OpenAI has launched ‘ChatGPT Agent’, which the company claims can “complete complex online tasks on your behalf”, including user requests like “plan and buy ingredients to make Japanese breakfast for four”.
The agent can “intelligently navigate websites, filter results, [and] prompt you to log in securely when needed”, it said, such as when it is necessary to sign in to a supermarket account or enter payment details.
Agent brings together prior releases – Operator and Deep Research. The former could be asked to “handle a wide variety of repetitive browser tasks” like ordering groceries, while the latter specialises in in-depth research on the public internet.
“We brought the best of both together,” OpenAI said this week.
“It can now actively engage websites – clicking, filtering, and gathering more precise, efficient results. You can also naturally transition from a simple conversation to requesting actions directly within the same chat,” the company added.
While the agent is carrying out the task, it provides an “on-screen narration” to give “visibility into exactly what ChatGPT is doing”. Users can interrupt and take control of the browser whenever needed.
ChatGPT Agent began rolling out this week to users with Pro, Plus, and Team subscriptions, including those in the UK.
The Verge’s senior AI reporter Hayden Field tested Agent and described it as like “a day-one intern who’s incredibly slow at every task but will eventually get the job done. Well…most of the job.”
Setting it the task of buying some flowers for a friend, Hayden found it good at researching which bouquet to buy and from where, but the tool “struggles to automate the most frustrating parts – actually filling out delivery details and making the purchase.”
Understanding AI newsletter writer Timothy Lee gave the tool a scan of a hand-annotated, printed shopping list, which it transcribed well, before initially struggling to log in to the Kroger website. However, it eventually completed the order.
He concluded that “ChatGPT Agent is dramatically better than its predecessor at grocery shopping. But it still made mistakes at this task. More broadly, the agent is nowhere close to the level of reliability required for me to really trust it.”
OpenAI said users should “expect continued improvements to ChatGPT Agent’s efficiency, depth, and versatility over time, including more seamless interactions as we continue to adjust the amount of oversight required from the user to make it more useful while ensuring it’s safe to use”.
ChatGPT is drawing huge interest from the grocery sector, having earlier this year introduced product recommendations and purchasing options to the chatbot. The new function means users asking questions such as “what’s a quick healthy meal for less than a fiver?” are not only served suitable recipes, but online stores where the ingredients can be bought, and a ‘buy’ button to take them straight there.
OpenAI is currently seeking to partner with retailers and was “exploring an easy way for merchants to provide product feeds directly to ChatGPT, helping ensure more accurate, up-to-date listings”, it said. Last week, the Financial Times, citing multiple people familiar with the proposals, reported OpenAI was planning to take a cut from online product sales made directly through ChatGPT, and was aiming to integrate a checkout system into the platform.
According to Similarweb, the ChatGPT website was last month the fifth most visited in the world. A March Adobe Analytics report measured a 500% increase in UK consumer traffic heading to retail sites from AI chatbots over the previous six months.
No comments yet