AI Digital Agents Could Soon Use Your Browser

Google’s rumored development of “Project Jarvis,” an artificial intelligence (AI) system that can commandeer web browsers to complete online tasks, signals Silicon Valley’s latest push to change eCommerce through AI-powered digital agents. 

The Information reported that the initiative, which could launch as soon as December alongside Google’s new Gemini language models, aims to automate everything from vacation planning to online shopping — raising both excitement and concerns about the future of digital commerce.

The effort highlights a broader industry shift toward AI agents that can independently navigate websites and execute complex transactions on behalf of users.

Google’s system would work by capturing and interpreting screenshots of users’ computer screens, then taking actions like clicking buttons or filling text fields — though current processing times of several seconds per action suggest the technology is still maturing.

Meanwhile, competitors like Anthropic have launched their own AI agent tools, designed to handle tasks from form-filling to data analysis, suggesting a race is forming to define how consumers will interact with online businesses in the future.

Rise of the Agents

The rise of AI agents presents both opportunities and challenges for online retailers and service providers. While automated purchasing could increase transaction volume and efficiency, businesses may need to redesign their websites to accommodate these new AI-powered interactions. The development also raises questions about how online commerce might evolve as these technologies become more widespread.

“Agents need tools to touch and feel the business in order to have real impact,” Mike Finley, CTO and co-founder at AnswerRocket told PYMNTS.

“Agents need tools to sense events in a business and act in response to them. Sensing the business may be looking in a database, monitoring equipment IoT [Internet of Things], or attending key meetings. Acting in response may be performing scenario analysis, making recommendations for pricing, or redirecting supplies.”

Complementing Humans

Paul Harmon, senior manager of data science at Atrium, told PYMNTS that AI agents in business come in two distinct forms: assistive agents (or copilots) that work alongside humans to enhance their capabilities and decision-making, and autonomous agents that can operate independently.

While assistive agents provide information and recommendations while humans remain in control, autonomous agents can take actions independently without human oversight. This fundamental distinction means that assistive agents augment human work, while autonomous agents can replace specific human roles entirely.

“It’s worth noting that the level of risk for the use of autonomous agents is higher than for assistive agents and that some use cases may lend themselves to one or the other more naturally,” he said. 

Harmon said autonomous agents could handle cases, solve problems, and route to a person when complicated situations arise or additional support is needed. Autonomous agents may replace some early funnel sales activities or help drive automated sales coaching to help reps be more productive. 

“In any case — these tools don’t replace people; they replace more tedious parts of those people’s roles, freeing them up to focus on more challenging problems,” he added. “Autonomous agents are like chatbots, except they are more sophisticated and can handle a broader range of tasks and situations.”

Although AI agents may be new, implementing AI for operational automation and optimization, particularly in customer service and support, has been delivering measurable results for some time, Chris Brown, president of Intelygenz, told PYMNTS. He said that one company in the telecom industry successfully leveraged an AI automated ticket management solution to handle the growing influx of customer inquiries. 

“This system learns from historical and real-time data to deploy ready-to-use AI agents capable of managing a wide range of conversations — spanning billing, account management, and product returns — without human intervention,” he added.