Anthropic, the research and safety company in AI, recently released its most developed model of flagship Claude: Claude 3.5 Sonnet and Claude 3.5 Haiku. But probably the most exciting innovation is "Computer Use."
This means that Claude will be able to work on screens, click on buttons, and type in text like a human having a conversation with a computer. And so, with this, for the first time in the history of humanity, industries whose workflows are complex could experience a change into unprecedented automation and efficiency.
Workflows Will Be Freed via AI Screen Interaction
A new "Computer Use" capability lets Claude do much more than just text-interact - it can even use its brainpower to function like a human being when using applications. Anything from opening software to interacting with user interfaces and filling out forms is now being done by the AI, which used to only have human hands.
As Chief Product Officer at Anthropic, Mike Krieger said, "This technology can do online research, manage repetitive tasks, and automate complex multi-step processes - that boosts productivity across sectors.".
Early Adopters Benefit
GitLab, Canva, and Replit are enjoying early benefits from Claude 3.5. The company GitLab reports, for instance, that with the new version, reasoning capabilities in its pipeline of software development are up by 10%.
In Replit, the belief is that Claude might be able to serve as a self-verifying verifier that helps smash open bottlenecks in coding projects. Canva is also exploring how it might use Claude more broadly to accelerate design work for its customers albeit greatly to their advantage.
How 'Computer Use' Works: Versatile Automation
What really stood out, though, was the fact that Claude could "see" a screen through screenshots in a way that would be impossible with traditional automation, according to VentureBeat. That means the AI can potentially perform tasks across applications and software environments.
In one demo, for example, Claude actually populated a vendor form by going through multiple systems, listing missing data, and filling out the form without requiring human intervention.
Such flexibility would mean Claude could transform sectors like finance, legal services, or customer support, where the worker needs to switch between applications multiple times. It would thus be able to perform jobs like running analyses on spreadsheets or retrieving customer information through CRM systems, completely minimizing human intervention.
Security and Privacy Safeguards
The capability of controlling a computer would be a security risk, but Anthropic has put very robust safeguards on this feature. It is only when developers introduce specific tools like the screenshot function and an action-execution layer that Claude may use a computer.
Systems to detect misuse have also been developed by Anthropic and sites that the AI should not access are blocked in order to give room for customer privacy and secured data.
The new feature is already in a limited public beta but can be accessed through an API that provides controlled environments to be tested by developers before its fuller release to the public.
The Future of AI-Driven Office Automation
This "Computer Use" feature by Claude marks the starting point for a new generation of office automation. It may, in the short run, lead to better productivity in entry work, customer service, and IT support, but over time, it will open wider implications.
Just think of AI handling advanced legal processes, compliance forms, or even assisting doctors by cross-checking patient records in medical databases. Well, the possibilities are endless, and Claude's new capabilities bring us one step closer to that future wherein AI comes closer to harmonizing every process of our digital workflow.
