What happened
Anthropic's desktop app now allows turning user actions into repeatable instructions using video and commentary.
Why it matters
This change lowers the barrier to entry for task automation by allowing AI to be trained via a 'show, don't tell' method, which could significantly accelerate the adoption of assistants in professional workflows.
Anthropic's desktop application, Claude Cowork, has received a new update enabling users to record their screens while performing specific tasks. According to a report by The Decoder, users can accompany such recordings with voice comments explaining the sequence of actions.
Upon completion of the recording, the system analyzes the video feed and audio track, converting them into a ready-to-use skill that can be reused. This mechanism aims to simplify the transfer of complex workflows from humans to artificial intelligence without the need for writing code or textual instructions.
This feature marks a shift from text-based interaction to demonstrative learning, where the AI learns directly from the operator's actions. This opens up possibilities for rapidly automating routine operations that previously required detailed step-by-step descriptions.
Facts
- Anthropic updated the Claude Cowork desktop application.
- The new feature allows recording the screen while performing a task.
- Users can add voice comments to the recording.
- The system converts the recording and comments into a reusable skill.
- The information was published by The Decoder on July 21, 2026.
Context
The development of computer use capabilities is a key trend in modern language model development. Previously, such systems required clear textual commands, whereas the new approach relies on multimodal perception of the visual interface and speech.
What remains unknown
- How accurate is the recognition of user intent within complex interfaces?
- Will the feature be available for all types of operating systems?
- How is the privacy of data visible to the camera during recording ensured?
AI analysis
The implementation of screen recording with voice-over indicates a strategic shift from programming skills to their natural demonstration. This could make AI agents more accessible to users without technical skills, but it also creates risks of reproducing user errors if the system misinterprets the context of actions.
Strategic AI conclusion
A likely consequence will be an increase in user libraries of skills for specific corporate tasks. The next observable signal will be the emergence of tools for editing and debugging skills created in this manner. The main uncertainty remains the model's ability to generalize demonstrated actions to slightly different conditions without additional instructions.