Google used a November 9 keynote to argue that AI agents are straining the way data centers were built, and it showed new chips, open software and a runtime for agents in response. The main takeaway is that Google is betting on hardware and openness together, with new TPU systems on one side and a donation of its agent runtime to the CNCF on the other. Here is what was announced and what it means for people building on this stuff.

Greg Brockman sat down with the Wall Street Journal the morning after an election that made data centers a political liability, and he spent half an hour defending OpenAI's way of building, selling and securing AI. The main takeaway is that OpenAI now sees itself as an enterprise company that needs enormous amounts of compute, even as voters, investors and rivals push back on how it gets there.

Rody Davis, Google's staff developer relations lead for Antigravity, used a 30 minute session to show how the tool works day to day, from permission settings to cloud agents to a demo that deletes old code. Much of what he covered landed in the past few weeks, so this post walks through what is new, what still has rough edges, and where the multi agent commands fit.

Walter Isaacson opened his new series The Builders with a long conversation with Elon Musk, and it covers more ground than the headlines suggest. Musk talks about why he thinks humans will not stay in control of AI, why he regrets how long he stayed in politics, and why he believes robots, not budget cuts, are the only way out of the national debt. The best way to read it is as a look at how the world's richest person thinks, including the claims that deserve a second look.

Demis Hassabis told Reid Hoffman at the Masters of Scale Summit that the world is not ready for what comes next in AI, with the greatest risk coming from closed loop self improvement, where AI builds its successors without human review. He called for hardened sandboxes, stronger monitoring and an industry run standards body under government oversight, while noting that weak international institutions make this harder. He also urged the world to build intelligent tools before conscious machines, and pointed to AI’s work on disease, from AlphaFold to Isomorphic Labs, as the clearest proof of its benefit.

At a Stanford HAI fireside chat, former Google AI head Jeff Dean joined Fei-Fei Li to discuss his new startup, Discovery Loop, which automates the scientific experiment cycle to boost expert productivity up to 50-fold. Dean demystified AI hype by distinguishing simple task-driven self-improvement from true recursive self-improvement, emphasizing that physical compute and time limits prevent infinite-speed AI loops. Reflecting on AI milestones like the Transformer and his past architectural misjudgments, Dean concluded that while machine automation will dramatically accelerate research and engineering, human direction, policy, and passion remain essential to shaping AI's societal impact.

At the Open Source Summit in Prague, Linus Torvalds reflected on 35 years of Linux, from hand-rewriting early patches and creating Git to maintaining today's strict 9-to-10-week release cycle. He noted AI's double-edged impact on open source, where maintainers fight an influx of AI-generated patches with AI review bots like Sashiko—and warned that while AI is a great gateway for beginners and quick UI prototyping, serious code still requires human understanding.

At GTC Berlin, Jensen Huang defended the economics of GPU-backed debt, arguing that Nvidia’s programmable chips stay productive for 7 to 8 years, far longer than skeptics claim. To help neoclouds turn idle capital into revenue faster, Nvidia unveiled its DSX POD container design to speed up data center deployment, though power constraints remain a bottleneck. Huang also addressed Nvidia’s financial backstops for customers, recommended a hybrid approach for enterprise AI models, and touched on global talent trends, noting China’s rapid rise in top AI researchers.

Apple's iOS developer preview lets Siri AI execute third-party app actions via custom tool intents and markdown-based App Skills. Custom tools route to Siri when an app is open, named, in a Live Activity, or in recent context, requiring developers to focus on slim tool sets, batch parameter support, and testing via Xcode's Siri Evaluations framework before spring App Store submission.

Japan Airlines (JAL) is rapidly scaling AI agents to combat a shrinking workforce, leveraging high employee engagement and rigorous security-focused evaluations. By combining a tailored four-level training curriculum with cross-functional teams, JAL has empowered employees to build over 700 agents for office and operational tasks, setting the stage to expand secure field integrations and achieve its 2030 vision as an AI-native airline.

Shiseido’s Japan team leverages AI behind the scenes to optimize its "Personal Beauty Partner" (PBP) operations—removing administrative burdens like training podcasts, complex multi-store scheduling, and social media compliance checks—so beauty advisors can dedicate more time and authentic warmth (omotenashi) to customer conversations.

Digital bank Revolut slashed its AI coding costs by 90% by moving 2,000 engineers from metered proprietary APIs to self-hosting the open-weight GLM 5.3 model on 56 Nvidia B200 GPUs. Providing predictable, flat-rate access eliminated token rationing, allowing developers to automate routine tasks and improve deployment lead times by 70%, while relying on strict inference tuning, compliance sandboxing, and keeping frontier models like Gemini and Claude Code as high-end fallbacks.

Demis Hassabis, the Nobel-winning CEO of Google DeepMind, envisions a future where AI transcends current hurdles—such as long-term planning and "true creativity"—through recursive self-learning, provided the field can bridge the gap in robust world models and verification. While he remains cautious about integrating advertising into AI assistants due to the potential for eroding user trust, he suggests that fierce commercial demand for reliable enterprise tools will paradoxically drive better practical safety guardrails as we move toward AGI.

Sulaiman Ghori's career at xAI ended in January 2026, just four days after a 71-minute podcast interview where he revealed sensitive details about the company's "move fast" culture. Ghori disclosed regulatory loopholes used for the Colossus data center, the stealth "Macrohard" project, and an audacious plan to use idle Tesla vehicles as a distributed computing network. His swift exit highlights the strict code of silence and extreme intensity that define Elon Musk's most secretive ventures.

Eric Schmidt delivered a compelling warning at the America Business Forum in Miami, arguing that we are on the cusp of a transformation as profound as any in human history – largely driven by rapidly advancing artificial intelligence. He highlighted breakthroughs like AI-generated Shakespearean sonnets, sophisticated language models, and reasoning systems capable of outperforming experts, predicting a future where access to “superhuman” intelligence could be democratized globally.

Jensen Huang argues that America’s AI competition with China is far more complex than headlines suggest, focusing on a “five layer stack” – energy, chips, infrastructure, AI models, and applications. While the U.S. maintains an edge in chip technology and AI models, China dominates energy capacity, manufacturing, and open-source development, creating significant competitive disadvantages. Huang stresses that America’s ambivalence towards energy growth and its reluctance to reindustrialize are key vulnerabilities.

In a wide-ranging conversation with Nikhil Kamath, Elon Musk argues that the core of entrepreneurship is being a "net contributor" to society by focusing on creating useful products rather than chasing wealth. He describes a rapidly approaching future where AI and robotics will trigger a "supersonic tsunami" of productivity, potentially leading to a post-work society where energy replaces currency and collective human consciousness is amplified through interconnected technology like X and Starlink.

At the Microsoft AI Tour in Mumbai, CEO Satya Nadella showcased a significant shift in the AI landscape, moving from theoretical potential to a "builder's era" where the barrier to creating sophisticated applications has drastically lowered. By demonstrating a personal cricket research tool built over a weekend, Nadella illustrated a new "test-driven" development philosophy where users utilize "context engineering" and multi-agent frameworks—like the "LLM council"—to solve complex problems through iterative debate rather than rigid specifications.

Despite mounting competition from rivals like Google and DeepSeek, OpenAI CEO Sam Altman maintains that ChatGPT’s dominance is secured by its AI-first user experience and deep personalization rather than simple distribution. The company is currently pivoting toward a massive enterprise infrastructure role, reporting that its API business is outgrowing its consumer sector and that its latest reasoning models now outperform human experts on over 70% of knowledge work tasks.

Anthropic CEO Dario Amodei highlights a stark divide in the AI industry between sustainable growth and "yoloing" financial risks, noting that while the technology’s capability follows a predictable upward trajectory, the business models supporting it are increasingly precarious. Companies are navigating a "cone of uncertainty," making $50 billion infrastructure bets today for customers they hope to have in 2027, often relying on "circular deals" where chip makers fund the very startups that buy their hardware.

Google co-founder Sergey Brin reflected on his short-lived retirement, calling it a "mistake" because the lack of intense technical challenges left him feeling intellectually dull. This realization, combined with his admission that Google initially "messed up" by underinvesting in the AI revolution it helped start, spurred his return to active duty to work on the Gemini AI project.

Google DeepMind is revolutionizing robotics by moving away from rigid, pre-programmed movements toward Vision-Language-Action (VLA) models that allow robots to comprehend and adapt to the real world. By utilizing the same reasoning architectures as large language models—including "chain-of-thought" processing—these robots can now perform complex, long-horizon tasks like packing luggage or sorting trash based on abstract goals rather than step-by-step instructions.

The current hiring slowdown is a correction from previous overhiring, shifting the demand toward candidates who can demonstrate real business value, execute projects independently, and collaborate effectively. To succeed, graduates must look beyond prestigious brand names and focus on working with high-quality teams, mastering both large-scale and specialized Small AI models, and maintaining a bias toward delivery by building functional solutions rather than just listing technical credentials.

At a July 2025 summit in London, French President Emmanuel Macron, alongside AI leaders, declared Europe's determination to become a major player in the global AI race. Acknowledging Europe's current lag in computing capacity—possessing only 3-5% of global capacity despite 20% of demand—Macron outlined a strategic shift. The focus is now on building robust AI infrastructure, fostering talent, and bridging the gap between research and commercialization. By prioritizing scientific excellence, open-source models, and democratic values, Europe aims to achieve technological sovereignty and shape the future of AI on its own terms.

Anthropic has launched "Claude for Financial Services," its first industry-specific AI solution, aiming to revolutionize how financial professionals conduct research, build models, and make investment decisions. This AI, integrating with major financial data providers like FactSet and S&P Global, enables analysts to synthesize information and generate comprehensive reports much faster, with early adopters like Norway's sovereign wealth fund reporting significant productivity gains (20% or 213,000 hours saved annually) and AIG improving underwriting accuracy from 75% to 90%.

Mark Zuckerberg is steering Meta towards an ambitious future centered on "personal super intelligence," a vision distinct from other tech giants' pursuit of artificial general intelligence. Instead of focusing on large-scale problem-solving, Meta aims to develop powerful AI that enhances individuals' daily lives, particularly in areas like relationships, creativity, and productivity. This endeavor is backed by a staggering investment in infrastructure, including "hurricane-proof tents" for rapid data center deployment, and a unique approach to attracting top AI talent by offering massive computational resources over large teams.

The AI community has long misunderstood intelligence, focusing on memorization rather than true adaptive reasoning. Despite massive computational scaling, AI models performed poorly on benchmarks like François Chollet's Abstraction and Reasoning Corpus (ARC), which demands genuine problem-solving in novel situations. A breakthrough in 2024 with "test-time adaptation" allowed models to dynamically learn and adapt, achieving near-human performance on ARC. This signifies a crucial shift from simply scaling pre-training to integrating both intuitive and symbolic reasoning, much like human thought.

UBTECH Robotics is making significant strides in the humanoid robot market with two key innovations: the Walker S2 and the Walker C. The Walker S2 is the world's first humanoid robot capable of autonomous battery swapping, enabling 24/7 continuous operation in industrial settings without human intervention, effectively creating a "never-ending shift" for robotic labor. Standing 163 cm tall and weighing 43 kg, the Walker C features multilingual interaction, autonomous navigation, and human-like motion, highlighting a shift toward robots augmenting human capabilities in service-oriented applications like exhibition halls, offices, and shopping malls.

OpenAI has achieved a significant breakthrough in AI reasoning, with its experimental model securing a gold medal at the 2025 International Mathematical Olympiad (IMO). This marks a pivotal moment as the AI demonstrated human-level creative thinking and the ability to construct complex, multi-page mathematical proofs under strict competition conditions (two 4.5-hour sessions, no external tools, and natural language proofs). The model's success in solving five out of six problems, earning 35/42 points, highlights its advanced capabilities in sustained, intuitive mathematical reasoning, a domain previously thought to be uniquely human.

Companies are increasingly using AI to create synthetic customers for market research, simulating consumer responses with surprising accuracy, as highlighted by Harvard Business School research. This innovation allows for rapid and cost-effective testing of product concepts and pricing, fundamentally changing how companies approach innovation by enabling them to test many more variations before investing in human research. Fine-tuning these AI models with a company's proprietary historical customer data dramatically improves accuracy, providing a competitive advantage and suggesting a future where market research combines synthetic insights for broad exploration with targeted human research for validation and deeper understanding.

Calvin French-Owen, a former engineer at OpenAI, offers a candid look into the company's culture from May 2024 to July 2025. He describes an organization driven by secrecy, speed, and relentless ambition, where the workforce dramatically expanded from 770 employees in November 2023 to 3,531 by September 2024. French-Owen highlights OpenAI's unique operational style, characterized by a near-exclusive reliance on Slack for communication, a "bottoms-up" meritocratic structure, and an absence of traditional corporate planning. The company operates under intense public scrutiny, with a "very secretive" atmosphere regarding internal operations and finances.

The International Space Station (ISS) serves as a unique orbiting laboratory, conducting groundbreaking research in microgravity that directly benefits humanity on Earth. Experiments on the ISS are revolutionizing medicine by providing unprecedented insights into cardiovascular and immune system behavior, leading to potential breakthroughs in treating heart disease, autoimmune disorders, and cancer. Beyond human health, the ISS is advancing life support technologies for future space missions, enabling the manufacturing of superior materials impossible to create on Earth, and providing critical Earth observation data for climate science and disaster preparedness

Andrew Ng argues that in the rush to build AI agents, developers mistakenly prioritize complex autonomous systems over simpler, linear workflows. He advocates for starting with straightforward business processes that can be broken into sequential steps, leveraging existing "Lego brick" tools, and implementing quick, iterative evaluation systems. Ng also highlights the underexplored potential of voice applications, the importance of the Model Context Protocol (MCP) for data integration, and the transformative impact of AI-assisted "vibe coding."

At I/O 2025, Google unveiled SignGemma, a groundbreaking AI model within the Gemma family designed to translate sign language, specifically American Sign Language (ASL), into spoken English text in real-time. This open-source model aims to significantly improve accessibility for the 70 million deaf people globally by addressing the complexities of visual language, including hand movements, facial expressions, and body positioning. SignGemma holds immense potential for applications in customer service, education, healthcare, and workplace accessibility.

France's Mistral AI has launched a new Agents API, a comprehensive platform enabling AI agents to perform real-world actions, maintain context, and coordinate complex tasks. It includes built-in connectors for code execution, image generation, web search, and document access, distinguishing itself from basic chatbots. By focusing on multi-agent orchestration and open standards like the Model Context Protocol, Mistral aims to offer a robust and flexible alternative for enterprise AI automation.

Anthropic has introduced voice mode for its AI assistant, Claude, built on the Claude Sonnet 4 model, enabling users to have natural conversations, interact with documents and images, and integrate with Google Workspace for paid subscribers. This move intensifies the AI voice assistant battleground, where competitors like OpenAI's ChatGPT and Google's Gemini Live are also vying for supremacy, focusing on natural interaction and enhanced productivity, despite early challenges in voice quality and human-like speech.