Google、Skills と Gemini Live を Web に展開へ

Google appears to be preparing two notable additions to Gemini on desktop and web. In recent builds, a Gemini Live button has surfaced in the web version, suggesting the real-time voice mode may finally move beyond mobile. Live remains absent even from the desktop app today, though testing is reportedly underway within the scope of Google's Trusted Tester program, and a simultaneous debut across desktop and web looks plausible. Google itself promised new voice capabilities for the macOS app at I/O in May, and earlier signs in that app pointed to voice selection options and a screen-sharing overlay, so the groundwork has been visible for months.

The same builds also carry traces of skills being wired into regular chats. Skills currently live exclusively inside Gemini Spark, the autonomous agent Google gates behind its AI Ultra subscription, where they act as reusable instruction packages the agent applies to recurring tasks. The traces suggest ordinary chat users could soon:

• Upload their own skills
• Pick from predefined ones
• Build new ones from scratch
• Have Gemini generate a skill on their behalf, even without Spark access

That would mirror what Claude and ChatGPT already offer, closing one of the more visible feature gaps in Google's assistant, and it fits a broader pattern: Chrome received a lighter prompt-shortcut take on skills in April, and Gemini Enterprise already lets workers invoke skills mid-chat.

Timing is the open question. Gemini 3.6 Flash appears to be in preparation while Gemini 3.5 Pro has slipped repeatedly, most recently past a mid-July target, so a stopgap model could arrive first. Something may surface on Google by the end of July, but nothing here has been announced, and plans of this kind can shift before launch. Together with Live on web, chat-level skills would make the everyday Gemini experience considerably more capable.