The video explains Claude’s shift to a cloud-first AI operating system that runs tasks in virtual containers, emphasizing new workflows for managing context, projects, and automation through structured skills and agents to optimize business processes. It highlights the importance of integrating SaaS tools, organizing data effectively, and leveraging hybrid local-cloud setups to build flexible, efficient AIOS solutions while encouraging iterative skill development and community engagement.
The video discusses the recent major update to Claude’s co-work platform, which has transitioned to a cloud-first model, fundamentally changing how AI operating systems (AIOS) are built and managed. Mansel, the presenter, explains that previous setups are now obsolete, and users must adapt to the new system where tasks run in virtual containers on Anthropics’ cloud, enabling always-on functionality without needing local machines to be active. However, this shift means local-only modes for co-work tasks no longer exist, and files required for tasks are uploaded to the cloud temporarily, with changes during sessions not syncing automatically. Mansel emphasizes the importance of understanding these changes to build effective AIOS setups that maintain current context and optimize token usage.
The foundation of building an AIOS lies in managing data and context effectively. Mansel outlines three key components of context: knowledge (curated facts about the business), state (dynamic information stored in databases), and memory (experiences learned by the AI over time). He guides viewers through connecting their SaaS tools via MCP connectors to pull in existing context, such as email voice and business information, and using onboarding skills to extract and organize this data into local folders. This groundwork ensures portability and flexibility across AI models and platforms, preventing vendor lock-in and enabling more accurate and personalized AI interactions.
Projects in Claude have been overhauled to support a structured framework based on four business pods: acquisition, delivery, support, and operations. Mansel demonstrates creating a project aligned with acquisition, adding context files, and explains the new agentic workflow involving a coordinator (project manager) and multiple worker threads that operate in isolated contexts to efficiently handle complex tasks. This design improves token efficiency and task management, allowing users to oversee and steer individual threads. He also highlights the importance of linking local folders for hybrid workflows and the implications of cloud-only versus local data access on task execution.
Skills and agents form the core of automating business processes within Claude. Mansel explains that skills are detailed, repeatable procedures that guide AI to perform tasks consistently and reliably, contrasting with ad-hoc agent commands. He introduces the pod mapping skill, which audits business workflows to identify pain points and automation opportunities, helping users design precise skills tailored to their needs. Skills include deterministic scripts and reference materials to ensure consistent output quality. Plugins, which bundle multiple skills, are also covered, along with the marketplace for discovering vetted skills and plugins. Mansel stresses the iterative nature of skill development to achieve operational excellence.
Finally, the video covers the interface and operational layers of Claude’s AIOS. Users can access their projects and context across desktop, cloud, and mobile interfaces, with some limitations on local folder access on mobile. Scheduled tasks and routines enable automation of recurring workflows like daily briefs, with options for notifications and permissions. While monitoring tools are currently limited for consumer plans, enterprise users have more advanced capabilities. Mansel concludes by encouraging viewers to focus on building grounded, business-relevant skills and workflows, offering additional resources and community support for those looking to implement AIOS solutions effectively.