AI Industry Midday Report｜From 'Chatting' to 'Delivering': AI Agents Are Entering the Engineering Phase This week, the focus isn’t on yet another chat model but rather on the maturing infrastructure for Agents. On July 7, Google expanded the Gemini API’s Managed Agents, adding backend tasks and remote MCP connections. This means developers can now delegate time-consuming processes to managed Agents and integrate external tools. Meanwhile, Microsoft announced a $2.5 billion investment and a 6,000-person team to help enterprises implement AI, signaling that the real bottleneck for companies has shifted from 'choosing a model' to data, permissions, processes, and evaluation. Practical advice: Start with a high-frequency, low-risk, and verifiable process as a pilot, such as converting meeting notes into to-do lists for manual confirmation. Keep logs for every step, including inputs, tool calls, and final outputs. Then measure the benefits using metrics like accuracy, time spent, and human intervention rate. No matter how powerful the model is, without permission boundaries and rollback mechanisms, Agents will struggle to move into production. #AI #AIAgent #TechNews #GoogleGemini #MicrosoftAI