Jan 23 (Reuters) - Generative artificial intelligence heavyweight OpenAI on Thursday

previewed an AI agent that can carry out tasks on the web for users, as it seeks to enhance its

chatbot amid intensifying competition.

The tool, called Operator, is powered by a model that allows it to interact with on-screen

buttons, menus and text fields.

"This capability marks the next step in AI development, allowing models to use the same

tools humans rely on daily and opening the door to a vast range of new applications," the

company said in a blog post.

Operator can perform a variety of tasks, like creating to-do lists or assisting with

vacation planning. It also takes user input once it decides a task is complete and seeks

confirmation for some tasks, such as entering login details on a website.

The tool is currently available to Pro users in the U.S. as a research preview, the

Microsoft ( MSFT )-backed startup said.

Agents, which are systems that can execute actions such as making purchases and scheduling

meetings without direct human intervention, are now at the forefront of companies' AI agenda.

OpenAI competitor Perplexity launched an agent-based assistant for Android devices earlier

on Thursday. This assistant can book dinner reservations, hail rides on apps and set reminders,

among other tasks.

Last year, Apple incorporated Apple Intelligence into its voice assistant, Siri, and - in a

partnership with OpenAI - the iPhone maker also introduced the use of ChatGPT, with user

permission.

While such agents had long been elusive to researchers, the emergence of step-by-step

reasoning approaches like those used in OpenAI's o1 model could make such tasks possible,

business executives told Reuters in December.