Anthropic has released a major update to its Claude Cowork AI agent, enabling users to teach the system new tasks through screen recording and voice narration. This feature, currently available on the macOS desktop app, allows the AI to transform human demonstrations into repeatable, autonomous workflows.
Teaching Claude via macOS screen recordings
Users can now train the Claude Cowork agent by navigating to the "+" menu in the macOS desktop app and selecting "Record a skill." By recording their screen while narrating their actions, users can show the AI how to execute specific tasks, such as formatting spreadsheets or navigating complex web-based workflows. As reported by HeadTopics.com, this capability allows the AI to identify individual steps and create reusable skills that can be triggered with a single command.
This update builds upon Claude Cowork’s existing "computer use" features, which allow the agent to see the screen and control a mouse and keyboard to manage files or edit documents. By adding the ability to learn from demonstration, Anthropic is moving the agent closer to a tool that can handle specialized, non-standardized professional tasks without requiring manual prompt engineering.
The $20 to $200 pricing ladder for skill learning
The new skill-learning feature is integrated into several Anthropic subscription tiers, including Pro ($20/month), Max ($100–$200/month), and Team ($30/user/month) plans. According to Anthropic's official Claude AI account, the feature is designed to help users codify repetitive tasks that can be scheduled or run on demand. This allows for a level of automation that goes beyond simple chat interactions.
However, there are caveats to the rollout. Because executing these learned skills consumes significantly more compute resources than standard text-based chat, Pro users may encounter usage limits more quickly than those on higher-tier plans. Additionally, Anthropic notes that Cowork currently runs sessions remotely in beta, meaning users can theoretically start a task on one device and monitor its progress from another.
A personalized challenge to Microsoft Copilot and Google Gemini
Anthropic is positioning Claude Cowork as a highly personalized alternative to major competitors like Microsoft Copilot and Google’s Gemini. While many AI assistants rely on text-based instructions, Anthropic’s emphasis on learning through demonstration offers a way for professionals to train the AI on highly specific, idiosyncratic workflows without writing a single line of code .
This move signals a broader industry shift from passive AI assistants toward "agentic" AI—tools that act more like junior employees capable of observing, learning, and executing long-running tasks autonomously. By focusing on the ability to "watch and learn," Anthropic is betting that personalization will be the deciding factor in the race to create the ultimate digital worker.
Security risks and the Windows support gap
Despite the productivity gains, Anthropic has cautioned that the agentic nature of Claude Cowork introduces unique security risks, particularly regarding the agent's ability to access the internet. the company recommends that users avoid performing tasks involving sensitive information while recording skills, as the AI may access various files and web services during its autonomous execution phase.
There are also practical limitations regarding platform availability. While the feature is currently exclusive to macOS, the timeline for Windows support has not been specified. This leaves a significant portion of the professional user base waiting to see if this level of demonstration-based automation will eventually bridge the gap to other operating systems, or if macOS will remain the primary playground for Anthropic's most advanced agentic features.
Comments 0