Companies have been racking their brains for years to automate repetitive processes. Anthropic AI wants to be your solution

Putting nuts on is no fun, but assembly line workers have been doing it for years. Robots eventually replaced humans in many of these processes, and for years they have been trying to do the same with tasks that are performed in front of a computer and are equally repetitive.

Robotic Process Automation (RPA). This discipline combines the use of APIs (application programming interfaces) with interaction with a computer’s user interface. The goal is to perform repetitive tasks in productivity and primarily in business applications.

Simulate human interaction. These processes attempt to complete autonomous execution of various actions and transactions in all types of software systems. The idea is to simulate and replicate human user interactions, thereby avoiding the time wasted on more complex tasks and thus providing a greater return on employee investment.

RPA wasn’t an AI (but now it is). Until now, this type of automation could not be considered an artificial intelligence system, since RPA is process-based and AI is data-oriented. Anthropic and its new Computer Usage feature have just shown that both things can be combined and interactions with the user interface can be automated to perform all sorts of repetitive tasks.

This image shows the Computer Usage AI feature filling out a form by getting information from a CRM system. The dream of many companies comes true.

Anthropic understands this clearly.. Anthropic CEO Mike Krieger discussed the capabilities of this feature in an interview with VentureBeat. He explained that “we expect this to be particularly useful for tasks such as online research, performing repetitive processes such as testing new computer programs, and automating complex multi-step tasks.” And who likes to automate complex tasks specifically for companies?

Business sweetie. The Computer Usage feature is able to recognize what is displayed on the screen and act accordingly. The level of automation is promising for sectors such as finance, as well as legal services and customer support. Krieger explained that “Claude can open spreadsheets, do analysis, and create visualizations. In terms of customer service, he will be able to navigate CRM systems to find and update information about any customer.”

And huge potential. Features like this are a promising solution for data entry or customer service and help desk applications, but this automation can go much further as it can recognize and analyze the information you see on the screen and then act on it. This is exactly what AI agents promise, and this could be a great example of their capabilities.

But. For now, Computer Usage has limited availability, and as Anthropic explains, “it can’t use your computer as such.” The tool is available as part of an isolated environment in which we can perform all types of tasks using mouse movements and autonomous keystrokes generated by artificial intelligence. Despite this, Anthropic has developed mechanisms to detect whether this feature is being used for dangerous activities, such as possible malicious interactions with government portals. The power of this tool can of course also be used for malicious purposes, as we see with generative artificial intelligence and deepfakes.

In Hatak | Microsoft is starting to offer autonomous AI agents. At the moment these are hypervitaminized IFTTT recipes.