Skip to content
Lenny's 🏦 Vault

AI Models Review Each Other

by Zevi Arnovitz on January 18, 2026

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 using AI tools to build products without coding knowledge. His approach transforms AI from a simple assistant into a collaborative technical partner that enables non-technical people to build significant products.

The Core Workflow

  1. Issue Creation and Exploration

    • Start by capturing ideas with /create issue to quickly document bugs or features in Linear
    • Use /exploration phase to deeply understand the problem and have AI ask clarifying questions
    • This mimics the initial PM-Engineer conversation about requirements and constraints

  2. Planning and Execution

    • Use /create plan to generate a detailed markdown implementation plan
    • The plan becomes a reference document for all models to follow
    • Execute the plan with /execute plan to generate the actual code
    • Different models excel at different tasks (e.g., Gemini for UI, Composer for speed)

  3. Multi-Model Review Process

    • Run /review with Claude to identify bugs and issues in the code
    • Run parallel reviews with other models (Codex, Composer) to catch different types of issues
    • Use /peer review to have models critique each other's findings
    • Models will defend their decisions or acknowledge and fix legitimate issues

  4. Documentation and Learning

    • Update documentation with /update docs to improve future code generation
    • Use /learning opportunity to understand technical concepts you're unfamiliar with
    • Conduct "postmortems" on AI mistakes to improve prompts and documentation

Key Principles for Non-Technical Builders

  • Start simple and gradually increase complexity: Begin with ChatGPT projects, then graduate to Bolt/Lovable, and finally to Cursor/Claude Code

  • Use "exposure therapy" for code: Gradually expose yourself to more code to overcome the initial fear

  • Leverage model specialization: Different AI models have different strengths - use them accordingly

    • Claude: Best for communication, planning, and collaboration
    • Codex (GPT): Excellent at solving complex coding problems
    • Gemini: Superior for UI/UX design work

  • Continuous improvement through reflection: When AI makes mistakes, ask "what in your system prompt made you make this mistake?" and update accordingly

  • Make AI-native codebases: Add markdown files and documentation that help AI understand the codebase structure

  • View AI as amplifying learning, not replacing it: "The expectation isn't being a 10x PM, but being a 10x learner"

  • Treat AI as a thought partner: Use it to explore ideas and get feedback, not just generate outputs

The Future of Product Management

  • "If people walk away thinking how amazing you are, you failed. If people walk away and open their computer and start building, you've succeeded."
  • Titles and responsibilities will collapse as everyone becomes a builder
  • The best time to be a junior contrary to popular belief - when else could you build a startup on your own right out of school?
  • "It's not that you will be replaced by AI, you'll be replaced by someone who's better at using AI than you"
View full episode