What if an AI agent never had to leave the browser? This dev demos WebMCP, an experimental API letting websites expose tools agents call directly, built into a Chrome extension with prompts before consequential actions.
{ author: @SylwiaLask8977 }
Two new Gemini models built for voice first apps: 3.8 Live for real time conversational reasoning, and 3.5 Transcribe for streaming speech to text across 85+ languages. This dev covers deploying both.
{ author: @thorwebdev + @GoogleAI }
Realistic load does not come from hammering a single endpoint. This dev lays out designing performance tests around actual user workflows, from setting expected load to reading percentiles and resource use instead of averages.
{ author: @dbalcare }
Cloud inference is not the only option for AI agents. This dev swaps it for a locally running Gemma 4 model on Docker Model Runner, keeping the same ADK framework while handling function calling limits.
{ author: Mazlum Tosun + @GoogleDevExpert }
Clean code and clear code are not the same thing, and comments were never the real problem. This dev makes the case for comments that explain a magic number, warn of consequences, and record reasoning git history misses.
{ author: @georgekobaidze }