AI as On-Call CTO

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

Zevi Arnovitz, a non-technical PM at Meta, has developed a systematic workflow using AI tools to build products without coding knowledge. His approach transforms AI from a simple coding assistant into a comprehensive development partner that handles everything from ideation to implementation.

The Core Workflow: From Idea to Implementation

1. Issue Creation Phase

   • Use a slash command to quickly capture feature ideas or bugs
   • AI formats these as structured Linear tickets with proper context
   • Allows you to continue working without losing momentum on ideas

2. Exploration Phase

   • AI analyzes the issue and scans the codebase to understand context
   • Acts as a technical partner asking clarifying questions about requirements
   • Helps identify technical constraints and implementation options
   • Focuses on understanding the problem before jumping to solutions

3. Planning Phase

   • AI creates a structured markdown plan with clear, trackable steps
   • Includes TLDR, critical decisions, and detailed implementation tasks
   • Serves as documentation for future reference and collaboration

4. Execution Phase

   • AI implements the plan by writing actual code
   • Different models can be assigned to different parts of implementation
   • Allows leveraging specialized capabilities (e.g., Gemini for UI, Claude for backend)

5. Review Phase

   • AI reviews its own code to identify bugs and issues
   • Multiple models review the same code to catch different types of problems
   • "Peer review" approach pits different AI models against each other

6. Documentation Update Phase

   • AI updates documentation based on what was built
   • Creates reference materials for future development
   • Ensures knowledge is captured for future iterations

Key Principles for Non-Technical Builders

Start Simple and Graduate Upward

• Begin with user-friendly tools like ChatGPT projects or Lovable/Bolt
• Progress to more powerful but complex tools like Cursor as you gain confidence
• Use "exposure therapy" to gradually become comfortable with code

Leverage Model Specialization

• Different AI models have distinct strengths and personalities
• Claude: Communicative, thoughtful, good for planning and collaboration
• GPT/Codex: Less communicative but excellent at solving complex technical problems
• Gemini: Creative and design-focused but sometimes unpredictable

Continuous Learning and Improvement

• Use the "learning opportunity" slash command to understand technical concepts
• After each project, analyze what went wrong and update prompts/documentation
• Ask AI "what in your system prompt made you make this mistake?" to improve workflows

Maintain Quality Through Multiple Reviews

• Have AI models review their own work
• Use different AI models to review each other's work
• Manually QA the final product before shipping

Broader Implications for Product Development

• Role Evolution: "Titles are gonna collapse and responsibilities are gonna collapse and everyone's just gonna be building"
• Junior Advantage: "It's the best time to be a junior contrary to what a lot of people are saying"
• Learning Focus: Be a "10x learner" rather than trying to be a "10x PM" from day one
• AI Partnership: Use AI as a thought partner rather than just a code generator

The workflow demonstrates how non-technical PMs can now build significant products independently, turning AI from a simple assistant into a comprehensive development partner that handles everything from ideation to implementation.