Summary
Product manager Daniel Bloom joins the How I AI podcast to walk through a Claude Cowork (Claude.ai Projects) system he built that compresses a week of PM coordination work into a single day. The system goes significantly beyond a standard morning brief: it ingests recent Slack messages, emails, personal notes, and connected data sources, then actively surfaces gaps in its own context — asking Bloom to clarify unfamiliar files, milestones, or goals so it can update its persistent knowledge base rather than silently working from incomplete information.
Bloom describes a layered architecture where Notion serves as a read-only priority surface managed entirely by Claude, while Jira, Slack, and browser access (via Chrome connector and MCPs) feed real-time context into the system. A key design principle is routing as much work as possible through Cowork so that context is captured consistently — he estimates he now manages 70–80% of his computer time through the interface. The self-healing element refers to Claude proactively prompting Bloom to define ambiguous terms and save them to its context rather than accumulating silent misunderstandings over time.
For teams evaluating enterprise Claude deployments, the video offers a practical look at what a mature single-user agent workflow looks like in a real product organization, including honest discussion of which connectors still require browser fallback versus native MCP integrations, and how the system scales — or doesn’t — when extended to teammates beyond the original builder.
📺 Source: How I AI · Published August 31, 2026
🏷️ Format: Workflow Case Study
