Tim Cook Hands Over to AI Skeptic John Ternus

Tim Cook Hands Over Apple Leadership to John Ternus

After almost 15 years at the helm of Apple, Tim Cook is stepping down as CEO in September and moving into the newly created position of Executive Chairman. The 65-year-old will be succeeded by John Ternus, the 50-year-old head of hardware engineering who has been with Apple since 2001. Under Cook's leadership, Apple's annual profit quadrupled to over $110 billion, while the company's value soared to $4 trillion. Ternus will be the eighth CEO in the company's 50-year history and faces the challenge of leading Apple at a time when the company has not introduced a new mainstream product in years and is lagging behind competitors in AI investments. Cook will remain with the company as Executive Chairman, focusing particularly on relationships with policymakers worldwide after years of navigating the often-conflicting agendas of Washington and Beijing. → The New York Times

Synthszr Take: Apple is undergoing the classic generational change of a mature technology corporation: The operations master hands over to the product engineer. Cook was the perfect CEO for the iPhone era, a supply chain virtuoso who orchestrated Apple's production machinery from China to Brazil, quadrupling profits in the process. Ternus is taking over a company that runs like a Swiss watch but is no longer producing revolutionary products. The parallel to Disney's Bob Iger succession is striking: both companies are searching for the next big growth story while milking their cash cows. The fact that Cook is handling diplomatic relations as Executive Chairman reveals Apple's true Achilles' heel: the company has become a geopolitical pawn between Washington and Beijing. Ternus' first task will be to lead Apple out of its comfortable iPhone dependency without jeopardizing its $4 trillion valuation.

What John Ternus Means for Apple

A trained mechanical engineer, John Ternus is praised internally for his humane leadership style—he even gave up a private office to continue sitting in the open-plan office with his team. Ternus was instrumental in the development of the first iPad and the AirPods and was most recently responsible for the Mac, Apple Watch, and iPhone product lines. Unlike Cook, who is considered a supply chain genius, Ternus is a 'true engineer' who understands the technical details of the products down to the smallest detail. However, he shares a certain risk aversion with his predecessor—Apple's hesitant stance on AI and smart home products is partly attributed to him. The biggest challenges for his tenure: Apple's weakening position in the Chinese market and the question of whether its restraint in the AI hype will prove to be farsighted or a missed opportunity. → gizmodo.com

Synthszr Take: Apple is switching from a business administrator to an engineer—a pattern that repeats itself in tech history like the seasons. After the founding generation (Jobs) comes the optimizer (Cook), then the product person (Ternus). IBM went through these cycles, as did Microsoft, and at Google, we are currently seeing the transition from Page/Brin to Pichai. The 'nice engineer' Ternus could be exactly the right type for a phase in which Apple is no longer defining the category but must build excellent products in existing markets. His risk aversion to AI could prove to be a feature, not a bug—while Meta and Google sink billions into LLMs, Apple might be perfecting the next generation of wearables or the mythical tabletop robot device. Silicon Valley loves its hero CEOs, but sometimes a competent craftsman is exactly what a $4 trillion company needs.

Qwen3.6-Max-Preview: Alibaba's Next Move in the AI Race

Alibaba is releasing an early version of its next proprietary language model with Qwen3.6-Max-Preview. The preview version shows significant improvements over Qwen3.6-Plus: better agentic coding capabilities (SkillsBench +9.9, SciCode +6.3), stronger world knowledge (SuperGPQA +2.3, QwenChineseBench +5.3), and more precise instruction following (ToolcallFormatIFBench +2.8). The model achieves top scores in six major coding benchmarks, including SWE-bench Pro and Terminal-Bench 2.0. It will be available via the Alibaba Cloud Model Studio API as 'qwen3.6-max-preview,' with a special 'preserve_thinking' feature for agentic tasks. Alibaba emphasizes that development is still ongoing and further improvements will follow → qwen.ai

Synthszr Take: Alibaba is playing the Silicon Valley playbook in reverse: instead of releasing a model and then patching it, it's deliberately releasing a 'preview' with an explicit promise of improvement. This is reminiscent of video game betas, where the community becomes part of the development process. The benchmark numbers (SkillsBench +9.9!) show real progress in agentic coding, the holy grail of AI development. The 'preserve_thinking' feature is interesting: while OpenAI hides its reasoning, Alibaba is making it a sellable product. The geographical API segmentation (Beijing, Singapore, US-Virginia) shows how geopolitical realities translate into technical infrastructure. Alibaba is betting that transparency in the thought process is the key to reliable AI agents.

Moonshot AI Releases Kimi K2.6 o and Challenges GPT-4

Moonshot AI is releasing Kimi K2.6, an open-weight model that is on par with GPT-4, Claude 3 Opus, and Gemini 1.5 Pro in coding benchmarks. The model achieves 94.0 on AIME, 93.2 on LiveCodeBench, and 83.2 on HumanEval. The real innovation lies in the 'Agent Swarm' system: K2.6 can coordinate up to 300 specialized sub-agents in parallel, each capable of executing 4,000 steps. The system automatically breaks down complex tasks into subtasks and distributes them to agents with different skills, such as web research, document analysis, or code generation. In a single run, it produces finished outputs like websites with animations and database connections, presentations, or spreadsheets. The license allows free use with one restriction: commercial products with over 100 million monthly active users or more than $20 million in monthly revenue must visibly mention 'Kimi K2.6' in the user interface. → the-decoder.com

Synthszr Take: Moonshot AI is turning AI models into ant colonies. While OpenAI and Anthropic are breeding their models to be ever larger, K2.6 is taking the path of swarm organization: 300 specialized agents work in parallel, like in a well-organized 19th-century factory, only digital and without breaks. This is reminiscent of the emergence of modern cities, where specialized districts (financial district, creative quarter, industrial area) are connected by efficient transportation routes. The license clause is a clever move: small developers can experiment freely, while tech giants have to advertise the 'Kimi' brand. Moonshot is betting that the future of AI lies less in a single super-brain and more in the skillful orchestration of many specialists.

OpenAI's Codex Films Your Screen and Sends It Home

OpenAI has given its Mac desktop client Codex a new feature called Chronicle, which periodically captures screenshots of the screen and sends them to OpenAI's servers for processing. The resulting text summaries are stored locally as unencrypted Markdown files, giving the AI assistant passive context about user activities. Chronicle is part of a larger update from April 16th titled 'Codex for (almost) everything,' which has transformed Codex from a pure coding assistant into a universal AI workspace. The feature requires a ChatGPT Pro subscription for at least $100 per month, an Apple Silicon Mac with macOS 14 or newer, and is not available in the EU, UK, and Switzerland—a geographical exclusion that strongly suggests OpenAI has recognized its incompatibility with GDPR. In contrast to Microsoft's Recall, which performs all processing locally on the device and stores screenshots encrypted, or the open-source project Screenpipe with its local approach, OpenAI deliberately chooses cloud processing. The screenshots are deleted after processing and not used for training, but the generated memories remain permanently as plain text on the computer—OpenAI explicitly recommends pausing Chronicle before meetings or when viewing sensitive content. → The Next Web

Synthszr Take: OpenAI is playing the same game as McDonald's with its franchise system, only in reverse: while McDonald's has central standards implemented locally, OpenAI centralizes local activities for processing. Chronicle is less a technical innovation and more a bet on user behavior—that people, for convenience (the AI assistant automatically 'knows' what I'm working on), will let their screen activities be funneled through OpenAI's servers. The geographical restrictions show that OpenAI knows exactly what it's doing: in markets with strict data protection, the feature is simply illegal; in the US, they are hoping for the well-known privacy paradox, where users express concerns but participate anyway. The comparison with the failed Rewind AI (bought and shut down by Meta) or Microsoft's Recall (39% subscriber loss) shows that screen capturing is a minefield. OpenAI is betting that its market power and the $100 price tag will be enough to win the trust that others have failed to secure.

Anthropic Secures $5 Billion from Amazon, Pledges $100 Billion in Cloud Spending in Return

Anthropic announced on Monday that Amazon is investing another $5 billion, bringing its total investment in the company to $13 billion. In return, Anthropic commits to spending over $100 billion with AWS over the next 10 years and will receive up to 5 GW of new computing capacity for training and operating Claude. The agreement is similar to a deal Amazon made with OpenAI just two months ago: Amazon participated with $50 billion in a $110 billion funding round that valued OpenAI at $730 billion. This deal was also partially structured as cloud infrastructure services instead of direct cash. At the center of the agreement are Amazon's custom chips: Graviton (a low-power CPU) and Trainium (an AI accelerator chip that rivals Nvidia). The Anthropic deal specifically includes Trainium2 to Trainium4 chips, although Trainium4 is not yet available. Anthropic has also secured the option to buy capacity for future Amazon chips as they become available. According to reports, VCs are offering the AI company capital in a deal that would value it at $800 billion or more. → techcrunch.com

Synthszr Take: Amazon is turning Anthropic into a captive major customer that has to pay back its own investment over ten years. $100 billion in cloud spending for a $13 billion investment sounds like a multiple, but with 5 GW of computing capacity and rising energy prices, Amazon could end up more profitable than any VC. The Trainium chips are the real leverage: Anthropic becomes the beta tester and reference customer for Amazon's Nvidia alternative while being locked into a proprietary ecosystem. The $800 billion valuation that VCs are supposedly discussing seems absurd when you consider that Anthropic has just mortgaged its next decade to a single cloud provider. Amazon is not just buying an AI champion here, but also control over its infrastructure destiny.

Adobe Fights Disruption of Its Own Business Model with Enterprise Agents

Adobe is responding to growing pressure from AI-native competitors with a new enterprise agent platform called CX Enterprise. The system aims to automate digital marketing, customer engagement, and sales, uniting three areas: an AI-powered content supply chain, the orchestration of customer interactions, and 'Brand Visibility'—the visibility of brands in a world increasingly shaped by AI agents. The 'CX Enterprise Coworker' can independently take on tasks, coordinate other agents, collect business data, and create and implement marketing plans. Adobe announced partnerships with over 30 AI platforms, including Amazon's cloud division, Microsoft, Anthropic, OpenAI, and Nvidia. Adobe's stock has fallen by about 30 percent this year after new AI tools from Anthropic and OpenAI spooked investors, costing software and data stocks hundreds of billions of dollars in market value. CEO Shantanu Narayen, who will step down after 18 years, admitted: 'There will be new AI-first applications. There is no doubt about that, and business models will change. → Techpresso

Synthszr Take: Adobe is acting like a franchisor who suddenly realizes that customers prefer to eat at food trucks. The CX Enterprise platform is an attempt to become a food truck itself while still running the large restaurant chain. The problem: Canva and Claude Design are already parked on the best corners. Adobe's partnership strategy with 30 AI providers is reminiscent of the city-states of the Italian Renaissance, which allied with all allies to avoid going under. The irony is that for decades, Adobe benefited from the interchangeability of its tools—despite high switching costs and network effects (anyone sending Photoshop files forces others to use Adobe). Now that AI agents can understand and convert these file formats, that moat is crumbling. Narayen's departure at this point seems like the captain abandoning ship before it's clear whether the new lifeboats will float.

The Story Behind China's Robot Marathon

Smartphone manufacturer Honor, once spun off from Huawei, achieved a remarkable victory at the Beijing E-Town Half Marathon in April: its humanoid robot finished the 21 kilometers in 50 minutes and 26 seconds—faster than the human world record. Honor took the top three spots with autonomously running machines, and even all six top spots based on weighted scoring. The team had only existed for a year. The robots navigated independently through 22 turns and various types of terrain. Established robotics companies like the Beijing Humanoid Robot Innovation Center and Unitree (with a pending stock market listing valued at $610 million) were pushed off the podium. While only six out of 102 teams reached the finish line in 2025, 47 did so in 2026—an explosion in performance within a single year. → Hello China Tech

Synthszr Take: Honor is turning humanoid robotics into a smartphone business: fast development cycles, vertical integration, and prices like those in consumer electronics. The parallel to China's electric car revolution is obvious—BYD and Geely learned from the mobile phone industry, and now that model is returning to robotics. Honor's head of engineering speaks of 'full-stack in-house development' within a year, while Unitree, despite a valuation of $610 million, sells mainly to research labs. The marathon is perfect marketing: visible superiority, measurable progress, viral potential. Smartphone margins plus a robot's ASP (Average Selling Price) could industrialize the industry faster than any Boston Dynamics approach.

Chinese Tech Workers Fight Back Against Their AI Clones

Technology companies in China are instructing their employees to train AI agents designed to replace them. A GitHub project called 'Colleague Skill' went viral after it showed how workflows and even personality traits of colleagues could be replicated in AI agents. The tool, which started as a parody, automatically imports chat histories and files from Chinese workplace apps like Lark and DingTalk and creates detailed manuals for automation. Employees report that their supervisors are actively asking them to document their workflows for AI tools like OpenClaw or Claude Code. As a counter-movement, product manager Koki Xu developed an 'anti-distillation' tool that sabotages the creation of useful work instructions by rewriting them into vague, non-actionable language. → Techpresso

Synthszr Take: Chinese tech workers are experiencing the digital equivalent of biological cell division: they have to extract their own DNA so the company can clone them. What makes 'Colleague Skill' so disturbing is not the automation of tasks, but the digitization of personality—including punctuation habits and communication quirks. The sabotage tools are more than technical resistance; they are an attempt to protect one's professional identity from commodification. Historically, this is reminiscent of the Luddites, except that today's machine-smashers are more subtle: instead of destroying looms, they feed the algorithms with meaningless noise. The real question is not whether AI agents will replace jobs, but whether companies understand that they stand to lose more than just their employees' quirks in the process.