OpenAI has introduced a new feature for ChatGPT – customisable AI agents – designed to perform tasks on behalf of users, marking a major step forward in the company’s ambition to integrate AI more deeply into everyday productivity and workflows.

Dubbed simply as “agents”, the new capability enables GPT-4-turbo to not only understand and generate text, but also take actions across apps and services, making it possible to automate complex, multi-step tasks. These include actions such as booking travel, managing files, retrieving data from emails, and even running code—all with minimal user input.

According to OpenAI, users will soon be able to create and deploy agents tailored to specific needs, such as scheduling meetings, conducting research, or interfacing with third-party applications through APIs.

“Eventually, you’ll be able to say ‘Hey GPT, book me a flight to San Francisco and find a hotel near the venue’—and it’ll handle the logistics,” OpenAI said in its announcement.

How It Works

These agents build on existing capabilities such as Code Interpreter (now called Python), DALL·E image generation, and browser tools—fused into a more autonomous and persistent AI experience.

Developers can define an agent’s instructions, memory, APIs to call, and even the ability to browse the web. Once configured, agents can act independently within their scope, significantly reducing the time users spend on repetitive or operational tasks.

An agent, for instance, can read a user’s recent emails, extract information, and summarise it in a document, or even respond automatically based on predefined preferences.

Still in Early Access

OpenAI noted that the agent feature is currently being tested with early users and developers and will gradually roll out more widely. The company is also building a marketplace for these agents, akin to an “app store” where users can discover and deploy agents built by others.

The addition of agents signals OpenAI’s shift from ChatGPT as a passive assistant to an active AI collaborator – blurring the lines between chatbot and digital employee.

This evolution poses significant implications for productivity apps, task automation platforms, and even enterprise software. By enabling AI to interface directly with apps and take real-world actions, OpenAI is laying the groundwork for a new category of AI-native tools.

Privacy and Control Still Under Watch

Despite the promise, the rollout also raises questions about user trust, data security, and permission boundaries, particularly as agents gain more autonomy. OpenAI said it is working on robust controls and transparency features to ensure users remain in charge of their digital environments.

More technical documentation and safety guidelines will be released in the coming weeks as OpenAI expands developer access.