Tech • AI • Robotics • Game

Morning edition

Monday, November 16, 2026

  • AI-generated illustration
    OpenAI launched ChatGPT Wallet, enabling autonomous agents to shop and pay for services on users’ behalf. The product adds secure, transparent transaction handling to agent workflows, moving ChatGPT from recommending purchases toward executing them. That shift raises the commercial stakes for merchants, payment processors and identity providers, because authorization and auditability become core infrastructure when software spends real money. At last, the shopping bot has a wallet, but hopefully not the Konami code for unlimited credit 😉
    OpenAI also opened beta access to Business Hub and Business Agents, extending its commercial stack for enterprise users. The release packages agent-driven workflows into a dedicated business environment, giving companies a clearer route from general chat to operational automation. It matters because enterprise adoption increasingly depends on centralized management and repeatable processes, not isolated prompts, and the beta broadens OpenAI’s competition with workplace software vendors. The office copilot has entered multiplayer mode, and IT still controls the server 😉
    OpenAI launched Daybreak, agents that autonomously probe applications for vulnerabilities, while Uber disclosed a framework that continuously judges whether an agent’s actions are appropriate across its lifecycle. Together, the releases show two complementary defenses: offensive testing before attackers strike, and dynamic governance once agents interact with tools, APIs and real-world data. Traditional identity controls alone struggle with autonomous behavior, making runtime oversight a growing deployment requirement. One agent picks the locks, while the other keeps sudo privileges behind glass 😉
    Full article
  • AI-generated illustration
    Google began selling full-rack TPU systems externally, claiming 50% higher inference performance per dollar than Nvidia’s GB200 platform. Moving beyond cloud-only access turns Google’s custom silicon into a direct hardware challenge to Nvidia, giving large AI operators another vertically integrated deployment option. Price-performance gains at rack scale matter because inference is becoming a recurring operating cost, not merely a training expense. The GPU boss battle just unlocked a very large TPU-shaped final form 😉
    Full article
  • AI-generated illustration
    Apple and OpenEvidence launched an experimental health-chat app alongside a 200,000-participant study announced at HLTH26. The project will examine how consumers use health AI, gathering behavioral and outcomes data intended to improve the accuracy and reliability of diagnostic and treatment tools. Its scale gives Apple an unusually large real-world test bed, while OpenEvidence gains a route into mainstream patient engagement. The doctor may be virtual, but its sample size has definitely installed the enterprise RAM upgrade 😉
    DIAGENS launched iMed-6.0, a CT-focused AI model covering 509 diseases and imaging findings. That breadth positions the system as a versatile radiology assistant rather than a narrow detector, expanding the range of conditions that one model can flag. For healthcare providers, broader coverage can simplify procurement and workflows, although clinical validation remains crucial when one algorithm spans hundreds of diagnoses. It is basically a Pokédex for scans, except every entry needs peer review 😉
    Full article