AI Product Manager Workflows: Strategy Documents, Prototyping, and Personal Operating Systems

Research Date: 2026-01-20 Publication Date: 2026-01-11 Source URL: https://x.com/petergyang/status/2010368153671647266

Reference URLs

• YouTube Video: Full Course - The AI Stack We Actually Use
• Peter Yang YouTube Channel
• X Post Announcement

Summary

This research note documents a 51-minute video discussion between three AI product managers—Peter Yang, Tal Raviv, and Aman Khan—demonstrating five practical AI workflows they use in daily product work. The central thesis, articulated by Peter Yang, is that effective AI workflows reduce to “text files all the way down”—meaning well-structured markdown context, templates, and instructions form the foundation of useful AI assistance rather than complex technical implementations.

The demonstrated workflows span strategy document creation using Claude Projects, rapid UI prototyping with Google AI Studio, initiative management through Obsidian and Cursor, personal operating systems built with Claude Code, and task coordination via Linear with AI agents. A meta-observation emerges: each participant has effectively built their own personal AI product that they continuously iterate on, functioning simultaneously as user and product manager for their system.

Participants

Workflow 1: Strategy Documents with Claude Projects

Overview

Peter Yang demonstrates using Claude’s Projects feature to write strategy memos and PRDs. The workflow combines custom templates, deep research capabilities, and meeting transcripts to iteratively develop documentation.

Components

1. Strategy Document Template: A markdown file containing structured sections:

   • Problem statement
   • Vision
   • Core product principles
   • Goals
   • Solution
   • Non-goals

2. One-Page Constraint: Documents are intentionally limited to one page maximum, based on the rationale that longer documents will simply be summarized by AI anyway.

3. Deep Research Integration: Claude’s research capabilities gather competitive intelligence and market context as reference material.

4. Meeting Transcript Integration: Tools like Granola capture meeting discussions that feed into document iterations.

Process Flow

Key Insights

• The initial AI output is typically “vague” and “generic”—the value comes from iterative refinement
• Context quality determines output quality: more meetings, more research, and clearer human opinions yield better results
• The workflow replaces isolated document writing with continuous AI collaboration

Workflow 2: Prototyping with Google AI Studio

Overview

Peter Yang demonstrates using Google AI Studio for rapid UI prototyping, arguing that prototypes should precede formal documentation and design.

Approach

1. Base Template Creation: Replicate an existing product UI in AI Studio to establish a foundation
2. Feature Prototyping: Add new features by prompting with reference images and descriptions
3. User Validation: Test prototypes with users before investing in formal design or documentation

Advantages Cited

• Speed: “By the time the Figma mocks were ready, I already tested the prototype with real users and got feedback”
• Gemini Model Quality: Noted improvement in UI generation quality with Gemini 3, moving beyond “purple AI slop”
• Cost Position: Google can afford to be the low-cost provider, subsidizing usage
• Enterprise Integration: Bundling potential with Google Workspace

Workflow Position

The discussion frames Google AI Studio as occupying the prototyping space rather than production code space, with competitors including Lovable, Replit, Figma Make, and Magic Patterns.

Strategic Observation

Multiple variations and mix-and-match capabilities between prototypes represent an underexplored feature space—designers typically create five variations, not one.

Workflow 3: Initiative Management with Obsidian and Cursor

Overview

Tal Raviv demonstrates using Obsidian as a markdown visualization layer for initiative tracking, with Cursor providing AI analysis capabilities over the same file structure.

Architecture

Implementation Details

1. Kanban Plugin: A free Obsidian plugin renders markdown as a Trello-style board
2. File Structure: Each column is a markdown section; each card is a link to another file
3. Cursor Ask Mode: Rather than coding, Cursor is used for strategic consultation

Example Interaction

Tal feeds an entire initiatives folder to Cursor and prompts:

“I need your help taking a look at this. I need a second pair of eyes. What do you think of this prioritization? What do you think the order of operations and how I’m approaching this?”

Linear Integration

Linear is mentioned as a complementary platform that:

• Provides free seats for AI agents
• Enables agent collaboration on development tasks
• Maintains structured rollout plans and milestone tracking

Key Quote

“Cursor is an IDE, but I use it mostly actually not to code.”

Workflow 4: Personal Operating System with Claude Code

Overview

Aman Khan demonstrates an open-source personal operating system built with Claude Code, featuring skills, MCPs, goal tracking, and automated daily planning.

System Architecture

Core Components

Skills

Skills are markdown files containing instructions Claude Code follows based on decision criteria. Example: morning-planning.md skill that:

1. Checks for new meetings in Granola
2. Lists tasks from custom MCP
3. Reads goals file
4. Presents morning overview with focus recommendations

Granola Integration

A Python script wraps Granola’s local cache (meeting notes and transcripts) in an MCP-compatible interface, enabling:

• Automatic meeting transcript ingestion
• Sync to knowledge base
• Context for daily planning

Goals File

A markdown file containing:

• Job description and context
• 12-month goals
• 5-year north star
• Quarterly objectives
• Key metrics

Accountability Feature

When used consistently, the system provides accountability:

“When I use my personal OS, it sometimes says ‘Hey, it looks like you’re drifting. You’re working on projects that you’re finding interesting, but are they really aligned with the goals that you care about?’”

Mobile Integration

Obsidian’s mobile app enables phone-based backlog capture that syncs to the desktop system, allowing workflow inputs from any context.

Open Source

Aman mentions the personal OS repository is public on GitHub with tutorials for setup.

Workflow 5: Linear with AI Agents

Overview

Brief discussion of using Linear for task management with AI agent integration.

Key Points

• Linear provides free seats for AI agents
• Agents can be assigned tasks like human teammates
• MCP integration (Linear MCP) enables Claude/Cursor to read and write Linear data
• Rollout stages and milestone tracking visible to both humans and agents

Meta-Observations

The Personal AI Product Paradigm

All three participants have built personalized AI systems they continuously iterate on:

“What all three of us are doing is we basically created our own personal AI product that we’re curating and iterating on. That in itself is like an incredible source of learning and building intuition.”

User and PM Combined

“We’re all building our own personal AI where we’re both the user and the PM. That in itself is an incredible source of learning.”

Minimal Vibe Coding Required

The workflows demonstrated require minimal traditional coding:

“Almost no vibe coding. Very little. When Aman was vibe coding, he was vibe coding a markdown file that was expressing how he wanted it to work.”

Tool Convergence

Despite using different tools (Claude Projects, Cursor, Claude Code), the underlying pattern is identical: carefully designed context in text files.

“It’s Claude Projects, it’s Cursor, it’s Claude Code—really impressive AI products that mean a lot and have at least one highly retained user for each of us, without vibe coding tons of HTML CSS.”

Core Thesis: Text Files All the Way Down

The video’s central insight, quoted directly:

“If I could summarize what I’ve learned about AI, it’s all text files all the way down.”

This implies:

• Context management is the primary skill
• Markdown files are the universal interface
• Plumbing (MCPs, skills, templates) is largely standardized
• Differentiation comes from personalized context, not technical implementation

Use Cases by Workflow

Practical Recommendations

Based on the discussion, entry points for adoption:

1. Start with context: Before adopting new tools, structure existing knowledge in markdown
2. Template your documents: Create reusable templates for common artifacts (PRDs, strategy memos)
3. Capture meetings: Use transcription tools (Granola, Zoom AI) to create analyzable records
4. Define goals explicitly: Write down objectives in a format AI can reference
5. Iterate daily: Use the system as both user and product manager, noting friction points

Limitations Noted

• AI output quality depends heavily on input context quality
• Initial outputs are often generic; value comes from iteration
• Non-deterministic behavior requires flexibility in expectations
• Systems require ongoing curation and maintenance

Video Metrics

• Duration: 51:33
• Views: 11,000+ (as of 2026-01-20)
• Channel Subscribers: 49,700
• X Post Engagement: 35,200 views, 240 likes, 495 bookmarks

References

1. Peter Yang X Post - 2026-01-11
2. YouTube Video - 2026-01-11
3. Peter Yang YouTube Channel
4. Google AI Studio
5. Obsidian
6. Linear
7. Granola