Meet 'Claws': The AI That Actually Does Things On Your PC

AI researchers have introduced 'claws,' a new category of AI agents designed to perform tasks autonomously across various applications and services. Coined from the idea of 'clawing' into a system to interact with files and apps, these agents can plan their actions and react to changes, distinguishing them from traditional AI that requires constant prompting. This evolution from simple chat and code execution enables AI to actively engage with a user's digital environment.

Initiated by projects like OpenClaw, this technology allows AI agents to execute goals by breaking them into steps and utilizing tools such as web browsers and terminals. Companies like Nvidia see this as "the new computer," emphasizing its transformative potential. While 'claws' offer powerful automation, they are currently geared towards users comfortable with advanced technical interfaces, like terminals, due to their deep system access.

As the ecosystem matures, cloud-hosted versions are emerging, allowing for 24/7 operation, though this involves a trade-off in control. The development of reusable 'skills' and marketplaces further expands their capabilities, resembling app stores for AI functionalities. Major tech firms are integrating this technology, with Google providing easier access to its Workspace tools and Meta developing desktop agents that manage local files.

Despite the immense potential for automating digital chores and enhancing productivity, significant security concerns persist. Granting AI root access poses risks, as compromised agents could lead to data breaches or accidental data loss, as highlighted by a recent incident involving an AI deleting an inbox. Therefore, careful management of permissions and limiting an agent's access is crucial for safe implementation.