Claude Cowork for PMs: My Self-Improving Productivity System

In this episode, I was so excited to sit down with Daniel Blum, a senior product manager at Melio. As a fellow product person, I'm constantly wrestling with the chaos of endless tasks, meetings, and Slack messages. Daniel came on the show to reveal a system so comprehensive and powerful that it's completely changed how he works. He’s managed to wrangle the complexity of product management into a cohesive, automated, and self-improving workflow powered by Claude Cowork.

Daniel's approach is a testament to how we can move beyond simple AI prompts and build a true digital partner. He was feeling the classic PM pain: the constant overhead of coordinating, remembering endless action items, and trying to find time for deep, focused work. His solution wasn't just to use AI for one-off tasks, but to build an interconnected system that learns, adapts, and manages his work on his behalf. Today, he estimates that 70-80% of his time at the computer is spent working directly within Cowork.

We’re going to walk through the key components of his setup. It starts with a simple Notion board that acts as a read-only dashboard, but the real magic happens in Cowork. We’ll explore his "Weekly Prep" and "Morning Brief" workflows, dive into a brilliant "self-healing" context loop that keeps his AI up-to-date with internal knowledge, and see how he created a "Workstation" skill to help his entire team get onboarded in minutes. This is one of those episodes that shows what's possible when you truly commit to rethinking your process.

Workflow 1: The Automated PM Dashboard in Notion

The first thing Daniel showed me was his central dashboard in Notion. It’s deceptively simple, organized into three key sections:

• Top of mind: The big, strategic initiatives he's thinking about.
• This week: His tactical priorities for the current week.
• Inbox: A catch-all for new tasks and action items that pop up from Slack, email, and meetings.

But here's the key: Daniel almost never touches this board himself. He described it as a "read-only" view for him. His Cowork system built this board for him—migrating him from a messy Google Doc—and now it manages the board entirely on his behalf. It’s the front-end UI for the complex AI engine running in the background. All the workflows we're about to discuss feed into and update this central view, giving him a constant, clear picture of his priorities without the manual upkeep.

Workflow 2: Kicking Off the Week with 'Weekly Prep'

Every productive week starts with a plan. Daniel automated this with a recurring Cowork task called "Weekly Prep" that runs on Sunday mornings. This isn't just a simple reminder; it's a comprehensive process that sets his focus for the entire week.

How the Weekly Prep Works

1. Information Gathering: The skill pulls in information from his entire digital ecosystem. This includes his calendar, Slack messages, Notion pages, and even meeting summaries from the transcription service Granola.
2. Priority Suggestions: Based on all that data, the system analyzes what's new and what's outstanding. It then suggests what should be in his "Top of mind" and "This week" columns in Notion. It will propose adding new items, removing completed ones, or moving things around.
3. Meeting Preparation: The workflow scans his calendar for the upcoming week. For each meeting, it asks him how he wants to prepare. He can choose to create a dedicated task for deep prep, set a quick reminder, or mark it as needing no preparation at all.

After he confirms the suggestions, the system automatically updates his Notion board. This single workflow ensures that he starts every week with a clear, realistic plan that's already aligned with all the inputs from the previous week.

Workflow 3: The 'Morning Brief' and its Self-Healing Context Loop

While the weekly prep sets the strategy, the "Morning Brief" handles the daily reality. This is where I think the system's brilliance really shines, especially with its ability to learn and adapt.

Part A: The Daily Catch-Up

Every morning, this task runs to bridge the gap between yesterday's chaos and today's plan.

• It reviews transcripts from yesterday's meetings via Granola, pulling out one-line summaries and, most importantly, any action items that were decided.
• It identifies urgent priorities or new tasks that have come in overnight from Slack and email.
• In the demo, it surfaced a key action item from a meeting: to share the walkthrough and recording of... a write-up in the feature channel. Without this, that task might have been forgotten in the rush to the next meeting.

Part B: The Self-Healing Context Loop

This is the part that made me say, "This is really genius." A huge challenge with using AI at work is that it doesn't understand your company's internal jargon, project codenames, or specific goals. Daniel solved this by teaching Claude to ask for help.

Here’s the process:

1. As part of the morning brief, Cowork scans Daniel's recent communications.
2. It actively looks for terms or concepts it doesn’t recognize from its knowledge files.
3. When it finds one, it prompts Daniel directly. In the episode, it asked:

"A term I didn't know, the settlement cap... I read the thread and understand it's a limit working through. Want me to save?"

1. Daniel simply replies "Yes," and Cowork saves this new definition to its context files, making it smarter for the next time.

This proactive learning loop ensures the AI's knowledge doesn't become stale. It's constantly enriching its understanding of Daniel's specific work environment, making its assistance more and more relevant over time.

Workflow 4: The Self-Improvement Engine

A static system is a dead system. Daniel has built a scheduled weekly task specifically designed to make his entire AI setup better over time, with minimal effort from him. It’s a meta-workflow that improves all the other workflows.

This "Self-Improvement Loop" has a few fascinating components:

Learning from Edits

When Cowork drafts a message for Daniel, and he edits it before sending, the system takes note. The loop looks for these instances, comparing the initial draft with the final version. It then analyzes the differences to learn his voice, tone, and common phrasing patterns, so its future drafts are better from the start.

Suggesting New Skills

The system monitors Daniel's activity for repetitive tasks. If it notices he's doing the same sequence of actions over and over, it suggests turning that sequence into a new, reusable Cowork skill. For example, it noticed he was doing a lot of prototyping and suggested creating a "Claude design handoff" skill to automate the process.

Fixing Friction

Daniel built feedback collection directly into his skills. If a workflow is clunky or he has to correct the AI, that "friction" is logged. This weekly task surfaces the top friction points and suggests concrete fixes to the underlying skills, smoothing out the rough edges of the system.

Critiquing New AI Hype

This might be my favorite part. Daniel has a skill called "Improve" where he forwards articles, X posts, and tips about AI (including How I AI episodes!). The skill acts as a critical auditor. It analyzes the new idea, compares it to his existing system, and gives a recommendation on whether it's worth implementing or if it's just "LinkedIn hype." It's a filter for the firehose of AI content we all face.

Workflow 5: Scaling Success with the 'Workstation' Onboarding Skill

A 10x system is great for one person, but the real impact comes from scaling it to a team. Daniel and his colleagues at Melio faced the challenge of making this complex system accessible to everyone, not just AI power users. Their solution is a shared Cowork skill called "Workstation."

Workstation is a guided, chat-based onboarding experience that gets a new user up and running in about 15 minutes.

Here's what it does:

1. Connects Tools: It walks the user through connecting their accounts like Slack, calendar, and Notion.
2. Maps the Org: It asks questions to understand their role, their manager, and their key colleagues.
3. Finds Their Voice: It helps them generate a personal voice profile so that AI-generated text sounds authentic to them from day one.
4. Sets Goals: It captures their primary objectives and priorities.

By turning the setup process into a simple, conversational workflow, they've made it incredibly easy for anyone at the company to adopt this powerful way of working. It removes the technical barriers and high learning curve, ensuring the whole team can benefit.

Conclusion: The Compounding Value of an AI Operating System

What Daniel has built is more than just a collection of scripts; it's a true AI operating system for his work life. He said something that really stuck with me: the value "tremendously compounds." The initial weeks of setting things up, contextualizing the AI, and centralizing your work can feel like a chore. There's friction, and the ROI isn't immediately obvious.

But as the system learns your "personal business logic," it becomes an incredibly powerful partner. Daniel says he can now accomplish in a day what used to take him a week. More importantly, it frees him up for deeper work—more customer conversations, more in-depth research, and more data-backed decisions instead of just operating on hunches.

This episode is a blueprint for what it looks like to go all-in on building an AI-powered workflow. It requires a willingness to push through the initial setup pains, but the payoff is a system that not only makes you more productive but also grows and improves right alongside you. If you're wondering how to take your own AI usage to the next level, start by building your own small, self-improving loop.

A word from our sponsors

Thank you to our sponsors for helping us make the show!

The Optimizely Agent Platform for Marketers is built on a bigger idea, a whole directory of agents now joined by a newly revealed squad of virtual teammates, each with a defined role and a personality of their own. Meet the squad and understand what virtual teammates can do for you at optimizely.com/howiAI.

The teamwork graph in Jira by Atlassian delivers forty-four percent more accurate agent results with forty-eight percent less token usage. The hardest part of shipping with AI isn't the code, it's the context. The teamwork graph pulls all of that from across your entire stack. Same team, smarter agents. Try them free at jira.dev.

Episode Links

• Listen to the full episode on your favorite podcast platform.
• Find Daniel Blum on his website.