Projects Prevent Memory Cross-Talk

Zevi Arnovitz's AI-Powered Product Development Workflow for Non-Technical PMs

Zevi Arnovitz, a non-technical PM at Meta, has developed a sophisticated workflow that allows him to build functional products without coding knowledge. His approach combines AI tools with structured prompts to create a complete development pipeline.

The AI-Powered PM Workflow

Core Tools and Setup

• Use Cursor as the primary IDE with Claude Code as the AI assistant
• Create slash commands (reusable prompts) for each phase of development
• Maintain documentation to help AI understand your codebase
• Use multiple AI models for different strengths (Claude for planning, Gemini for UI, etc.)

The Development Process

1. Issue Creation

   • Use /create issue to quickly capture bugs or feature ideas
   • AI formats the issue for Linear with proper structure
   • This allows you to continue working without losing your train of thought

2. Exploration Phase

   • Use /exploration phase to analyze the issue in depth
   • AI examines the codebase to understand current implementation
   • AI asks clarifying questions about requirements and constraints
   • This phase is about understanding the problem, not solving it yet

3. Planning

   • Use /create plan to develop a structured implementation approach
   • AI creates a markdown file with TLDR, critical decisions, and tasks
   • Each task includes status tracking for progress monitoring
   • The plan becomes documentation for future reference

4. Execution

   • Use /execute plan to implement the planned changes
   • AI writes the actual code based on the plan
   • Different AI models can be used for different aspects (backend vs. frontend)

5. Review Process

   • Manual QA testing first to catch obvious issues
   • Use /review to have the AI review its own code
   • Use multiple AI models to review the same code (peer review)
   • Use /peer review to have models critique each other's feedback
   • This creates a "team" of AI reviewers with different perspectives

6. Documentation Update

   • Update documentation based on what was built
   • This helps future AI interactions understand the codebase better

Continuous Improvement

• After encountering errors, ask AI "what in your system prompt made you make this mistake?"
• Update prompts and documentation based on these insights
• This creates a virtuous cycle where your tools get smarter over time

Practical Applications and Advice

For Beginners

• Start with ChatGPT projects before moving to code editors
• Create a "CTO" project with a custom prompt to be your technical partner
• Gradually move to more advanced tools as you gain confidence
• Use the "learning opportunity" prompt to understand technical concepts

For Teams

• Make your codebase "AI-native" with clear documentation
• Start with contained UI projects rather than complex database changes
• Use AI as a collaborative learning tool with your development team
• Focus on areas where each AI model excels (Claude for planning, GPT for debugging, Gemini for UI)

Mindset Shifts

• Think of AI as a learning tool, not just a production tool
• Own your outputs - never blame AI for poor quality work
• Use AI to play at a higher level than your current experience
• Focus on being a "10x learner" rather than a "10x PM"
• Remember: "You can just do things" - the barrier to building is lower than ever

Addressing Common Concerns

On Skill Atrophy

• Using AI doesn't weaken skills if used intentionally
• AI allows you to get more "reps" at higher-level thinking
• The PM's job is to deliver solutions, not to always have the answers
• Set AI up for success with proper context and guidance

On Code Quality

• Use multiple models to review code from different perspectives
• Have models critique each other's work
• Use the /deslop command to clean up code
• Always manually test before shipping