For years, tech companies have promised AI assistants that do things rather than merely suggest them. Google is now making a credible attempt to deliver on that promise, with its March 2026 Pixel feature drop turning Gemini into what the company calls an "agentic" assistant — one capable of opening apps, filling in forms, and assembling orders on your behalf.
The headline capability allows Pixel 10 owners to instruct Gemini to complete tasks inside third-party apps. As reported by The Verge, those apps include Uber and Grubhub, with DoorDash also confirmed as a partner. TechCrunch reports a broader list at launch in the US, covering services such as Instacart, Lyft, McDonald's, Starbucks, and Uber Eats as well. Gemini handles multi-step tasks across food, grocery, and rideshare apps, running inside a secure virtual window while the user continues using their phone normally.
The key guardrail is a mandatory human confirmation step. Gemini never hits "confirm" or "pay" without the user's final tap. Google also notes that users can supervise or interrupt the process at any time, and the AI agent operates in a virtual environment and does not gain unrestricted access to data stored on the device. That matters: giving any AI agent access to external accounts carries real risk, as Google itself recommends supervising Gemini closely and interrupting it when necessary.
Strip away the buzz and the fundamentals show a meaningful but tightly scoped debut. The initial beta is starting in the United States and South Korea only. The agentic features are restricted to the Pixel 10 series, with older handsets excluded — though it remains unclear whether this reflects a hardware requirement tied to the Tensor G5 processor, or simply Google reserving new capabilities for its latest flagships. Australian Pixel users, regardless of model, are not in scope for this rollout at launch.
The competitive context is pointed. Apple has been promising a more capable Siri since at least 2024, and those features remain delayed. Google's consumer rollout precedes broader iOS equivalents, letting it capitalise on Android's roughly three-billion-device global base for scale. Samsung, which previewed the same Gemini agentic features at its Galaxy Unpacked event last week, is shipping the Galaxy S26 with these capabilities on March 11. Samsung is layering Gemini, Perplexity, and a revamped Bixby into what it calls a multi-agent stack, marketing the S26 as the world's first "agentic AI phone."
More from the March Drop
Beyond the agentic headline, the March update carries several practical additions. Google is updating Magic Cue so that when friends mention restaurants in a conversation, the Pixel proactively surfaces search suggestions based on criteria drawn from the chat. Circle to Search gains the ability to identify individual pieces of an outfit in a photo and surface places to buy them, with a virtual try-on option also included for Pixel 10 users.
A new Comfort view mode offers a refined display colour-filtering option to reduce visual stimulation from very bright or saturated colours, designed for sensitive users on the Pixel 10, 10 Pro, 10 Pro XL, and 10 Pro Fold. Older Pixel models are not left out entirely: the Pixel 8 and above gain a desktop mode when connected to an external display, Pixel 7 and newer devices pick up commute updates via At-a-Glance, and Pixel 6 phones and newer can now generate custom AI app icons, with Google offering styles including Scribbles, Cookies, Easel, Treasure, and Stardust.
Separately, as reported by Engadget, Google Home is gaining a Gemini-powered Live Search feature that allows users to query their cameras in natural language — for example, asking whether a car is in the driveway. The capability is limited to Google Home Premium subscribers, who pay USD $20 per month or $200 per year. Google Home's head, Anish Katturkaran, said the team launched the Gemini for Home service in October 2025 specifically to gather real-world feedback, and that regular voice improvements are being pushed based on user input.
What the market hasn't priced in yet
Agentic AI is genuinely new territory for mainstream smartphones, and Google deserves credit for shipping something real rather than staging another demo. Privacy advocates will reasonably ask how much cloud processing accompanies these virtual sessions, and regulatory scrutiny on AI autonomy and data handling could influence the rollout pace as the feature expands beyond the US and Korea.
For Australian consumers, the more immediate question is access. None of the confirmed app partners for agentic features — DoorDash, Grubhub, Uber — map neatly onto the local delivery market as it stands, and Google has not specified which Australian services, if any, will be included when the feature eventually broadens. More comprehensive details on how Gemini AI agents will evolve are expected at Google I/O 2026, scheduled for May 19.
The honest assessment is that this is an early beta with real promise and real limitations. The privacy architecture — a sandboxed virtual window, human confirmation before payment, the ability to interrupt at any time — reflects a measured approach. Whether Google can keep that architecture intact as it scales to more apps and more markets is the question worth watching. Agentic AI that works reliably is transformative; agentic AI that occasionally orders the wrong thing, or worse, is a different story entirely.