Phone AI assistants can start shopping and booking, but rarely finish

In short: Phone AI assistants are getting better at doing multi-step tasks across apps, but most still stop before final checkout and need a person to approve the last step.

What's going on

AI on phones is starting to act more like a helper that can do chores, not just answer questions. This is sometimes called “agentic AI”, meaning the AI can take actions across different apps (like a personal assistant who can open the right apps and fill things in for you).

Google has shown beta features for Gemini that can handle tasks like building a grocery cart or setting up a ride on some devices. Samsung has also described its newest phones as “agentic AI” devices, with similar goals around automation.

Delivery and ride companies are moving in the same direction. DoorDash has added AI tools that can create an order from a written prompt or a photo, and it can build carts from recipes or grocery lists. Uber launched “Cart Assistant,” which turns text or image prompts into a grocery basket at participating stores.

What to watch

A key question is whether people actually want an AI to do the whole job, end to end. The reporting so far shows many companies offering these features, but less proof that most shoppers want to hand over full control.

There are practical reasons for caution. People often want to confirm brands, prices, and substitutions before paying. Orders can also get complicated because items sell out, coupons change, and each store has its own rules. For now, the most common setup looks like a hybrid, where the AI drafts the cart, then you review and approve the final purchase, like a friend who helps write your shopping list but still lets you check it before you buy.

Source: NYTimes