Radical Strategy Shift at SAP and Meta Employees Rebel

Klein Announces Big News: SAP Proclaims the Autonomous Enterprise

SAP is overhauling its entire enterprise software strategy, promising the “Autonomous Enterprise”—supported by Anthropic, Amazon, Google, Microsoft, Nvidia, and Palantir. The core promise: Over 50 specialized AI assistants are set to automatically handle business processes from finance to supply chain. The new “Autonomous Close Assistant,” for example, compresses financial closing from weeks to days by independently handling postings, reconciliations, and error corrections. SAP is luring partners into this new world with 100 million euros, while existing customers will only get access to the AI features if they commit to cloud migration. The Joule platform orchestrates over 200 specialized agents built on the new SAP Knowledge Graph—a structured map of all business entities and processes in the SAP universe. → The Deep View

Synthszr Take: SAP is making a bet here that's bigger than it first appears: It's not software eating the world, but autonomous agents eating software operation. The 100-million-euro war chest for partners shows how serious the Walldorf-based company is—and how much it wants to retain control. The real trick will be to marry the promised autonomy with the necessary governance (in financial closings, 'almost right' is indeed not good enough). The forced cloud migration for AI features is brutally clever: SAP is using the AI hype as leverage for the largest customer migration in its history. What Klein is announcing here is nothing less than the end of classic enterprise software: Instead of people typing data into SAP forms, agents will soon rule, negotiating with each other. The only question is whether SAP's notorious complexity can truly be transformed into autonomous elegance—or if we'll just get Byzantine processes with AI agents.

Meta Employees Protest Against Mouse Tracking in the Office

At Meta, employees are rebelling against new surveillance software that tracks mouse movements and keyboard inputs in the office. The technology is intended to increase productivity and identify 'unproductive time.' Employees at several US locations have started an internal petition that has already gathered over 2,000 signatures. The software not only records activity patterns but also creates individual productivity scores accessible to managers. A Meta spokesperson defended the system as an 'optimization tool for hybrid work models.' The irony: While Meta invests billions in VR worlds to define the future of work, the company is installing 1990s-era surveillance technology in-house. → Techpresso

Synthszr Take: The mouse-tracking affair at Meta reveals the dark side of the AI revolution: surveillance capitalism is eating its own children. While everyone outside is raving about agent AI and productivity boosts, tech companies are internally installing micromanagement systems that log every click. This isn't AI-powered work optimization; it's digital Taylorism with a machine-learning veneer. The real punchline: The same companies that promise us AI will liberate creative work are treating their own developers like assembly-line workers with a mouse-click quota. Meta could take the 2,000 protesting employees seriously and shut down the system tomorrow morning. Instead, they are probably optimizing the algorithm that identifies protest organizers.

Claude Automates the Legal Profession

Anthropic is bringing Claude for Legal from the lab to law firms. The new plug-ins and MCP connectors automate standard legal work: document review, case law research, and drafting legal briefs. Specifically for contract law, data protection, labor law, and AI governance. The connectors link directly to DocuSign, Box, and Westlaw. Paying customers can start immediately. The market is exploding: Harvey raised 200 million at an 11 billion valuation in March, while competitor Legora countered in April with 600 million and Jude Law as its celebrity spokesperson. At the same time, accidents are piling up: dozens of lawyers have submitted AI-generated briefs full of fabricated citations, federal judges have used ChatGPT for rulings, and California imposed the first penalty on a lawyer for AI hallucinations. The automatically generated complaints are already clogging the courts. → Techpresso

Synthszr Take: The legal profession is the perfect use case for AI automation: highly repetitive text work, clear rules, expensive per hour. An average associate spends 80% of their time on document review at 300 euros per hour. Claude does it for 3 cents per thousand words. The Jevons paradox hits hard here: when legal services become a hundred times cheaper, demand increases a thousandfold. Every rental agreement becomes a battlefield, every T&C a lawsuit. The law firms that bet on automation now will buy out their competition in two years. The others will wonder why they were still creating PowerPoint decks about 'AI risks' in 2026 while Harvey and Legora were already worth billions.

Anthropic Brings Back OpenClaw

After just one month of suspension, Anthropic has re-enabled the use of third-party agents like OpenClaw for Claude subscribers. The catch: Instead of unlimited usage, there are now separate 'Agent SDK Credits' worth $20 to $200 per month (depending on the subscription tier), which expire if unused. What began as a capacity problem in April 2026 is now turning into clever price differentiation: Inefficient agents like OpenClaw bypassed Anthropic's caching mechanisms and sometimes consumed thousands of dollars in tokens on $20 subscriptions. The new solution separates interactive from programmatic use; as soon as you use the Claude-p command or GitHub Actions, the separate budget applies. Once it's used up, usage stops unless you activate paid extra credits at API prices. → VentureBeat

Synthszr Take: Anthropic elegantly solves the classic platform dilemma between openness and profitability here. The $20 to $200 in agent credits are not generosity, but a precise calculation: those who use inefficient third-party tools pay their true costs. The expiration principle of the credits is the real masterstroke (use it or lose it). While OpenAI bets on a closed integration with ChatGPT Workspace, Anthropic chooses the middle path: controlled openness with transparent cost allocation. For enterprise customers at $200 per seat, it's a reasonable deal. For hobbyist developers at $20, it's effectively a price hike through the back door.

Stripe: Forward Deployed Agents Instead of Forward Deployed Engineers

Stripe is permanently stationing AI practitioners in its marketing organization: one specialist for every 20 employees. These 'Forward Deployed AI Accelerators' are not just meant to introduce tools, but to fundamentally change the way each employee works. The fintech company is turning its own workforce into a controlled experiment in human-AI collaboration. The data gathered from this flows directly into Stripe's agentic commerce infrastructure, which the company is positioning as the 'Visa for machines.' In parallel, Circle is building an exit strategy disguised as a product launch with its new Agent Stack—a pattern we will see more of in 2025. The integration of Anthropic's Claude agents and Google's Commerce agents shows that the market is already consolidating around a few central orchestration layers. → Linas from Linas's Newsletter

Synthszr Take: Stripe is doing something brutal here: they are treating their own organization as a product development lab. One AI specialist per 20 employees means about 400 full-time positions for operational AI integration alone among 8,000 employees. That costs at least $80 million per year. But Stripe isn't concerned with ChatGPT licenses or productivity tools. They are systematically building behavioral data: How do people really work with agents? Where do workflows break? What scales and what doesn't? This data is the real competitive advantage for their commerce infrastructure. Meanwhile, Circle's 'Agent Stack' shows how fintech companies create exit options: infrastructure layers positioned as standalone products can be sold off or licensed separately later. Practitioners should learn from both moves: anyone who only buys tools in 2025 instead of changing ways of working will fall behind.

China's AI Efficiency Moat – How Western Controls Created Stronger Competition

Azeem Azhar was in Beijing, Hangzhou, and Shanghai and visited 14 Chinese AI labs. His takeaway: US export controls have put China three years behind in compute, but at the same time created a 4- to 7-fold efficiency advantage in intelligence extraction per GPU. DeepSeek, MoonshotAI, ByteDance—they all operate with a fraction of the hardware available to OpenAI or Anthropic. While American labs are signing 10-gigawatt deals for Nvidia's latest Blackwell chips, Chinese researchers get their H100s via Singapore, declared as 'tea' or 'toys.' Despite a three-year hardware handicap, Chinese open-source models are only six to eight months behind the US frontier. → Azeem Azhar, Exponential View

Synthszr Take: The export controls are a perfect example of the Jevons paradox in geopolitics: resource scarcity leads to higher efficiency, and higher efficiency leads to more output. China has made a virtue of necessity—while Meta burns through half a million H100s for Llama 4, Chinese teams achieve comparable results with a tenth of the computing power. The average age in the labs: 25. This generation is growing up with compute discipline the way we in Germany grew up with waste separation. The real irony: America wanted to slow down China's AI development and instead has bred the most efficient AI engineers in the world. Anyone looking for the winners of the global AI race in 2030 might not find them among the players with the most GPUs, but among those who achieve more with less.

The 10x Developer is Dead

Investor Geoff Woo served founders a hard truth this week: '10x productivity' as a pitch is dead. Anyone who still comes with broad productivity promises today signals a weak positioning. Reality speaks a different language: Challenger data shows 83,387 job cuts in April, 38% more than in March. Tech companies are cutting the fastest and redirecting the money directly into AI infrastructure. Meanwhile, the first companies are testing organizational charts where three people report to twenty agents. This is no longer science fiction, but the next quarterly report. After 18 months, Mira Murati's Thinking Machines Lab is showcasing its first interaction model: real-time audio, video, UI generation, and micro-turns are taking AI out of the turn-based chatbot. → MyClaw Newsletter

Synthszr Take: The industry talks about productivity but means substitution. The 83,387 layoffs in April are not an efficiency gain but the beginning of an organizational upheaval where people report to machines. What Woo is touching on here goes deeper: the 10x developer was always a myth, but now it's being replaced by 20 agents who don't need sleep and don't negotiate salaries. The real question is not whether AI kills jobs (it does), but whether the new AI-powered business models are defensible at all when the next open-source model turns every advantage into a commodity. Murati's real-time models show where the journey is headed: agents are becoming collaborators that you can steer, interrupt, and redirect during work. The organizational chart of the future is no longer a pyramid, but a network of humans and machines—and the machines are in the majority.

Alexa Shops on Amazon – The Assistant Becomes the Point of Sale

Amazon is turning Alexa into a shopping interface: effective immediately, the AI assistant is directly integrated into the search functions of Amazon.com and the app. Instead of traditional search results, 'Alexa for Shopping' answers complex questions like 'When did I last order AA batteries?' or sets up automatic reorders. This replaces the previous Rufus assistant and marks a fundamental shift: the search box is becoming a conversational commerce interface. Alexa can now independently monitor prices, automatically buy when thresholds are met, and even shop for the user on other websites—the controversial 'Buy for Me' feature turns the assistant into an autonomous shopping agent. The rollout is starting in the US for all Amazon customers; an Alexa Plus membership is not required. → Techpresso

Synthszr Take: Amazon is finally solving the central weakness of e-commerce: the search box has so far only been a filter for existing products. With Alexa as a shopping agent, an intent-based system emerges—the user formulates their need, and the AI does the rest. This is reminiscent of Perplexity for shopping, but with a direct purchase connection and Amazon's massive product database behind it. The 'Buy for Me' feature shows where this is headed: autonomous agents that operate across platform boundaries. For brands, this will be brutal—if you don't land in Alexa's recommendation logic, you practically cease to exist. The next step is obvious: Alexa will become a personal shopping manager that proactively anticipates needs. Amazon is transforming from a marketplace into an AI-driven supply infrastructure.

AI Bros: Half-Open Laptops are the New Patagonia Vest

The finance bro's Patagonia vest has competition: the half-open laptop is the new status symbol of the AI coding scene. Business Insider documents a phenomenon observed everywhere from airports to ice skating rinks. AI power users are leaving their laptops demonstratively open so their coding agents can keep running. The reasoning sounds technically plausible: many laptops pause processes when closed. The solution seems archaic: jamming a finger between the display and keyboard or carrying the device around completely open. One user describes herself as 'the equivalent of an iPad kid for middle-aged women.' The comment section is exploding as expected: everyone knows you can adjust the power settings. Third-party software exists. 'Seriously?! These aren't engineers!' one writes. → Business Insider

Synthszr Take: The open laptops are perfect projection screens for the cultural divide between old-school coders and the new generation of vibe coders. The technical elite gets worked up over a lack of shell script knowledge, while the newcomers just build software. This is reminiscent of the reaction from computer science departments when Visual Basic suddenly turned everyone into a programmer. The difference today: AI tools not only democratize coding but also the visibility of technical work. The open laptop becomes the physical manifestation of always-on computing. What the gatekeepers read as technical incompetence is possibly the first authentic gesture of a post-code era: when intent becomes more important than implementation, the visibly working agent is more important than elegant power settings.

New York Times Enforces AI Ban for Freelancers

The New York Times sent out a 'periodic reminder' to its freelance contributors this week: no AI-generated content, no AI editing, no AI assistance in writing. The document explicitly prohibits ChatGPT, Claude, Perplexity, and even AI-powered Google searches. After several embarrassing incidents involving AI-fabricated quotes and plagiarized book reviews, the newspaper is drawing hard lines. Even 'high-level brainstorming' with AI tools is only reluctantly tolerated. The compliance department is redefining creative work: 'All writing must be the product of human creativity and craft.' → The Deep View

Synthszr Take: The Times is reacting to its AI mishaps like a 90s corporation reacting to the internet: with bans instead of skill-building. While experienced developers, according to a Fastly study, already have AI generate 32% of their code (increasing their productivity by 22%), the editorial team is even banning the rephrasing of single sentences. The real question is not whether to use AI, but how. Those who categorically exclude AI tools are losing touch with a reality in which the best practitioners have long since learned to use AI as an amplifier for their expertise. The Times is confusing risk management with a refusal to innovate.