Product Manager Handles 70-80% of Workday with Claude
Daniel Blum shares how he developed his personal AI assistant.
1Core Architecture of the AI Assistant
Daniel Blum, a product manager at Melio, delegates 70-80% of his daily tasks to an AI assistant named Claude. The system manages his Notion board, prepares his weekly plans, and scans Slack and emails to capture important context. Blum emphasizes that the system's real power lies not in the AI tool itself but in its ability to update its own core files and integrate seamlessly with existing tools.
2Continuous Context Updates
Blum spent months feeding Claude voice memos, links, presentations, and verbal brain dumps to create detailed context files for every aspect of his work. These files are updated every few weeks, ensuring Claude's knowledge stays closely aligned with real-time company developments. This continuous update mechanism keeps the AI informed and relevant.
3Identifying the Unknown
One of Claude's most impressive features is its ability to identify what it doesn't know. Each morning, it scans Slack, emails, and notes for unfamiliar terms, projects, or goals missing from its context files. For instance, when it encountered the phrase 'settlement cap,' it had already read the relevant thread and grasped the general idea, only needing Blum to confirm the meaning before saving it.
4Slow Start, Big Payoff
Blum admits the first few weeks can be frustrating. The system lacks sufficient knowledge, and its outputs often require minor tweaks. However, once the groundwork of centralizing information and building context is done, the payoff is substantial. Blum can now complete a week's worth of PM work in just one focused day, thanks to his personalized AI system.
5Self-Improving Loops
Blum developed a weekly skill that compares Claude's original drafts with his final versions, using the differences to improve future work. This system, resembling Alex Lieberman's 'write like me' loop, relies less on explicit feedback and more on observing Blum's actual edits. It learns instinctively from small adjustments, refining its outputs over time.
6Feedback Telemetry
Every one of Blum's skills captures moments of friction. If something isn't working or needs correction, the system logs these signals. A weekly improvement loop then identifies common issues and suggests updates. This functions like analytics for internal tools, providing a structured way to refine the system based on real-life performance.
7Simplifying AI Adoption
Blum addressed one of the biggest barriers to AI adoption within companies: personalization. Observing other PMs struggle with Spectacular—a tool tailored to his workflow—he created an onboarding flow that connects each employee's tools, maps their colleagues, and learns their voice in about 15 minutes. This shared foundation, personalized to individual needs, eliminates the need to start from scratch.
8The Future of AI Systems
Blum estimates Claude currently handles 70-80% of his workday. A remaining limitation is its inability to operate autonomously in the cloud when his computer is off. Preparing for this future, Blum teaches Claude to recognize the 'closed state' of tasks. For example, if a drafted Slack message is no longer saved, Claude infers it was likely sent. When full autonomy arrives, Blum's system will already understand what completion looks like.
Original article: Lenny's Newsletter — Ürün Yöneticisi, İş Gününün %70-80'ini Claude ile Yönetiyor