🤖 Copilot Fusion turns GitHub Copilot into a model manager 🤖
Model routing integration lets Copilot balance speed and quality.

GitHub has launched Copilot Fusion, a new orchestration layer that automatically routes and combines multiple AI models for each task inside GitHub Copilot. Instead of forcing developers to pick a single model, Fusion reads the effort level the user selects and decides which models to call, and how to chain them, to complete the job. Quick fixes can go to lighter, faster models, while heavier refactors can draw on more capable ones working together. For enterprises, administrators keep control over which models Fusion is allowed to use, so compliance and cost policies stay in place. The move shifts Copilot from a model picker to a model manager, raising competitive pressure on Cursor and other AI coding tools that still rely on manual model selection.

🧩 Google unifies Connected Apps in one place 🧩
One hub simplifies access and integration for Google developers.

Google has introduced a centralized platform to build and manage Connected Apps, pulling the tools developers need into one place instead of scattering them across separate pages and consoles. From a single workspace, teams can create, configure and oversee the apps that connect to Google services, with less back-and-forth between settings, credentials and permissions. The update arrives as developers race to ship AI agents and assistants that must plug into user data and third-party services, where setup friction often slows launches. Simplifying that first step fits Google's broader effort to make its ecosystem, including Gemini, easier to extend. It is a quiet distribution play: easier building means more apps living inside Google's ecosystem, not a rival's.

🧠 AI performs autonomous brain surgery on a mouse, Oxford PhD claims 🧠
A viral claim of hands-off robotic neurosurgery still awaits independent proof.

Gabriel Ocana-Santero, an Oxford PhD and CEO of Exin Therapeutics, says he handed an AI full control of a custom motorized surgery rig fitted with cameras, with a single goal: reduce seizures in an epileptic mouse. According to his account, the AI chose the drug, targeted the brain region and carried out the injection on its own, while a vet stood by with a kill switch. If confirmed, it would be one of the first examples of AI moving from planning procedures to physically executing surgery on a living animal. The caveat is significant: no outside sources have verified his claims yet, and the details come from a viral thread rather than a peer-reviewed study. For now, it is a striking demo that needs independent validation before anyone calls it a milestone.

INTERESTING TO KNOW

🎨 Microsoft ships MAI-Image-2.7 image model 🎨

Microsoft AI has released MAI-Image-2.7, the latest version of its in-house image generation model, extending the rollout of first-party models built outside its OpenAI partnership. The release signals Microsoft's push to reduce dependence on OpenAI for creative features in products like Copilot and Bing, and adds competitive pressure on rivals such as Google's Gemini image models and Midjourney.

💰️ Nvidia invests $3B in Meta's equity offering 💰️

Nvidia will invest $3 billion in Meta's equity offering, deepening an AI infrastructure partnership tied to the rollout of Meta's growing data center capacity. The deal makes Nvidia both supplier and shareholder, a circular financing pattern increasingly common in AI, and it locks in long-term demand for Nvidia chips as Meta races to scale compute for its superintelligence ambitions.