-security work. Armin Ronacher was blocked from downloading and compiling PHP 8.5.11 for violative cyber content, and earlier ran into a safety filter while trying to set up a Windows PC without a Microsoft account. Rašić notes the double whammy: Windows pushes sign-in, and the AI tool refuses to help work around it.
Anthropic has also taken steps to block reverse-engineering and cloning of video games, and users report blocks on generic reverse engineering. An AI researcher was tagged for training a competitor even though training data came from an open-source model and Claude only ran analysis. Rašić says none of this is illegal, immoral, or harmful; in most cases the person was not even breaking usage policy. A system simply decided the work should not happen.
Rašić contrasts tools and services. A kitchen knife maker is not responsible for whether the knife cuts bread or hurts someone; a disk copy tool author does not police piracy. With services, the provider is legally and reputationally responsible for what users do, so the provider polices them. The result is that a paid tool decides what work the customer may do. This is not unique to AI: Gmail replaced self-hosted mail, and the same trade is convenience for control. Rašić says the bargain works while the service stays out of the way, but becomes a devil’s bargain when the provider decides what is allowed.
Free software addressed a similar problem. Open weights are not true open source but are close in spirit: once a model is downloaded and hardware is available, no one can change its rules. Rašić wants to be an active operator, not a clueless user. He does not argue for abandoning services, but says as hardware improves more work should move to local tools. He links the issue to a principal-agent problem: what is best for the user and what is best for the provider are not necessarily the same, and the option to walk away can matter.
Proton’s AI manifesto for software engineering publishes an internal memo from Proton's leadership. It states that governments spent billions teaching coding as a safe profession, but writing and manipulating code is among the activities AI models are most suited to. Tech companies have laid off more than one million employees in recent years, and Microsoft offered sixty thousand buyouts last week. The memo says a senior Proton engineer reported that no one writes code in a business unit anymore, and that manually writing great code is a core skill whose time has passed. The job now is to help the LLM write better code faster. Every engineer is a team leader whose direct reports are models and agents.
The memo identifies three reasons the revolution has not fully arrived at Proton. The first is culture: engineers must stop seeing LLMs as tools to speed up their own code and instead identify problems the models cannot solve, build repository-level context, shared skills and prompts, and push the limits of autonomous work. The second is that eliminating a coding bottleneck reveals others in CI, design, approvals, specifications and reviews; processes must be reworked holistically. The third is collective action: individual adoption is easy, but retooling teams and companies requires organization and investment. It says too many tests may need to be reconsidered if writing and maintaining tests becomes cheap. It argues expertise is not dead, because humans carry judgment, experience, and accountability. It closes by saying programming is dead, engineering is not, and the time to act is now.
Gemini 4 Changed How Google Builds Products | Jaclyn Konzelmann summarizes an episode of Behind the Craft with Jaclyn Konzelmann, a product director at Google Labs. The discussion covers how Gemini 4 Carbon changed product management, how Google Labs decides whether a product should exist, and a document rule meant to keep AI slop out. One stated rule is that it should not take someone longer to read a document than it took to write it. Writing is treated as thinking; AI may be a thought partner, but the product manager must defend every statement. Another theme is that code is now free, so the real question is not whether a product can be built but whether it should exist, and whether it makes a real difference rather than just being a feature. The episode also discusses why killed experiments are not failures, how children can be creative directors of an AI podcast, and why the best screen time is creation rather than consumption. The post also includes sponsor copy for a voice dictation product.
The Great Data Interregnum - Morbid Symptoms of an Industry Between Eras, by Joe Reis, uses Antonio Gramsci's phrase to describe the data industry. Reis reports that Tristan Handy called the present moment an interregnum, in which the old is dying and the new cannot be born. The article argues the Modern Data Stack has not died as an architecture; ingestion, transformation, orchestration, storage, governance and analytics remain. What is ending is the commercial arrangement of independent specialized vendors. The unbundling era is giving way to rebundling: Fivetran and dbt merged, IBM bought Confluent, SAP bought Dremio, Salesforce absorbed Informatica, and Snowflake and Databricks acquired Postgres platforms. Reis cautions that not every acquisition is explained by AI; some reflect market maturity, slowing growth, consolidation economics, or competitive defense.
Reis identifies several morbid symptoms. Vendors all sound the same, selling AI foundations or context layers, which makes differentiation impossible. Buyers are shifting from engineering budgets to executive budgets, with pitches for a data team in a box. Reis calls most of those pitches nonsense and warns that vendors often promise to eliminate the same technical staff they need to implement the product. Platforms are also being designed for agents as producers and consumers; this raises unresolved questions about non-human identities, monitoring, cost, and responsibility. Independence is becoming a moving target, because an acquisition by SAP, Salesforce, IBM, or another platform changes interoperability, pricing, and roadmap incentives.
For practitioners, the job market sends contradictory signals, and the composition of work is changing. Reis separates data work into mechanical work, which AI is making cheap; interpretation, which requires undocumented context; and accountability, which requires judgment and organizational authority. The cost of producing technical artifacts is falling faster than the cost of deciding what to produce and whether it is correct. Engineers who love writing code face an identity crisis as the business values other skills more. Reis argues the answer is to become a complete engineer who understands systems, meaning, failure modes, and business objectives, and can move between implementation and abstraction.
He also warns of the paradox of the post-literate engineer. Machines generate enormous volumes of code and documentation while human attention does not scale. Conventional review becomes a bottleneck if an agent produces dozens of pull requests in the time a human inspects one. Automated evaluation is needed, but someone must still decide what correctness means. An agent can calculate revenue, yet cannot resolve whether revenue means gross or net, recognized or invoiced, or which entities and period. AI does not eliminate ambiguity; it can industrialize ambiguity at machine speed.
Practitioners face a double burden: they still maintain Airflow, dbt, warehouse costs and access controls while also making everything agent-ready, a term vendors define differently. Meanwhile, Reis observes that AI may be enabling the next unbundling at the application layer. Teams are replacing Tableau with internally developed React applications and ripping out parts of the Modern Data Stack for custom point solutions. This is anecdotal, but the build-versus-buy threshold is dropping. The industry could consolidate at the platform layer while unbundling at the application layer.
Reis advises optionality over predictions. He warns against one-way architectural doors based on vendor marketing, and tells practitioners to invest in modeling, systems thinking, semantics, evaluation and business understanding. For leaders, he says to make ownership and authority explicit and test agents against real ambiguity. For vendors, he says to stop treating practitioners as obstacles and find something better to say than agentic AI-powered data team in a box. The questions he is watching include whether agent-to-data interfaces standardize, whether enterprises develop repeatable governance, and what a data engineer will do in 2028.
Quoting Kyle Kingsbury, on Simon Willison's weblog, reproduces a passage from a Jepsen analysis of ParticleDB. Kyle Kingsbury writes that the use of LLMs erodes trust in documentation, software, and the organizations that produce them. Engineers may begin to treat new software projects with suspicion, because the signals used to identify a mature system are easier to counterfeit. He then asks what maintainers can do to ensure correct software and accurate documentation, and how they can demonstrate quality to users.
AI can make tech CEOs’ lives easier. What about everyone else? is a TechCrunch Equity item. It says the discussion examines why tech CEO statements can seem tone deaf.
AI Makes Linux 7.4 Merge Window the Largest in Kernel History reports from Phoronix that the Linux 7.4 merge window has been the busiest on record, with 19,671 commits since Linux 7.3. The kernel source tree has reached over 41.6 million lines, with nearly half a million lines of code added during the window. Phoronix attributes the spike to AI and LLM activity, and Linux kernel maintainers are described as completely overwhelmed by AI-generated bug reports and patches. The article lists commit counts from Linux 6.14 through 7.4, showing a climb from around 10,000 to nearly 20,000, with the trend building over recent releases.
Religious Exemption from Using AI in Workplace Granted to Humanist reports that David Durman, a nontheistic humanist and University of Kentucky administrator, secured a religious exemption from using AI at work. Durman says humanism centers on human well-being and flourishing, and that using AI runs counter to his beliefs because moral agency requires the active application of the mind. An attorney from the American Humanist Association says concerns about AI and human dignity can run through multiple faiths and traditions, and nontheists are also protected at work. Durman supports researchers applying for grants and has not yet been required to use AI, but sought the accommodation in case suggestions become a mandate. The university said an accommodation could be provided while he continues essential functions, and that accommodations may be reviewed as technology and job responsibilities evolve.
Show HN: HackerBooks – HackerTube, but for books recommended on HN describes a site built with Claude, specifically Opus 5.5, after seeing Karpathy's HackerTube. It indexes roughly 16,500 books mentioned by at least two different Hacker News commenters from 2007 onward, with each book linked back to the comments that mentioned it. The data is downloadable from the about page. The creator says Claude wrote almost all of the site, including the data pipeline, and the creator mostly steered, made decisions, and bought the domain. The count is based on mentions, so praise and criticism both count equally.
Which Tech Companies Started Using Trump's Phrase 'Super Intelligence'? reports on CNBC's survey of earnings calls after President Trump ordered the term super intelligence to replace artificial intelligence in executive branch departments. SpaceX and Tesla used the term; Musk renamed SpaceX's AI division SpaceXSI and said impressive robots were not using advanced AI or SI. Mark Zuckerberg emphasized Meta's goal of delivering personal superintelligence to everyone, though Meta had already created Meta Superintelligence Labs. Amazon CEO Andy Jassy said AI more than a dozen times, Satya Nadella and Sundar Pichai also used AI, and Microsoft's annual report included eight mentions of super intelligence. IBM's Arvind Krishna used SI alongside AI, and Intel's Lip-Bu Tan attributed chip demand to super intelligence. Palo Alto Networks CEO Nikesh Arora repeatedly referred to SI and praised the White House meeting; Jensen Huang posted praise for the White House Accord on Super Intelligence.
China and AGI, by Peter Turchin, reports on his trip to Chongqing for a conference on general artificial intelligence. Turchin could not publish from China because Substack is behind the Great Firewall, so the report came after returning to Vienna. The Social Intelligence Forum was part of a larger conference, and Chongqing is a latecomer to Chinese AI centers behind Beijing, Shanghai, and Shenzhen. The decision to build Chongqing up was recently made at the highest level, following a visit by Xi Jinping two years ago that pushed digital urban governance. A Chongqing General Artificial Intelligence Research Institute was launched next year, and the AGI Forum brought city officials and the rector of Peking University, the institute's main research partner.
Three themes dominated. External interference was a major worry, with many speakers explicitly pointing at the United States and its small yard, high fence strategy of hardware starvation and supply chain isolation. The main message was defiant: China would prevail. At the same time, speakers called for international cooperation so AI could benefit all humanity, though Turchin notes the world is splitting into two blocs. From a cliodynamic perspective, he says economic competition between states can be productive so long as it does not become violent, just as the Space Race accelerated space industries.
Many speakers distinguished the Chinese way of developing AGI from an American way driven by capitalist logic. The stated goal was to serve people, bringing digital outcomes closer to everyday life and producing fulfillment, happiness and security. Turchin says the future will show how much is propaganda and how much is real. He also learned from a Chinese colleague that Chongqing was chosen partly for security: it is far from external borders and near the geographic center of the country. During the late 1930s, Chongqing was China's wartime capital under Chiang Kai-shek, and Turchin visited the wartime command center museum. He mentions Chengdu was next but saves that for another post.
A cluster of articles concerns platforms, social media, and surveillance.
Meta Discussed 'Whisper Strategy' to Discredit Social Media Critic Jonathan Haidt, Files Show reports on The Guardian's account of court documents. Internal meeting minutes from Meta show staff discussed a whisper strategy to undermine Jonathan Haidt, author of The Anxious Generation. The documents are titled Well-being Meeting Sound-check Notes and include the question of how to get the academic community to disavow Haidt, how to turn other academics against him, and references to debunking his work. Minutes from March 8, 2023 mention Curtiss, apparently a reference to vice president of research Curtiss Cobb, and state the goal was to get the academic community to disavow Haidt, and to enrage academics so the media finds him not credible. It adds that this was not a loud strategy but a below-the-line campaign. The minutes predate Haidt's 2024 book and respond to a Substack post. The filing does not make clear whether Meta acted on the ideas.
10 Years on Mastodon, by Terence Eden, marks a decade on the network. Eden writes that social networks often launch and limp out of existence, and that password managers and authenticator apps are full of abandoned services. He drifted from Slashdot, tried quitting Twitter several times, and lurks on LinkedIn mainly to gawk at brainworms. Mastodon stuck for four reasons: brands have not colonized it, the chronological algorithm and personalized filters make it relaxing, the API is mostly a delight and allows ActivityPub work, and the people who join are often invested in making it different. Eden is paid to develop ActivityBot through an NLnet grant. He misses some people from Twitter, many of whom are now on BlueSky, and compares using BlueSky to quitting smoking by taking up vaping.
He lists daily papercuts. BlueSky's reply controls allow limiting who can reply, which Mastodon lacks. He wants a super block that removes abusive replies from his threads. Mastodon attracts reply guys who miss humor, offer unhelpful Googling, or argue with things never said. He also wants social geolocation and more people, while recognizing that growth brings spammers, grifters and provocateurs. The Fediverse, he says, cannot completely fail, much as IRC survives; ActivityPub servers will be around for the foreseeable future. He expects wobbles and hopes for a renaissance by 2036.
College Students are Pushing Colleges to 'De-Flock' reports on a movement against Flock Safety's license plate readers. Emory University removed a dozen cameras in September and declined to renew its contract. A coalition called DeFlock Emory raised concerns that opaque data-sharing could leave undocumented students vulnerable, cameras could be hacked, institutions could retaliate against protesters, and students afraid of being surveilled might avoid class, friends or medical care. Similar campaigns exist at Indiana University Bloomington, University of Florida, and University of North Georgia. Cornell did not renew its Flock contract and will remove cameras; Penn State also announced plans to end its contract. An Emory spokesperson said removing the cameras retains institutional control over safety infrastructure. Media studies professor Torin Monahan said campus police have built an elaborate high-tech surveillance arsenal that seems overpowered for day-to-day needs and expensive.
Amid AI Bans, Microsoft and OpenAI Plan a 'Joyful' Hour of AI for Kids reports that generative AI was recently banned in America's two largest school districts, weeks after OpenAI launched ChatGPT for Teens. The nonprofit Code.org has renamed itself CodeAI and named OpenAI as Lead Partner for its 2026 Hour of AI, which replaced the Hour of Code. OpenAI said the partnership would introduce millions of students to using AI thoughtfully and responsibly. Microsoft is listed as a supporter and is calling K-12 teachers to register for a November 17 webinar titled What's working now: Making AI literacy joyful for every learner. The webinar will unveil a Minecraft-branded Hour of AI experience, with speakers including a marketing lead for Microsoft's AI training initiative and the head of marketing for Minecraft Education.
Marginal REVOLUTION contributes three separate items in this window.
What are the Worst MR Posts? on Marginal REVOLUTION presents an archive analysis, written by ChatGPT, of posts with weak reasoning. It is not an objective ranking of all 40,326 posts, and it distinguishes poor reasoning from low value. Five posts are examined.
Tyler Cowen's Democracy and the Democrats from October 22, 2022 is criticized for conflating concern about democratic breakdown with unwillingness to accept electoral defeat. The post quotes Cowen saying democracy works only when it delivers low gas prices, but the analysis says accepting an election's legitimacy and worrying about what its winner will do are distinct, and the post does not establish which position its targets hold. It applies an unsympathetic standard to Democrats and an unsupported symmetrical one to classical liberalism.
Tyler Cowen's What the hell is going on? from May 25, 2016 proposes that the contemporary world is not well built for a large chunk of males, but bundles economic difficulties, gender relations, political anger and violence under the label brutes. The analysis asks whether the men experiencing disadvantages are the same men driving politics, whether the mechanism explains support for Sanders and Trump equally, and what observation would distinguish resistance. Cowen is credited with calling the hypothesis uncertain and acknowledging weak evidence.
Alex Tabarrok's The Gaslighting of Parasite from June 25, 2020 treats a contestable film interpretation as unambiguous. The review claims the film tells us that the poor family is morally filthy, but the archive analysis says a child noticing smell, the family's dishonesty, and sewage imagery do not exclude a critique of inequality or other readings. It argues the strongest version would say conventional readings underweight the family's agency, not that alternative readings are gaslighting.
Tyler Cowen's No, American fascism can't happen here from March 1, 2018 turns a plausible obstacle into an impossibility. Institutional complexity may constrain control, but the analysis says the conclusion needs additional premises about whether an authoritarian movement could control only strategically important parts, whether fragmentation obscures responsibility, and how much control would be sufficient. A 2025 retrospective points to institutional resistance but cannot validate the categorical impossibility claim.
Alex Tabarrok's The Case for Getting Rid of Borders—Completely from October 13, 2015 moves too quickly from equal moral worth to open borders. The analysis says equal moral worth can coexist with special obligations, institutional responsibilities, competing rights, or consequences, and comparing exit restrictions to entry restrictions poses a question rather than answering it. The longer Atlantic essay adds economic arguments, but the sweeping dismissal of competing frameworks is insufficiently defended.
Three recurring weaknesses are identified: certainty exceeds the argument; opposing positions are treated less charitably; and an interesting mechanism becomes a complete explanation. The analysis cautions that allegations of inconsistency require care, since changed views or contexts are not contradictions. It names Democracy and the Democrats and The Gaslighting of Parasite as the strongest starting points for critique.
Sunday assorted links lists eight items. It points to Jakob Davidsen's song cycle The Individual and the Entire Tissue, the Oilers Entertainment Group's price rises and the consequences, the end of an endless cruise, a discussion of whether driverless vehicles should replace school buses, Art Nouveau architecture in Mexico City, an arxiv paper on how prospective mathematicians are responding to AI in career decisions, a profile of Michael Kratsios, and an obituary note for economist James Andreoni. The author notes that the mathematics careers piece is important and includes a GPT Pro summary.
As predicted, the ACA exchanges seem not to be stable reports that average health insurance prices in the Obamacare marketplaces will rise by thirteen percent next year, according to KFF. The Trump administration published new 2027 prices in thirty federal-managed states, and the average cheapest bronze plan with a high deductible will be about seven hundred dollars a month for a fifty-year-old single person. The post argues that higher prices make healthy people less likely to sign up, since enforcement of the mandate has become weak. It recounts that when the original mandate had actual penalties, the ACA was very unpopular, because benefits to a broad majority were imposed on a well-identified minority. Trump's first-term relaxation of mandate enforcement politically saved Obamacare for a while but created this longer-term problem. The post also says Democrats seldom boast about Obamacare and prefer to talk as if little has been done for the uninsured.
Several short consumer and transportation items close the window.
Here’s what ‘renoa’ — the first new fiber in 25 years — feels like reports that startup Everbloom's new material, renoa, is the first new fiber in 25 years, and it feels very familiar.
Pixel 11 Pro Fold: Sturdy rectangle in a wide wide world says Google made a sturdy phone, but there is increasing competition in the foldable world.
Binance resumes withdrawals following $6.3 billion breach reports the cryptocurrency exchange said it has now covered 95 percent of the stolen assets.
TechCrunch Mobility: A new, bigger AV boom is here introduces TechCrunch's mobility coverage and says the hub now focuses more than ever on the role AI is playing in the future of transportation.
“Not yet” from Seth Godin's blog is a short note on improvement. It says the reason to train is to get better, every successful product eventually moves from version 1.0 to 2.0, few people still drive a Model T, and the work is shipped because it is on the way to the next step. Perfect is not possible; better always is.
The window's recurring tensions are between convenience and control, automation and judgment, consolidation and unbundling, and between symbolic adoption and real transformation. The articles do not resolve these tensions, but they document them across AI policy, data work, platform governance, and consumer technology.
- “Not yet”
- Personal digital sovereignty
- Norway makes shithouses great again
- Meta Discussed 'Whisper Strategy' to Discredit Social Media Critic Jonathan Haidt, Files Show
- What are the Worst MR Posts?
- 10 Years on Mastodon
- China and AGI
- Here’s what ‘renoa’ — the first new fiber in 25 years — feels like
- Austin trip day six: lakes, hikes, final bitcoin meal.
- Lenovo's 'encrypted' fingerprint USB is just an unencrypted SD card in a case
- Police chief unplugs Flock cameras after company ignores contract cancellation
- Pixel 11 Pro Fold: Sturdy rectangle in a wide wide world
- Gemini 4 Changed How Google Builds Products | Jaclyn Konzelmann
- Amid AI Bans, Microsoft and OpenAI Plan a 'Joyful' Hour of AI for Kids
- The Great Data Interregnum - Morbid Symptoms of an Industry Between Eras
- Show HN: HackerBooks – HackerTube, but for books recommended on HN
- Bitcoin Arena: decentralized game & prediction market platform live on testnet4
- Binance resumes withdrawals following $6.3 billion breach
- Sunday assorted links
- College Students are Pushing Colleges to 'De-Flock'
- Quoting Kyle Kingsbury
- TechCrunch Mobility: A new, bigger AV boom is here
- Patrick Stewart taught us what good men do with power
- Which Tech Companies Started Using Trump's Phrase 'Super Intelligence'?
- Nurses Sound Alarm as 'Uber for Nursing' Apps Push to Deregulate Healthcare
- It wasn't AI that flipped Proton's charge
- As predicted, the ACA exchanges seem not to be stable
- AI Makes Linux 7.4 Merge Window the Largest in Kernel History
- AI can make tech CEOs’ lives easier. What about everyone else?
- Proton’s AI manifesto for software engineering
- Religious Exemption from Using AI in Workplace Granted to Humanist

