How a Meta PM ships products without ever writing code | Zevi Arnovitz

Key Concepts

• AI-Powered Product Development: AI tools like Cursor, Claude, Codex, and Gemini are democratizing product development, enabling individuals without traditional coding skills to build and ship products.
• Iterative Improvement & “Postmortems”: A core workflow involves analyzing AI failures, understanding the root cause, and updating prompts/tooling for continuous improvement.
• AI as a Learning Partner: AI is not just a tool for output, but a valuable resource for learning and understanding technical concepts.
• The Evolving Role of the Product Manager: PMs can increasingly contribute directly to code development, blurring traditional role boundaries and becoming “builders.”
• AI-Native Codebases: Designing codebases with clear documentation and structure optimized for AI understanding enhances collaboration and maintainability.

AI-Driven Product Development & the Democratization of Building

Zevi Arnovitz demonstrates how non-technical product managers can leverage AI – specifically Cursor utilizing Claude Code as its primary engine, alongside Codex and Gemini – to independently build and ship products. This represents a shift towards a future where traditional job titles blur and AI proficiency becomes a crucial skill. He previously experimented with Bolt and Lovable but found them too restrictive. The current moment is seen as an opportune time for junior individuals to enter the field, as AI levels the playing field. A key component of this workflow is careful planning, exemplified by creating a “CTO” persona within ChatGPT to challenge product thinking and ensure sound technical decisions.

Workflow & Tooling

Zevi’s workflow is structured around a series of commands within Cursor, including /create issue to generate Linear issues, /exploration phase for problem analysis, /create plan for detailed implementation plans, and /review for code review utilizing multiple AI models (Claude, Codex, Composer). The /peer review command then prompts Claude to justify or address feedback from other AI models, simulating a collaborative review process. Continuous learning is fostered through a /learning opportunity command to deepen understanding of technical concepts. He emphasizes the importance of updating documentation and tooling after each AI mistake, creating a feedback loop for improvement.

Examples of products built using this workflow include StudyMate, a platform for creating interactive quizzes from uploaded materials (with a fill-in-the-blank feature being added live during the podcast), and a personal website built in just 1.5 hours. Zevi characterizes each AI model with a distinct personality: Claude as a collaborative CTO, Codex as a highly skilled but uncommunicative coder, and Gemini as a creative but unpredictable designer.

The Iterative AI Workflow & Learning from Failure

A central tenet of Zevi’s approach is an iterative workflow focused on identifying AI failures, analyzing the root cause, and updating the system (prompts, documentation, tooling) to prevent recurrence. This is presented as a more productive approach than simply attempting to force a solution. AI is framed not as a replacement for skill, but as a powerful learning tool; analyzing why an AI made a mistake is as important as obtaining the desired output. This process of prompt iteration is a key differentiator between casual and effective AI users.

Building for AI & the Future of Roles

Zevi advocates for building “AI-native” codebases designed to be easily understood and navigated by AI agents. This involves incorporating plain text documentation (Markdown files) explaining code structure and functionality. The conversation explores the potential for Product Managers to contribute directly to code development, even without extensive coding experience, starting with UI projects before tackling more complex tasks like database migrations. The long-term prediction is a blurring of traditional role boundaries, with everyone becoming a “builder.”

Addressing AI “Sloppiness” & Ownership

The segment acknowledges that AI-generated outputs can be flawed (“slop”) and stresses the importance of ownership and quality control. Tools like Cursor’s “deslop” command are mentioned for post-generation refinement. Guidance and context are crucial for AI to produce valuable results. Zevi emphasizes that users must take responsibility for the quality of AI-generated content, avoiding the excuse of “the AI did it.”

Real-World Applications & Interview Preparation

Zevi details his successful use of AI (Claude, Comet, and a question bank by Louis Lynn) to prepare for his Meta PM interview, creating a personalized “coach” AI and a quiz game for segmentation practice, supplemented by human mock interviews. He also recounts a high school entrepreneurial venture selling thermal clothing, illustrating his proactive approach to business. His brother’s replacement of Zapier and Airtable with a custom-built CRM and automation system using AI demonstrates the potential for individuals to build sophisticated tools without traditional development expertise. Zevi’s early experience at Wix highlights the importance of seeking mentorship and framing oneself as a “learner.”

Technical Foundations

Key technical terms include: Cursor (AI-powered code editor), Claude Code (Anthropic’s coding model), Codex (OpenAI’s coding model), Composer (fast coding model within Cursor), Gemini (Google’s multimodal AI model), Linear (project management tool), MCP (Anthropic’s Model Capabilities Platform), System Prompt, /commands, Vibe Coding, and API. LLMs (Large Language Models) are central to the discussion, as are concepts like Prompt Engineering and Markdown.

Conclusion

The conversation highlights a significant shift in product development, driven by the increasing accessibility and power of AI tools. The key takeaway is that AI is not a replacement for human skill, but a force multiplier that empowers individuals to build and learn more effectively. Success hinges on embracing an iterative workflow, prioritizing continuous learning from AI failures, and designing systems – both codebases and processes – that are optimized for AI collaboration. The future of work appears to be collaborative, with blurred role boundaries and a growing emphasis on becoming a “builder” proficient in leveraging AI.