Anthropic surges ahead in enterprise adoption; Apple wants Answers
Get bigger weekly updates! Free subscribers receive the top stories each week, while Paid subscribers will get a few extra stories. All support for Handy AI directly helps us maintain the newsletter and keep the information flowing.
last week’s top stories
💰 Anthropic surges ahead in AI race amid huge funding round. Anthropic’s Claude model has quietly overtaken OpenAI in enterprise adoption, now commanding 32% of enterprise LLM usage vs OpenAI’s 25%. The startup is doubling down on this lead with plans to raise $3–5 billion at a jaw-dropping $170 billion valuation. The massive funding, led by Iconiq and others, would cement Anthropic as one of the world’s top AI labs and fuel its rivalry with OpenAI (valued ~$300B). Read more
🍏 Apple forms new “Answers” team to build a ChatGPT rival. Apple has quietly created an Answers, Knowledge, and Information team (known as “Answers”) to develop an in-house AI answer engine that can respond to user questions with web-sourced knowledge. Headed by Siri/Search veterans, the group is reportedly working on a large language model system to integrate across Siri, Spotlight search, and other Apple apps. This initiative suggests Apple is aiming to reduce its reliance on Google (which pays billions to be default search on iPhones) and offer a more personalized, privacy-centric answerbot for Apple users in the future. Read more
🎓 ChatGPT gets a new ‘Study Mode’ to act as your AI tutor. OpenAI launched Study Mode in ChatGPT, a feature that guides learners step-by-step instead of just spitting out answers. When activated, ChatGPT will ask you Socratic questions, give hints, and even quiz you on material to deepen understanding. The goal is to curb students from cheating and instead turn ChatGPT into a 24/7 interactive tutor. Read more
📢 OpenAI’s GPT‑5 rumored to launch imminently with massive upgrades. Rumors swirl that OpenAI’s next-gen model could arrive as early as this week. Leaked codenames (“Nectarine,” “Lobster,” “Starfish”) hint at various GPT‑5 model tiers in testing, fueling excitement that a major ChatGPT upgrade is just days away. Read more
🤖 Zuckerberg touts “personal superintelligence” for everyone. Meta’s CEO Mark Zuckerberg outlined his vision for ubiquitous personal AI assistants, calling it a “new era of personal empowerment”. In a public letter, he predicted that AI will become deeply personalized (“…superintelligence that knows us deeply, understands our goals, and helps us achieve them…”) likely accessed through smart glasses as our primary devices. Zuckerberg’s plan involves AI that sees and hears what you do (via AR glasses) to provide constant, context-aware. Read more
👐 Z.ai releases a powerful open-source AI model to rivals. Chinese startup Z.ai (formerly Zhipu) has open-sourced GLM‑4.5, a massive 355 billion-parameter AI model that’s challenging the best proprietary models. GLM‑4.5, built on a Mixture-of-Experts architecture, ranks among the top performers on reasoning and coding benchmarks. By open-sourcing under Apache 2.0, Z.ai aims to let companies fine-tune and deploy GLM‑4.5 freely, a direct challenge to closed U.S. models and part of a broader push in China for homegrown, open AI alternatives. Read more
🔬 Stanford debuts AI-powered “virtual scientists” that make discoveries. Stanford researchers created a virtual AI lab, a team of AI “scientist” agents plus an AI principal investigator, that autonomously brainstorms, experiments, and finds solutions in biomedicine. In a test, the AI scientists were tasked with designing a better COVID vaccine and came up with an unconventional nanobody (mini-antibody) approach in days. When synthesized in real life, the AI-designed nanobody bound to the virus better than existing antibodies. Notably, the virtual lab runs research meetings in seconds and needed human help only ~1% of the time. Read more
🧪 AI Research of the Week
Working with AI: Measuring the Occupational Implications of Generative AI
From Microsoft Research
Jake’s Take: This study looks at how generative AI impacts different jobs by analyzing real-world interactions with Microsoft's Bing Copilot. The researchers sifted through 200k anonymized user-AI conversations, categorizing tasks users sought help with versus tasks AI directly performed.
They found that generative AI primarily excels at information gathering, writing assistance, and delivering guidance. Notably, knowledge-based roles (especially those involving data, writing, sales, and customer support) are most influenced. Despite popular narratives, here they show that there’s no strong correlation between AI applicability and job wages, though occupations demanding bachelor's degrees are somewhat more susceptible.
This work offers a much-needed, reality-based lens on how AI reshapes the workforce, highlighting both its promise and its limits.
and then, even more news…
🏹 Meta tried (and failed) to lure OpenAI’s CTO with $1.5 B offers. In the ongoing talent war, Meta reportedly targeted Mira Murati’s new AI startup (Thinking Machines Lab) by offering enormous pay packages (in one case $1.5 billion over several years) to poach her and her team. Murati (formerly OpenAI’s CTO) and co-founder Andrew Tulloch turned down Zuckerberg’s eye-popping bids, rebuffing both individual mega-deals and an acquisition offer for their fledgling company. Meta downplayed the exact dollar figures but didn’t deny aggressively recruiting AI elites. The episode shows that top researchers are increasingly choosing to build their own ventures rather than join Big Tech – even when offered “generational wealth” sums. Read more
🏗️ OpenAI builds $1B “Stargate” AI supercluster in Norway. OpenAI announced its first European data center (a Stargate Norway facility) to expand AI compute capacity overseas. The $1 billion project with partners Aker and Nscale will start with 100,000 Nvidia GPUs and could grow tenfold to meet surging demand. Slated to run on hydropower near the Arctic Circle, the site is among Europe’s first “AI gigafactories,” aimed at giving European developers and researchers local infrastructure for cutting-edge model training. Read more
💻 Tesla inks a $16.5 B deal for its next-gen AI chips. Elon Musk’s Tesla signed a $16.5 billion agreement with Samsung to fabricate Tesla’s in-house “AI 6” superchips at Samsung’s new Texas semiconductor plant. The AI 6 chip (Hardware 6) is Tesla’s all-in-one AI processor designed to power everything from Full Self-Driving car systems to Optimus robots and data-center training clusters. Musk noted Tesla’s spending on Samsung chips could ultimately be “several times” $16.5B as they scale production. This is a huge win for Samsung’s foundry (which already made Tesla’s prior autopilot chips) and reflects Tesla’s strategy to control its AI hardware stack as it transforms from an automaker into an “AI + robotics” company. Read more
⚖️ McKinsey sounds the alarm on AI upending consulting. The world’s top consulting firm, McKinsey, has entered “panic” mode internally as generative AI threatens to automate big parts of its business. In every board meeting, McKinsey’s leaders discuss the “existential” risk – AI that can analyze data and draft slick strategy decks in seconds could make traditional consultants far less essential. The firm has already cut ~5,000 staff since 2023 and deployed 12,000+ AI chatbot “agents” internally to assist with slide-making, note-taking and even checking the logic of recommendations. McKinsey’s CEO even envisions a future of “one AI agent per consultant”. Read more
🔒 Zoom backtracks after backlash over AI training on user data. Video-conferencing giant Zoom stirred uproar with a Terms of Service update that seemed to grant it rights to train AI on all user data (from meeting videos to chat texts) without opt-out. After public outcry, Zoom clarified that customer audio, video and chats won’t be used for AI without consent, and it amended the terms to say so explicitly. However, Zoom will still leverage “service-generated data” (like telemetry and engagement metrics) to improve AI features. Privacy advocates point out a remaining concern: if an enterprise enables Zoom’s new AI tools (like meeting summaries), individual employees may have little real choice: if your boss opts in, do you truly consent? Read more
💻 Mistral launches an AI coding assistant stack for enterprises. French startup Mistral AI rolled out a comprehensive AI coding toolkit built around its upgraded code model, Codestral 25.08. The new Codestral model boosts coding productivity, producing 30% more accepted autocompletions and 50% fewer gibberish “run-on” outputs than prior versions. Mistral’s solution combines the code AI with a vector search engine (for searching codebase), an “agent” to automate tasks, and plug-ins for VS Code/JetBrains IDEs. The result is an integrated AI dev assistant that companies can self-host (cloud or on-prem) to speed up code writing, reviews and testing by up to 50% while keeping sensitive code in-house. Read more