A non-technical PM ships real products by running Claude and Cursor through a disciplined slash-command workflow.
Plan before any code is writtenLetting an eager coding agent start writing immediately caused the gnarliest bugs, so exploration and a written plan come first.
Let rival models review each otherClaude, Codex and Composer each catch different issues, and the peer review command makes the lead agent defend or fix them.
Ease in gradually, then keep learningStart with a GPT project, graduate to Bolt or Lovable, then Cursor — and use a learning opportunity command whenever something is unclear.
Executive SummaryAI
Lenny introduces Zevi Arnowitz, a PM at Meta who was previously a PM at Wix, as a non-technical product person who taught himself to build real products with Cursor and Claude Code.
Zevi explains that a year ago, travelling in Asia when Sonnet 3.5 came out, he watched a YouTube video of people building apps with Bolt or Lovable and started building the day he got home from Japan.
Because Bolt and Lovable rushed into code, he built a 'CTO' as a ChatGPT project with a custom prompt that owns the technical side, challenges him and refuses to be a people pleaser.
His current workflow is a chain of Claude Code slash commands — create issue into Linear, exploration phase, create plan as a markdown file, execute, review, peer review and update the docs — demonstrated live by adding fill-in-the-blank questions to his side product StudyMate.
For review he runs Claude, Codex and Cursor's Composer over the same branch and then feeds their findings back through a peer review command that makes the lead agent defend or fix its code, plus constant postmortems that update documentation and tooling.
Key Quote
“I have zero technical background.”
— Zevi Arnowitz
Key Quote
“So let's say Claude, she would be the perfect CTO.”
— Zevi Arnowitz
Key Quote
“And I think the only way that AI makes you worse at your job is if you're using it wrong.”