After the Death of Blogs: Major Publishers Are Also Giving Up on the Web

Google AI Overviews Claim First Bauer Sacrifice: The End for Cosmopolitan & Co

The Bauer Media Group is largely shutting down its digital channels in Germany at the end of the year. This affects the online presences of magazines such as Auto Zeitung, Bravo, Cosmopolitan, and Intouch—only TV Movie, Lecker, and Astrowoche will remain online. The publisher is thus burying its digital unit Bauer Xcel Media, founded in 2019, and laying off all 160 employees. In the UK, Bauer is planning similar steps following the loss of about a third of its digital reach and over 100 jobs. Head of Publishing Ingo Klinge justifies the move with the “rapidly changing environment”: tech platforms are swallowing more and more traffic, barely referring users to publisher sites anymore, and AI platforms are using publisher content without permission. An attempt to counteract this with Content Commerce also failed due to declining reach. → horizont.net

Synthszr Take: Bauer is executing a controlled retreat, reminiscent of abandoning colonies when supply lines become too long. The three remaining portals (TV Movie, Lecker, Astrowoche) function like research stations in Antarctica: minimal staffing, maximum efficiency, observation posts in case conditions change. What Bauer is really doing here is a brutal form of portfolio consolidation based on the principle of permaculture: anything that costs more energy than it yields is cut off. The decision reveals a bleak truth: for many publishers, digital is no longer a growth area but a loss-making business where Big Tech reaps the harvest while publishers till the fields. After the mass extinction of AdSense-funded blogs over the last 18 months, it's now hitting the major publishers.

Claude Code Desktop Becomes an Orchestrated Development Environment

Anthropic has fundamentally redesigned its Claude Code Desktop app, positioning it as a command center for simultaneous coding agents. The new version brings an integrated terminal, a side-chat feature for intermediate questions during active coding sessions, an improved diff viewer for large changesets, and flexible windows for preview, plan, diffs, tasks, and terminal. The focus thus shifts from linear chat interaction to a parallel orchestration of multiple agent activities. While OpenAI's Codex has offered a terminal for some time, its absence in Claude Code was a frequent criticism—especially given the rapidly dwindling token budgets with intensive use. The change also affects the general Claude app: the modal switcher moves to the sidebar and is replaced by icons, while the entire app has been rebuilt for better performance and streaming responses. → thenewstack.io

Synthszr Take: Anthropic is transforming Claude Code into a kind of air traffic controller's workstation for code generation. This is reminiscent of the evolution of music production software: in the past, you recorded linear tracks; today, in Ableton Live, you juggle loops, effects, and MIDI controllers simultaneously. The side-chat feature solves a fundamental problem of agentic AI: how do you intervene without interrupting the ongoing process? It's like the difference between a conversation and a conductor's score—you can give instructions while the orchestra continues to play. The fact that Anthropic is launching this update at the same time as the new Routines shows their strategy: Claude should not only write code but also manage code development. The irony is that the better the tools get, the more tokens users burn—a classic example of Jevons' Paradox, where higher efficiency leads to more consumption.

OpenAI Takes Off the Kid Gloves

The rivalry between OpenAI and Anthropic is escalating. In a leaked internal memo, OpenAI's Chief Revenue Officer Denise Dresser accuses the competitor of inflating its revenue forecasts by about $8 billion. Anthropic is a “single-product company in a platform war” and is building its brand on fear. Dresser outlines five strategic priorities for OpenAI's dominance and casually mentions the upcoming model “Spud.” Simultaneously, Meta is working on a photorealistic AI clone of Mark Zuckerberg, trained on his public statements and internal strategic considerations. Meanwhile, the app Skye is revolutionizing the iPhone home screen with a dynamic AI layer that prepares relevant information and triggers actions directly. → Superhuman – Zain Kahn

Synthszr Take: OpenAI is practicing classic expectation management through targeted leaks. The memo serves three functions at once: it positions Anthropic as a bluffer, prepares the market for “Spud,” and signals combat readiness to investors. The timing is no coincidence: while Anthropic is integrating Claude into Microsoft Word, OpenAI needs to show it builds more than just models. The accusation of being a “single-product company” hits a sore spot: Anthropic's entire strategy is based on the assumption that safe AI models are sufficient. OpenAI's platform argument is reminiscent of Microsoft's Windows strategy in the '90s: whoever controls the operating system wins in the long run against specialized applications.

Microsoft's Copilot Crisis: Everyone Has It, Nobody Wants It

Microsoft is facing a precedent in enterprise software: 68% of Copilot users use the tool only because their company mandates it, while a mere 8% would use it voluntarily. This gap between forced adoption and actual preference reveals a fundamental problem in the perception of its output quality. Microsoft is responding with a multi-model approach called “Critique,” which combines GPT and Claude. The question is whether this architectural adjustment is enough to improve the quality perception for critical tasks. A band-aid alone won't close this preference gap if users perceive the product as fundamentally inadequate. → The Business Engineer

Synthszr Take: Microsoft is experiencing what's known in biology as “obligate parasitism”: the product survives only through its forced symbiosis with Office 365. The 8% preference rate is more brutal than any bad app store review because it comes from users who use the product daily. The multi-model approach with Claude is reminiscent of 1960s urban planning, when it was thought that more highways would solve traffic jams. The real problem probably lies not in the model architecture but in the product context: Copilot tries to automate universal office work without understanding that every team has its own working language. Microsoft is betting that users will get used to mediocre AI output. The old rule applies: the worse the coffee in the office, the more Microsoft on the computers.

Anthropic Introduces Automated Routines in Claude Code

Anthropic is enhancing Claude Code with automatable routines that can be triggered on a schedule, via API, or by GitHub events. This allows developers to automate recurring tasks such as processing bug tickets from Linear overnight, reviewing pull requests, or reacting to cloud events. The routines run on Claude Code's web infrastructure, so no local machine needs to be online permanently. Three variants are available: Scheduled Routines (hourly, nightly, or weekly), API Routines (custom endpoint with an auth token), and Webhook Routines (initially only GitHub events). Early users are using the feature for automated backlog management, deployment verification, and custom code reviews. The feature is now available to Pro, Max, Team, and Enterprise users, with daily limits of 5 to 25 routines depending on the plan. → TAAFT - There's An AI For That

Synthszr Take: Claude Code is becoming the janitor of software development. What used to be handled by Jenkins, GitHub Actions, or custom cron jobs is now taken over by an AI model with a natural language interface. The franchise system analogy fits perfectly here: Anthropic provides the infrastructure and the reasoning model, while developers define local rules (“fix the top bug every night at 2 a.m.”). The real genius lies in the abstraction: instead of writing YAML files and Bash scripts, you write a sentence, and Claude handles the implementation. This is reminiscent of the evolution of urban planning: from detailed development plans to framework guidelines that allow for local adaptation. Anthropic is betting that the future of development automation lies not in better CI/CD tools, but in delegating to AI agents that understand what is meant.

China Erases the US Lead in AI

The Stanford HAI Index 2026 documents a tectonic shift: China has eliminated the American lead in AI performance. While the US leads in capital, infrastructure, and chips, China dominates in patents, publications, and physical AI (robotics). South Korea, with the highest “innovation density” per capita, shows that it's no longer a two-country race. 44 nations now operate state-owned supercomputing clusters, while South America and the Middle East are falling behind—creating a new form of digital divide. Private companies develop over 90% of all significant AI models, with Google, Anthropic, and OpenAI no longer disclosing the dataset sizes and training duration of their latest models. Only 31% of US citizens trust their government with AI regulation, the lowest value after China (27%). Adoption is exploding globally with 53% regular users, surpassing the PC, internet, and smartphone—paradoxically, the US ranks only 24th worldwide with a 28.3% usage rate. → SiliconANGLE

Synthszr Take: The AI world order is restructuring like an ecosystem after an ice age: new species emerge, old dominances crumble. China's catching up is reminiscent of the Japanese semiconductor industry in the 1980s, but this time with a crucial difference: state-owned supercomputing infrastructure as a backbone. The 90% dominance of private actors, coupled with a loss of transparency, creates a sort of “feudal system of algorithms,” where companies rule over their black-box castles like medieval lords. South Korea's innovation density per capita shows that it's not size but concentration that makes the difference—like Singapore in finance or the Netherlands in agriculture. The irony is brutal: the US develops the technology that its citizens barely use (28%), while Asia, with an 80% usage rate, drives cultural integration. We are not experiencing a technological revolution, but a geopolitical redistribution of digital power.

After McKinsey, AI Agents Now Hack BAIN — in 18 Minutes

Cybersecurity startup CodeWall hacked Bain & Company's AI tool Pyxis in just 18 minutes. The one-person company's autonomous AI agent gained access to nearly 10,000 AI conversations, including confidential inquiries from Bain clients about their competitors, using publicly visible login credentials. The vulnerability lay in a username and password noted in the public web code. This is already the third successful attack by CodeWall on major consulting firms, following McKinsey and BCG. Bain promptly fixed the security flaw but emphasized that Pyxis only aggregates third-party data and does not store any proprietary company information. → AIDirectory

Synthszr Take: CodeWall demonstrates what happens when David fights Goliath with AI agents: a single founder dismantles the IT security of billion-dollar consulting firms. The pattern is reminiscent of the early days of white-hat hackers, except this time it's autonomous agents, not humans, finding the vulnerabilities. The 18-minute mark is less impressive than the fact that basic security practices (no passwords in the code!) are slipping through the cracks at the elite of strategy consulting. CodeWall's business model works like a digital immune test: the AI agents are the viruses that show where the organism is vulnerable. Bain's response (“only third-party data”) sounds like the classic consultant defense: downplay the problem while frantically fixing it in the background.

Stanford Report Shows a Growing Divide Between AI Experts and the Public

Stanford's annual report on the AI industry documents a widening split: while 56% of AI experts expect the technology to have a positive impact on the US in the next 20 years, only 10% of Americans share this assessment. The discrepancy is particularly evident in healthcare: 84% of experts see positive effects there, but only 44% of the population agrees. According to a Gallup poll, Gen Z is leading the growing AI skepticism, although about half of this age group uses AI tools daily or weekly. The public's concerns focus less on theoretical AGI risks and more on concrete impacts: job losses, rising electricity costs from energy-hungry data centers, and the concentration of power in the hands of a few tech corporations. Online reactions to attacks on Sam Altman's house showed a sentiment reminiscent of the comments following the attack on the United Healthcare CEO in 2024. → Techpresso

Synthszr Take: The AI elite live in a filter bubble reminiscent of Marie Antoinette's “Let them eat cake.” While Altman and Anthropic's founders philosophize about existential AGI risks, ordinary people worry about their electricity bills and wonder if their jobs will still exist next year. This gap between experts and the public follows a familiar pattern from economic history: the 19th-century railroad barons also didn't understand why farmers sabotaged their tracks. What the tech industry dismisses as irrational fears are rational responses to a technology whose benefits are concentrated among a few, while the costs (energy consumption, job losses, market power) are socialized. The 84% of experts who expect positive effects in healthcare probably don't work as nurses or radiologists. AI leaders would do well to spend less time on AGI scenarios and more time on the concrete fears of the people before online comments turn into real protests.