AI Agents Spot 76% API Bugs, Google Debuts Gemma & Flow

KushoAI evaluated seven AI systems on a black‑box API bug detection benchmark covering 97 planted bugs across 20 live APIs. KushoAI identified 76% of complex bugs, beating the best coding‑agent workflow (53%) and general‑purpose LLMs (34%). The study highlights that simple schema tests are now baseline, while cross‑field logic detection separates top performers.

A TechCrunch piece cites research showing developers now refuse to work without AI, but AI‑written code can double maintenance burdens, turning short‑term speed gains into long‑term debt. Recent Amazon and Uber setbacks illustrate how unchecked AI coding can inflate token costs and trigger costly outages.

Satya Nadella explains how Microsoft is doubling down on AI infrastructure, deepening its partnership with OpenAI, and refocusing on core software and cloud competencies. He also discusses Project Solara, data‑center economics, and the shift toward an agentic platform that could reshape the company’s growth trajectory.

Google Flow is an AI‑powered creative studio that lets users generate, edit, and remix text, images, and video with Gemini’s generative models. It offers natural‑language editing, custom tool building, and collaborative sessions, aiming to streamline the whole creative workflow from concept to final assets.

Google unveiled Gemma 4 12B, a 12‑billion‑parameter model that processes text, vision, and audio without separate encoders. Its encoder‑free architecture lets it run locally on consumer laptops with just 16 GB of VRAM, and it’s released under Apache 2.0, enabling developers to build on‑device multimodal AI.

Google’s Gemini Omni adds multimodal AI video generation and avatar cloning to the Gemini app, letting users turn text, photos or video into fully edited clips via natural conversation. The feature replaces the older Veo model and is available to Google AI Plus, Pro and Ultra subscribers 18+.