How I AII built a Claude Cowork system that does a week of PM work in a day
CHAPTERS
- 0:00 – 4:14
From PM overhead to “a week of work in a day”: the problem Daniel set out to solve
Daniel explains that classic PM overhead—coordination, action items, Slack/email chaos, and constant context switching—was blocking deep work. Cowork became the first system that meaningfully reduced that overhead while improving the quality of his outputs.
- •PM work is dominated by coordination overhead, not just specs and research
- •The core pain: keeping track of endless tasks across Slack, meetings, and email
- •Goal: reclaim focus for deeper, more data-backed work
- •Claimed outcome: compressing a week of PM work into a day
- 4:14 – 5:20
Daniel’s constraints-first AI stack at Melio (and why Cowork was the unlock)
Daniel describes how his tool choices were largely determined by Melio’s enterprise stack (Gemini, then Claude Enterprise + Cowork). He emphasizes that powerful systems can still be built under real-world constraints like budgets, tokens, and IT policies.
- •Enterprise constraints shape what ‘average PMs’ can realistically do
- •Melio moved from Gemini Enterprise to Claude Enterprise + Cowork
- •Cowork changed the game, but the underlying principles matter more than the brand
- •A viable system must work inside bureaucracy and budget limits
- 5:20 – 6:43
The two rules of a genuinely powerful AI system: self-rewriting + deep integrations
Daniel lays out two design rules: the system must be able to rewrite/improve its own core files, and it must connect broadly into your ecosystem. Together, they enable both continuous improvement and day-to-day usefulness.
- •Rule 1: the system can rewrite its own ‘core files’ to improve over time
- •Rule 2: connect to as much of your work ecosystem as possible
- •Self-improvement + integration turns a chatbot into an operational system
- •Earlier tools (e.g., Gems) helped, but lacked these compounding properties
- 6:43 – 8:37
The Notion board Cowork built—and actively manages as the PM’s operating hub
Daniel shows a simple Notion board (Top of Mind / This Week / Inbox) that functions as his lightweight source of truth. The key twist is that Cowork built the board and maintains it on his behalf, making Notion mostly “read-only.”
- •Notion structure: Top of Mind, This Week, Inbox
- •Cowork created the board autonomously (replacing a messy Google Doc)
- •Cowork populates items from Slack/email/calendar rather than manual entry
- •Notion becomes a focus dashboard while Cowork does the upkeep
- 8:37 – 10:05
Context bootstrapping: topic files, voice memos, and ongoing knowledge refresh
Daniel explains how he “teaches” Claude his environment using structured context files per domain (goals, colleagues, projects) and rapid voice dictation. The system then refreshes those knowledge files periodically to stay current.
- •Context files organized by topic/area: goals, people, projects, etc.
- •Manual upfront seeding: links, decks, docs, and other references
- •Voice dictation (‘whispering’ long explanations) as a fast context input method
- •Recurring updates refresh knowledge files based on recent changes
- 10:05 – 11:54
Weekly Prep automation: turning the ecosystem into a prioritized plan + meeting prep
Daniel walks through his “Weekly Prep” recurring task composed of multiple skills. It pulls signals from Notion, calendar, Slack, and meeting transcripts to recommend weekly priorities and help him prepare for upcoming meetings.
- •Weekly Prep is the backbone recurring task that sets the week up
- •Pulls from calendar, Slack, Notion, and Granola meeting transcripts
- •Recommends adds/removals for Top of Mind and This Week priorities
- •Prompts for meeting prep depth (serious prep task vs. quick reminder)
- 11:54 – 16:30
Morning Brief live demo: meeting catch-up, action items, and staying aligned with reality
The daily Morning Brief bridges the gap between a tidy plan and the messy reality of constant inbound work. It summarizes meetings using transcripts, extracts action items, and keeps Daniel oriented despite back-to-back schedules.
- •Uses Granola transcripts to summarize each meeting in a one-liner
- •Extracts and surfaces concrete action items from yesterday’s meetings
- •Designed for PM reality: meetings generate work faster than you can track it
- •Keeps the ‘clean board’ and ‘messy day’ continuously reconciled
- 16:30 – 18:47
Context self-healing: Claude detects unknown internal terms and saves definitions
A standout feature: the Morning Brief scans recent Slack/email/notes for terms Claude doesn’t understand and asks Daniel to define them. Those definitions are then saved into context, solving the ‘internal jargon isn’t in training data’ problem.
- •Claude flags unknown org-specific terms (e.g., “settlement cap”)
- •Prompts Daniel: what is it, is it important, should it be saved?
- •Definitions get written back into long-term context files
- •Prevents slow context drift and repeated misunderstandings over time
- 18:47 – 19:57
Operating 70–80% of the workday through Cowork: centralization as a force multiplier
Daniel explains that the system works best when most work flows through Cowork, because that’s how context and task state stay accurate. Notion becomes mostly a viewing layer while Cowork becomes the interface for executing work.
- •Daniel routes ~70–80% of computer time through Cowork
- •Centralizing work improves context capture and reduces blind spots
- •Notion is primarily read-only; Cowork updates and manages tasks
- •The system learns Daniel’s personal workflows (e.g., inbox zero signals)
- 19:57 – 20:35
Tools and execution: Chrome connector vs MCPs/connectors for missing integrations
Daniel compares approaches for tool access: native connectors/MCPs when available, and Chrome/browser automation when they aren’t. The Chrome connector remains critical to cover gaps in the integration landscape.
- •Early on, Daniel relied more on Chrome-based interaction
- •As MCPs/connectors expand, they become the preferred pathway
- •Chrome connector is essential for tools without direct integrations
- •Practical point: execution coverage matters as much as model quality
- 20:35 – 27:31
The real ROI curve: why early weeks feel slow and why the value compounds
Both Daniel and Claire stress that switching systems initially feels inefficient because the old workflow is muscle memory. The payoff comes from relentless contextualization and centralizing work, after which speed and depth compound dramatically.
- •Early setup is frictionful: low trust, lots of double-checking
- •Two behaviors to push through: share context ruthlessly + centralize work
- •Compounding benefit: faster throughput and higher-quality, data-backed work
- •Claim: can complete a typical week’s priorities in a single strong day
- 27:31 – 30:03
Self-improvement loop: learning from edits, suggesting new skills, and fixing friction automatically
Daniel showcases a weekly “self-improvement loop” that helps Cowork get better with minimal manual effort. It learns from how Daniel rewrites drafts, detects repeatable workflows worth turning into skills, and reviews logged friction to propose fixes.
- •Draft delta learning: compare Claude drafts vs what Daniel actually sent
- •Skill mining: detect recurring tasks and propose automating them as skills
- •Telemetry-driven fixes: skills collect feedback/friction during normal use
- •Weekly review surfaces top improvements without constant tinkering
- 30:03 – 33:33
“Improve” skill: filtering AI hype into actionable upgrades for your specific system
To avoid drowning in AI tips and trends, Daniel built an “Improve” skill that audits advice from X/LinkedIn/newsletters. Claude critiques relevance, maps ideas to the current setup, and helps decide what to implement versus ignore.
- •Addresses overload: endless AI ‘must-do’ tips and setups online
- •Claude acts as a critical auditor: what’s real, what’s noise?
- •Adapts ideas to Daniel’s existing architecture rather than chasing novelty
- •Turns content (including other episodes) into a structured upgrade pipeline
- 33:33 – 39:12
Scaling to the org with the Workstation plugin: onboarding UX and the ‘Spectacular’ lesson
Daniel explains how Melio scaled AI workflows with a shared “Workstation” plugin that onboards users via an in-chat guided flow. A key lesson from distributing a spec-writing gem (“Spectacular”) is that great internal tools still need great UX to be adopted.
- •Workstation plugin: an ‘operating system’ that started with PMs, then expanded
- •Guided onboarding connects tools, maps role/org, captures voice and goals
- •UX focus beat more technical options (e.g., terminal-heavy setups) for adoption
- •Lesson from ‘Spectacular’: tools tailored to one person need hand-holding UX for others
- 39:12 – 46:07
What’s still missing (the last 20%) and what Daniel does with reclaimed time—plus Claude rage
Daniel identifies the remaining gap as true cloud/away-from-computer operation and more autonomous execution of small tasks. With reclaimed time, he doubles down on deep PM work (customers, research) and continues improving the system—while admitting to occasional ‘Claude rage’ when tools misbehave.
- •Remaining 20%: agent autonomy when the laptop is closed + more direct action-taking
- •Workarounds today: Slack channel dropbox for later processing
- •Reclaimed time goes to deep work: user/customer conversations and research
- •Human reality: frustration still happens (‘Claude rage’), especially via voice dictation
