The non-technical PM’s guide to building with Cursor | Zevi Arnovitz (Meta)

Episode Summary

Executive Summary: Zevi Arnowitz, a non-technical PM at Meta, explains how AI tools let him build and ship real products without coding experience. He shares a step-by-step workflow using projects, slash commands, Cursor, Claude Code, Linear, and multiple model reviews to plan, build, review, and document features. The conversation argues that AI makes builders out of everyone, especially juniors willing to learn.

Main Topics: Non-technical product building with AI (Priority: 5/5): Zevi describes how someone with zero technical background can now build serious products by combining conversational AI, coding agents, and disciplined workflows. A structured AI workflow for product development (Priority: 5/5): He walks through his custom process: capture issues in Linear, explore the idea, create a plan, execute with Cursor/Claude, review with multiple models, and update docs. Why Cursor/Claude Code beat more opinionated tools for serious work (Priority: 4/5): Zevi explains that tools like Lovable, Bolt, Replit, Base44, and v0 are great for fast starts, but Cursor and Claude Code give him more control for complex production work. Multi-model code review and peer review (Priority: 5/5): Because the hardest problem is now reviewing AI-written code, he uses Claude, Codex, Cursor, and peer-review prompts to catch mistakes from different angles. AI as a learning accelerator for PMs and juniors (Priority: 4/5): He frames AI as a way to learn faster, get more reps, and operate at a higher level rather than outsource thinking or let skills atrophy. AI in hiring and interview prep (Priority: 3/5): Zevi used AI projects, mock interviews, and search/analysis workflows to prepare for and land his Meta PM role, while still emphasizing human mocks as essential. The future of roles collapsing into builders (Priority: 4/5): He predicts titles and responsibilities will blur as more people become capable of building products directly with AI.

Key Arguments: AI has made it possible for non-technical PMs to build real products end-to-end, even without writing code. The main challenge is no longer code generation but code review and understanding failures in AI output. A custom workflow of issue capture, exploration, planning, execution, review, and documentation makes AI-assisted building reliable. Opinionated low-code tools are excellent for getting started, but more advanced work requires tools that give users control over architecture and implementation choices. Using multiple models for code review is valuable because each model has different strengths and catches different issues. AI should be used as a learning partner; asking it to explain mistakes and updating prompts/docs improves future output. Junior professionals now have unprecedented leverage because they can build startups, ship products, and learn faster than before. At larger companies, AI-native codebases and contained UI work can enable PMs to contribute more directly, though heavy database changes should still be handled carefully. AI is not replacing strong performers; people who use AI better will outperform those who don't.

Data Points: Companies mentioned as using 10Web: Over 2 million websites - Sponsor mention: 10Web says its Vibe Coding Platform has generated more than 2 million websites in the last three years. StudyMate quiz generation: 30% - Zevi wants 30% of tests generated as fill-in-the-blank questions. Fill-in-the-blank format: 6 potential answers for 2 blank spots - Feature spec: two blanks, each with one correct answer and two incorrect answers as distractors. Model review process: Multiple reviews - He runs code review across Claude, Codex, and Cursor/Composer to catch different errors. Locali­zation project speed: 2 days - He said he localized StudyMate from Hebrew to English in two days. Personal site launch speed: 1.5 hours - He built a personal site from no domain to live on a domain within an hour and a half. Code review issue severity: Critical / high / medium bugs - Claude’s review surfaced a critical prompt bug plus high and medium issues during the feature review. Product pass count: 19 premium products - Sponsor offer from Lenny’s newsletter annual subscription. Starting salary/margin context: $4 per sale - In high school thermal clothing business, he made about $4 profit per item when selling at retail. Negotiated source price: $12.50 per piece - He negotiated directly with the importer to improve margins in his thermal clothing business.

Pivotal Quotes: "It's not that you will be replaced by AI. You'll be replaced by someone who's better at using AI than you." — Zevi Arnowitz / host framing the episode: Central thesis repeated during the conversation about AI adoption and career survival. "The expectation of me was being a 10X learner." — Zevi Arnowitz: He reflects on failing his first Wix product review and realizing junior PMs are valued for learning speed, not having all the answers. "If people walk away thinking how amazing you are, you failed. And if people walk away and open their computer and start building, you've succeeded." — Claude (as relayed by Zevi): Used to frame the purpose of the episode and the value of sharing actionable workflows.

Implications: AI is turning product management into a more hands-on, builder-oriented discipline. For listeners, the key is to start small, learn the workflow, use multiple models, and treat AI as a tutor and collaborator—not a shortcut to slop.