AI agents can do your tasks: How, and how much should you let them?

The Australia incident illustrates a complication that comes with delegating tasks to increasingly capable AI systems: how much of the process does a user hand over along with the task?

What changes when consumers delegate to AI?

For years, consumers have relied on AI to ease their lives in incremental ways: seeking a recommendation, getting answers to search queries, finding the fastest possible route to a destination, or translating a piece of text. In most such interactions, AI helps the consumer make or carry out a decision, but the decision to act on it rests with the consumer.

AI agents change this relationship by allowing users to hand over the task itself. A user can set a goal for the agent to achieve, while leaving it to figure out the steps required to achieve it. A user may now ask an agent to plan a holiday, from finding suitable flights and drawing up an itinerary to making reservations on their behalf.

An AI agent vs an AI chatbbot

In a 2021 paper titled Consumers and Artificial Intelligence: An Experiential Perspective, Professor of Marketing at Wharton, Stefano Puntoni and others examine this “delegation experience”, in which AI performs a task that a consumer would have performed themselves. The paper argues that such delegation can be experienced in two opposing ways: as empowering, when AI enables consumers to accomplish their goals, and as replacing, when handing over a task diminishes their sense of autonomy or control.

This makes control integral to the decision to delegate. What consumers are willing to delegate depends on factors including their trust in the AI and its perceived competence, Puntoni told The Indian Express. He described control as “the other side of the delegation coin”.

The more an agent can do independently, the less closely a user may need to direct or monitor how it gets done.

So what are consumers actually handing over?

Some early evidence of what users are actually willing to delegate comes from Perplexity’s Comet browser. A study of its use examined hundreds of millions of anonymised interactions between July and October 2025, allowing researchers to observe what users asked its agent to do, rather than what they said they might be willing to delegate.

The study counted an interaction as “agentic” only when the agent took control of the browser or acted on an external application, excluding ordinary searches and interactions that merely exchanged information.

The delegated tasks were mundane, ranging from researching and editing documents to searching for products and managing account settings. More than half (55%) of agentic queries were for personal use, compared with 30% for professional use and 16% for educational purposes.

However, the study does not show that consumers at large are ready to delegate these tasks. Comet was then a new product, and its users were early adopters who may have been more technologically inclined than the wider population. Moreover, observing what people delegated cannot, on its own, tell us why they were comfortable handing over those tasks, or which ones they were unwilling to hand over.

Willingness to delegate is unlikely to be uniform across tasks. Puntoni said there would be areas where most consumers remain reluctant to hand decisions to an algorithm, while some people may resist using AI agents in most situations. He expects this reluctance to ease as consumers build what he called “calibrated trust” through experience with agents.

“As AI agents demonstrate their usefulness, consumers will be more likely to adopt,” Puntoni told The Indian Express.

He cited the example of autonomous vehicles such as Waymo, saying people sceptical of the technology can change their minds after experiencing it themselves.

There may be limits to how much autonomy consumers want an agent to have. Research cited in The Wharton Blueprint for AI Agent Adoption, co-authored by Puntoni and AI researcher Thomas McKinlay, suggests that people were more willing to adopt AI agents with a moderate level of decision-making autonomy. Too little autonomy made the agent seem less useful, while too much reduced users’ sense of freedom and control.

Will consumers choose to adopt agents?

Consumers may not consciously choose to “adopt” an AI agent. Puntoni said adoption could instead happen gradually through products people already use.

“Agent adoption may in many situations be a relatively slow process where consumers learn to delegate more and more tasks as new automation features become embedded in products over time,” he told The Indian Express.

He cited Microsoft adding agentic features to products such as Office as an example of how this could happen.

Josh Miller, CEO of The Browser Company, argues that AI agents may be more useful as an underlying technology than as a stand-alone consumer product. In an interview with WIRED this month, he pointed to a personalised morning briefing in the company’s Dia browser, which he said was its most popular feature. An AI agent powers the feature, but users do not need to know that, he told WIRED.

As agentic capabilities become embedded in products consumers already use, what it means to be in control may also change. The Wharton Blueprint points to the ability to edit, pause, stop or reverse an agent’s actions as ways for users to retain control after delegating a task.

Whether users need to direct an agent closely to feel in control remains an open question.

“The notion of extended self might suggest that people may grow to see AI agents as an extension of themselves, and if so, perhaps experience a relatively high sense of control even in the case of highly autonomous agents,” Puntoni told The Indian Express. “It’s very speculative though, and I’m not sure how likely that is.”