{
"id": "p2qmX6TM0kw",
"source": "https://www.youtube.com/watch?v=p2qmX6TM0kw",
"uploader": "How I AI",
"duration": 2768,
"published": "2026-08-31",
"language": "en",
"transcript_source": "native_subtitles"
}Daniel Bloom, a Product Manager at Melio (B2B payments), shares how he built a comprehensive Claude Cowork-based productivity system that manages his entire PM workflow — from task tracking to meeting prep to self-improvement loops.
Daniel argues that any AI system becomes truly powerful when it satisfies two conditions:
- Self-rewriting core files — the system can modify its own instructions and improve over time
- Deep integrations — connected to as much of your work ecosystem as possible (Notion, Slack, email, calendar, meeting transcripts)
He emphasizes this isn't specific to Claude Cowork — the same principles apply to Codeex or ChatGPT Work.
Claude built and manages a Notion board with three sections:
- Top of Mind — large strategic initiatives
- This Week — weekly priorities
- Inbox — items from Slack, email, calendar picked up by Claude
Daniel describes Notion as essentially read-only for him now. He only checks it for focus; Claude manages everything.
Every Sunday morning, a recurring "Weekly Prep" task runs automatically. It:
- Pulls data from Notion, calendar, Slack, and Granola meeting transcripts
- Recommends top-of-mind additions/removals
- Suggests weekly priority moves
- Fills the inbox with new items
- Preps for upcoming meetings (categorizing them by prep level needed)
The daily Morning Brief is where "the real magic happens." It:
- Reviews yesterday's meetings using Granola transcripts, extracting one-liners and action items
- Proactively flags unknown terms — when Claude encounters jargon, milestones, or files it doesn't understand, it asks Daniel to explain and saves them to context
- This creates a self-maintaining knowledge base of company-specific terminology
Daniel estimates 70-80% of his computer time now flows through Claude Cowork alone. The reason: anything not done through Claude loses context capture. Centralizing work in the AI system is key to ROI.
Early weeks are frictionful and slow — the system doesn't know you, makes mistakes, and you double-check everything. But value compounds tremendously over time. Daniel says he now does in a day what used to take a week, and more importantly, does deeper work — more research, more data-backed decisions, less operating on hunches.
A scheduled weekly task with four components:
- Draft Analysis — Claude compares drafts it wrote with what Daniel actually sent, learning the gap
- Skill Suggestions — identifies recurring patterns in Daniel's work and suggests new skills to build
- Friction Tracking — each skill has built-in feedback collection; weekly review surfaces top frictions for fixing
- "Improve" Skill — a critical auditor for the flood of AI tips from X/LinkedIn/newsletters. Daniel shares articles and asks: is this real? Is it needed? How does it fit my current system?
Daniel built a "Workstation" plugin (with colleagues) that onboards any Melio employee into a personalized Claude setup in ~15 minutes. It:
- Connects all tools automatically
- Confirms roles and maps colleagues/management
- Analyzes calendar and Slack patterns
- Captures the user's writing voice
- Sets goals and gets them productive immediately
Key UX lesson: they tried Claude Code first (terminal-based, 3 days to set up) and switched to Cowork for simplicity. Good UX is critical even for internal tools.
What's missing:
- Cloud/headless operation — Claude needs the computer open. He wants it running autonomously (mentions OpenClaw and Hermes as examples of this pattern)
- True autonomous task execution — he's building toward Claude resolving simple tasks (scheduling, file lookups, answering common questions) without human involvement
- In-depth research and customer/user conversations
- Improving the AI system itself (researching new techniques, implementing improvements)
Daniel admits to "Claude rage" — getting frustrated and raising his voice (via Whisper voice input) when Claude doesn't perform. He jokes that Whisper doesn't capture his passion well enough.
- Claude Cowork for PMs: My Self-Improving Productivity System
- Create a Meta-Workflow to Continuously Improve Your AI Assistant's Performance
- Build a Self-Improving AI Morning Brief to Capture Action Items and Learn Company Jargon
- Automate Your Weekly Planning with an AI-Powered PM Assistant
- Claude
- Notion
- From a $6.90 newsletter to $3M API: How a non-coder built Memelord | Jason Levin
- How the founder of Morning Brew built a Claude content machine | Alex Lieberman
- Daniel Bloom — LinkedIn
- Daniel Bloom — Website
- Claire Vo — ChatPRD
- Claire Vo — Website
- Claire Vo — LinkedIn
- Claire Vo — X
- How I AI — Website
- Optimizely
- Jira
- Penname (production)