Claude Cowork : La méthode complète (vous loupez le meilleur de l'IA !!)

This video introduces Cloud Cowork as a powerful, yet underrated, AI tool that can significantly enhance personal and professional productivity. The speaker emphasizes a pragmatic approach to using Cloud Cowork, focusing on how users can integrate it into their daily routines rather than just showcasing its features. The tool is described as a "game-changer" for 99% of people, not just coders or tech enthusiasts. While other AI tools like Cloud Code (for coding) and Open Cloud have generated significant buzz, Cloud Cowork, which was developed in just ten days using Cloud Code, is designed for everyday work and the general public. It's available as a desktop application for Windows and Mac, not as an online service or CLI, which the speaker suggests might be why some early adopters overlooked it. Cloud Cowork is presented as a personal "collaborator" rather than a mere assistant or chatbot. It operates on your computer, accessing your files, directories, and even your web browser to perform tasks. This collaborative nature means it works *with* you and *on* your PC, reading your files, managing work structures, and handling multitasking. It can process long tasks without interruption and can even be automated for recurring activities. The speaker highlights its ability to structure work, manage files across directories, and facilitate complex, multi-agent reasoning for tasks, such as creating a comprehensive project brief by subdividing the work among internal "sub-agents." The core of effective Cloud Cowork usage, according to the speaker, lies in the user's personal organization. Before diving into the tool's advanced features, users must first structure their own work. This involves creating a dedicated "workspace" directory on their computer, which Cloud Cowork will access exclusively. Within this workspace, users should create subdirectories for different aspects of their work, such as "archives," "outputs," "resources," and most importantly, "projects." The "projects" directory is crucial for organizing distinct work areas (e.g., "teams," "sponsors," "finances" for a sports club). Cloud Cowork can then be directed to work within a specific project folder, ensuring it has the relevant context. The speaker demonstrates how to set up this structure by creating a new workspace for a fictional basketball club. Users can either manually create these directories or ask Cloud Cowork to help structure the workspace, as it acts as a collaborator. Cloud Cowork uses markdown (.md) files to store contextual information and notes within each directory, which it refers to for ongoing tasks, effectively offloading mental load from the user. The speaker then clarifies the distinction between user-defined "projects" within the file system and the "Projects" menu item in the Cloud Cowork interface. The latter, added after the initial release, is merely for organizing conversations, not for structuring actual work files. This can be confusing, as it creates a "conversational memory" rather than a "work memory" tied to your file system. The speaker advises against relying on this feature if the user's file system is well-organized. Next, the video delves into Cloud Cowork's "superpowers," starting with its integration with Chrome. By enabling the "Cloud in Chrome" extension, Cloud Cowork gains the ability to use the user's *actual* web browser, complete with logged-in sessions (e.g., Facebook, LinkedIn). This allows it to perform tasks that require authenticated access, such as posting on social media or gathering information from sites that require login, mimicking how a human collaborator would use your computer. The speaker illustrates this by having Cloud Cowork browse a sports club's website to gather team information, even navigating to different pages not initially specified by the user. The second superpower is "Connectors," which allow Cloud Cowork to integrate with third-party tools. These connectors are categorized into three types: 1. **Web:** Direct API connections to online services like Google Calendar, allowing data exchange. 2. **Interactive:** Official partnerships with tools like Figma or Canva, providing deep integration where parts of the connected application can even appear within Cloud Cowork's interface. 3. **Desktop:** These are essentially plugins for the Cloud Cowork desktop application itself, adding local capabilities like navigating the file system beyond the designated workspace or generating PDFs directly. The speaker mentions a "File System" plugin for broader file access and a PDF creation plugin. The third and final superpower is "Skills." These are custom capabilities or competencies that users can teach Cloud Cowork. Unlike simple prompts, skills are enduring instructions that Cloud Cowork learns and can execute autonomously, improving over time. Users can create skills by providing instructions, uploading existing markdown files (allowing skill sharing among users), or, most effectively, by collaborating with Cloud Cowork to develop them. Cloud Cowork, leveraging its Cloud Code heritage, can even test, refine, and self-correct its own skills. The speaker provides an example of teaching Cloud Cowork to create interview preparation sheets, where the AI learned to first ask clarifying questions before generating the sheet. Skills can also be assigned "slash commands" for quick access within the interface. Finally, the speaker addresses "Plugins," which are often misunderstood. Plugins are not separate agents or job roles (e.g., a "marketing plugin" is not a marketing agent). Instead, they are "packages" or "bouquets" of pre-configured skills and connectors. While Anthropic offers official plugins (e.g., "Marketing," "Sales"), the speaker argues that users should create their own skills and connectors tailored to their needs rather than relying on generic packages. The only recommended plugin is "Productivity," which includes "Memory Management" and "Task Management" skills. These skills enhance Cloud Cowork's ability to understand and manage tasks and retain information across conversations, further reducing the need for manual organization. The video concludes by highlighting the "scheduling" feature, which allows users to program Cloud Cowork to perform tasks at specific times or frequencies (e.g., a daily "morning brief"). This requires the user's PC and the Cloud Cowork application to be running. In essence, the video argues that Cloud Cowork is a powerful "Uber collaborator" that can multiply productivity and quality by ten. The key to unlocking its potential lies in the user's commitment to structuring their own work, providing clear context within the designated workspace, and then leveraging the tool's capabilities (Chrome integration, connectors, and custom skills) to automate and enhance daily tasks. The speaker encourages sharing skills and best practices within the community to collectively advance productivity.