Realtime AI News
Google widens access to Gemini Tasks, which can place calls and make purchases
Google is expanding access to Gemini's Tasks mode, which can coordinate phone calls, sign into websites, and complete purchases on a user's behalf. TechCrunch reported the move on November 17, framing it as a shift from a conversational assistant toward one that acts for the user.

Google is widening access to Gemini's Tasks mode, which can do considerably more than answer questions. According to TechCrunch's November 17 report, the mode can coordinate phone calls, log into websites, and complete purchases on the user's behalf.
The three capabilities map onto communication, identity and payment — precisely the domains that have long been treated as off-limits for automated systems. Letting software dial a number, hold a session on an account and spend money means the assistant now operates inside the parts of the stack where mistakes are expensive and hard to reverse.
What Google announced is an expansion of access, not a new product. That distinction matters: Tasks mode has evidently been running with real users long enough to justify a wider rollout, which moves it from a controlled test into a scale test. Edge cases, ambiguous authorizations and failed rollbacks surface much faster at volume.
The hard part of shipping delegated action was never whether a model can click a button. It is how permission is granted, how the process stays visible, and who owns the failure. Users need to know which identity the assistant is using, on which site, drawing on which payment method; platforms need a way to interrupt at any moment and reconstruct what happened afterwards.
The upside for Google is easy to see as well. Search and assistant products are built around the question of what the user wants to do next. If Gemini can finish that next step rather than merely describe it, it stops being an information gateway and becomes a transaction gateway — a billable path that sits alongside advertising and subscriptions.
The risks are equally legible. Placing calls raises recording and privacy questions, signing in raises account-security questions, and buying raises disputes over payment. A single public failure can damage trust in the whole category, which is the shared vulnerability of every company building action-taking agents.
What to watch next is unglamorous: which regions and which sites or call scenarios are supported, and whether Google exposes a full activity log and an undo path. The real bar for Tasks mode will be set by those product mechanics, not by a demo video.
Why it matters
Once an assistant can complete a purchase, competition shifts from raw model capability to permission design and liability. Solid logging, interruption and undo controls are what would move agentic tasks into everyday use.
Nearby Updates
All11/18, 03:05
Cato Networks unveils AgentSense to stop harmful AI agent behaviour in real time
Cato Networks has unveiled AgentSense, a capability it says stops harmful AI agent behaviour in real time, presented as a patented breakthrough. The product puts protection at the runtime layer, judging what an agent does rather than only filtering what goes in.
11/18, 02:48
Google to keep Gemini chats for two months, up from 72 hours, even with history off
Google is extending Gemini's retention of temporary conversations from 72 hours to two months, rolling out globally over the coming weeks. It is also separating chat history from AI training, which had been tied to one Keep Activity setting.
11/18, 03:32
Model ML raises $250M to put AI agents on investment banking grunt work
Model ML has raised $250 million for AI agents that understand the full context of a deal and automatically update the financial models and presentations behind it. The startup is targeting the repetitive spreadsheet and slide work that consumes analyst hours in banking and consulting.
11/18, 02:10
Sqwish Labs offers decision-model customization trained on customer data
Sqwish Labs is offering customization for decision models, letting customers train the system on their own data to extend what it can do, TechCrunch reported on November 17. The launch is framed against a backdrop of rapidly rising decision-model adoption.