Think about how much time you spend navigating software systems, entering data and uploading documents. Think about the routine tasks that require significant time and concentration but no ingenuity. Think of how often you must remember how to work through antiquated or poorly designed software simply because the costs of upgrading or switching are too high. While technology can change processes for the better, it can still create processes that make our work more difficult or less rewarding.
OpenAI opened up a new era of digital AI-based technology a couple of weeks ago with the research preview of their latest offering, Operator. OpenAI, one of the most influential players in the AI model development industry, calls this product a “Computer-Using Agent” or a “CUA.”
This name, no doubt an effort to clarify the confusing term “AI agent,” is currently available only to U.S.-based users who choose to pay for OpenAI’s $200/month Pro plan. Given the proper instructions (i.e., written prompts), Operator navigates to websites and uses them as a person would. It can click, type, and scroll through pages just like a person would, entering data and fulfilling your instructions. If AI chatbots are our research assistants who can only answer questions, Operator is our full-fledged assistant who can complete tasks on the web. Demonstrations on OpenAI’s Operator introduction page show the Operator booking online campsites, buying ingredients for a recipe using Instacart, and scheduling appointments.
This kind of technology means that our time spent navigating software and websites is about to change drastically. Instead of navigating to all the specific places in your case management software to enter data, you could now provide access to your case notes and have your CUA navigate the system and enter the data, asking for your approval whenever it is about to complete a critical step such as saving a form. Instead of booking your own appointments, initiating messages sent to your clients, and scheduling services, you could have your CUA do it. CUAs could potentially post documents in other systems and make multi-system input a much faster process. In other words, we can shortcut the time we spend waiting for our software systems to improve because the machines will start using it for us.
It’s important to understand that Operator and other CUAs will not be 100% reliable. Operator has integrated all sorts of compensatory processes, which help us control the scope in which it is allowed to work. This release is considered a “research preview” for a reason, and it will take time for this technology to mature. However, even with this early version, we can start to see many of the possibilities that are on the horizon while anticipating the pitfalls. And as always, we have the choice to run and hide or to dive in bravely.