Matt Lane
Transcript of video
Kia ora,
In this video I will showcase capabilities in the latest version of ChatGPT Pro that you may not yet be aware of, specifically, OpenAI’s new tool called Operator. The difference with Operator is that while classic ChatGPT with web search can retrieve and summarize information, Operator can actively perform actions within a browser. Operator is more like the classic “robot agent”, not just answering questions but actually doing stuff.
To illustrate this change, I’ve chosen 3 examples of how external customers might interact with our services using AI-driven automation. Obviously in the future this kind of tool will also be able to be used by staff, but to avoid touching any sensitive data or internal systems from the backend, this video just interacts with existing public digital services.
Example 1: Finding Rubbish collection information
In this scenario, I use ChatGPT with web search enabled to determine my rubbish collection day and whether it is a glass or mixed-plastics collection week. While ChatGPT successfully searches the web and provides instructions on how I can find this information, it does not directly provide the answer.
By contrast, Operator opens a browser window and actively navigates the Wellington City Council website. As seen in the demonstration, there are two cursors: mine on the left and the AI-controlled cursor on the right. Operator autonomously locates the relevant page and retrieves the specific rubbish collection details, demonstrating its ability to perform tasks rather than just guide users to the correct information.
Example 2: Reporting a tree with hazardous branches
The second example involves reporting a tree that has dead branches requiring trimming at a specific location in Wellington. Using ChatGPT, I receive step-by-step instructions on how to report this issue to the Council. While helpful, this still requires me to manually follow those steps.
Using Operator, however, the AI takes control of the browser, finds the appropriate page on the Council website, scans the content to understand the required steps, and begins logging the issue on my behalf. The AI enters the necessary details and, before submitting, prompts me for confirmation. Once I approve, the AI submits the request automatically. This illustrates how Operator can perform administrative tasks autonomously while allowing me to focus on other work.
Example 3: Paying an invoice from the Council
For this final scenario, I imagine a future use case where Operator could process invoices received via email. Suppose I receive an invoice from the Council for $1.50 to pay for a smart cat collar, with the invoice number ‘CoolCat.’
Using ChatGPT, I receive a response directing me to the correct payment page, instructing me on where to click and what to do next. However, Operator takes this a step further by actively navigating to the Council website, searching for the ‘Pay Invoice’ page, selecting the appropriate category (Cats > Smart Cat Collar), and completing the online form with details gathered from my input.
At a later stage, when prompted for personal information such as my name, address, and email, I choose to provide these details. Operator enters them on my behalf and advances to the payment gateway. When asked for my credit card details, I decline, as I’m not personally ready to enter my credit card details into a free text field. However, envisioning the future, I could see a scenario where I pre-authorize my payment details and receive a simple text message asking if I approve the payment. Something like a confirmation text reply could then trigger the entire transaction while I am on the go.
Conclusion
These examples illustrate how AI-driven automation is rapidly transforming the way people interact with online services. While ChatGPT provides helpful guidance, Operator demonstrates a leap forward by actively performing tasks on behalf of users. This is reminiscent, but better than, the promises made by Robot Process Automation or RPA in the past. As this technology continues to develop, we should anticipate and prepare for a future where customers, and probably staff too, increasingly rely on AI to navigate and complete transactions seamlessly. Understanding these advancements will help ensure that services remain accessible, efficient, and aligned with evolving user expectations.
