By Roman Aglaganov
Most people still use AI assistants like a search box: ask, read, repeat. Google’s 13 October update to the Gemini app pushes harder towards an assistant that does things. It arrives with Gemini 4 Argon, Google’s new frontier model, for Google AI Pro and Ultra subscribers.
What is new
Chat or tasks: a toggle in the main window switches Gemini from chatting to “tasks”. In task mode, Google says, Gemini can keep working when you are not at your computer, across your Google apps and more.
More than 40 connected apps: Gemini now connects to services such as Instacart, OpenTable, Adobe and Shopify, so a task can run from ordering groceries or booking a table to building an online shop.
Coding on the Mac: the agentic coding interface from Google Antigravity is now built into the Gemini macOS app, so you can move from chatting to coding without changing tools.
Fewer filler words: Google says Argon answers more directly, with less hedging and fewer clichés, and reports its progress more often on long tasks.
Before you hand over a task
An assistant that acts on your behalf can also make mistakes on your behalf. A few habits help:
Start small: try low-stakes tasks first and check what it actually did.
Check connections: look at which apps you have linked, and remove the ones you do not need.
Confirm spending: keep a human “yes” before anything that costs money or sends messages.
Why it matters
Every big AI company is racing towards agents that act for you; Meta, for one, launched its Muse agent app last month. Google’s advantage is that many people already live in its apps. Whether that becomes an advantage in trust as well depends on how clearly Gemini shows what it is doing, and asks before it does anything that matters.
