OpenAI is taking another leap toward mainstream AI adoption with the rollout of a new ChatGPT agent — a feature designed to perform complex, multistep tasks on behalf of users with minimal supervision.
Unveiled during a livestreamed event, the ChatGPT agent aims to simplify both personal and professional projects. Think: planning a meal, ordering all the necessary ingredients online, or drafting a business presentation from scratch.
The feature builds on two tools OpenAI introduced earlier this year:
- Operator — capable of browsing, clicking, and typing across the internet like a human.
- Deep Research — a system designed for in-depth, time-consuming online research.
By combining these capabilities, the ChatGPT agent can now execute detailed, goal-oriented tasks — from shopping for vintage lamps on Etsy and filling your cart, to creating early drafts of PowerPoint presentations that users can later refine.
Who Gets Access?
The new features will be available immediately to Pro, Plus, and Team subscribers, with a wider rollout planned for enterprise and education customers later this summer.
Why AI Agents Matter
Tech leaders, including OpenAI CEO Sam Altman, have called AI agents “the next giant breakthrough” — promising to save time and boost productivity by handling repetitive, multistep workflows.
The AI agent uses a virtual computer and web browser to perform tasks, adjusting its approach based on user feedback mid-process. However, OpenAI notes that while ChatGPT typically responds in seconds, agent-level tasks can take several minutes to complete.
Early Use Cases & Limitations
In a live demonstration, product manager Neel Ajjarapu tasked the agent with:
- Searching Etsy for vintage-style lamps under $200 with free shipping.
- Adding the best ones to the cart.
- Providing purchase links for each.
The tool has also been tested for drafting PowerPoint slides — though Ajjarapu cautions it’s best for “early rough drafts”, not polished presentations.
Safety First
With great capability comes risk. AI agents can be misused or make errors, so OpenAI has implemented safeguards:
- Refusing certain tasks (financial or legal advice).
- Seeking user permission for sensitive actions, such as purchases.
- Requiring supervision for tasks like sending emails.
Room for Improvement
OpenAI’s Chief Product Officer Kevin Weil admits the technology isn’t perfect yet:
“It is far from perfect. But if we had gone back six or 12 months ago and said this was going to be possible today, we would have been pretty excited about it.”
As the AI agent matures, it could redefine how people shop, research, and prepare for work — bringing us closer to a future where AI handles the busywork so humans can focus on higher-value tasks.