Google is putting more of its AI agent into more hands. The company is expanding access to Gemini Tasks, a feature that lets the assistant take real-world actions — coordinating phone calls, logging into websites, and completing purchases — without a person clicking through each step.

Tasks mode has been in limited testing, and the wider rollout means more Gemini users will be able to hand off errands that previously required manual follow-through. Instead of asking Gemini for restaurant recommendations and then booking the reservation yourself, you can ask it to make the call. Instead of finding a product and checking out, you can ask it to buy it.

This is part of a broader shift Google has been building toward for over a year, including earlier work on an agent called Project Mariner that could navigate websites on a user's behalf. Gemini Tasks appears to be the more consumer-facing, less experimental version of that same idea — an AI that doesn't just answer questions but completes transactions.

The feature requires giving Gemini access to accounts, payment methods, and in some cases phone capabilities, which is a meaningfully bigger permission ask than a typical chatbot interaction. Google has not published detailed guardrails publicly, but AI agents with purchasing authority typically ship with spending limits, confirmation prompts before final checkout, and logs of what the agent did — features worth checking before enabling this for anyone other than yourself.

Why it matters

Gemini Tasks is arriving alongside a wave of similar moves from competitors. OpenAI has its own Operator-style agent for browsing and task completion, Amazon has been pushing Rufus deeper into its shopping funnel, and Perplexity has tested shopping and booking features inside its assistant. The common thread: every major AI company is racing to move from answering questions to executing actions, because that's where the subscription revenue and lock-in potential lives.

The pattern with agentic features so far has been cautious, staged rollouts followed by expanded access once a company is confident the failure rate — wrong purchases, bad bookings, miscommunicated calls — is low enough to not generate bad press. Google's expansion suggests it's reached that comfort threshold, at least for the current feature set.

What this means for small businesses

If your business takes orders or bookings over the phone, website, or app, expect a slow but real increase in interactions coming from AI agents rather than people directly. That has practical implications for customer service scripts, CAPTCHA and bot-detection systems, and how your checkout flow handles an entity that behaves differently than a typical shopper — faster, more literal, and less likely to abandon a cart out of hesitation.

For business owners using Gemini personally or across a team, the task-automation angle is worth testing on low-stakes errands first: scheduling calls, comparing vendor options, or handling simple reorders. Save purchases above a certain dollar threshold for manual review until you've seen how the agent handles edge cases like unclear pricing, substitutions, or canceled orders.

There's also a liability question nobody has fully answered yet: if an AI agent makes an erroneous purchase or miscommunicates something on a business call, who is responsible — the user, Google, or the business on the receiving end? That's unresolved territory, and contracts and terms of service haven't caught up.

What to watch

Watch for Google publishing specific spending caps or confirmation requirements for Tasks, which would signal how much autonomy it's actually comfortable giving the feature. Also track whether competitors respond with their own expanded rollouts in the next few months, and whether any high-profile mistakes — wrong orders, unauthorized calls — surface publicly, since that has historically been the trigger for companies to pull back agentic features or add more friction.

The bottom line

Gemini Tasks is now reaching more users, giving Google's AI the ability to act on a person's behalf rather than just advise them — a capability still in its early, imperfect stages industry-wide. Small businesses should start thinking now about how their phone lines, websites, and checkout processes handle a caller or shopper that isn't human.