A non-technical PM ships real products by running Claude and Cursor through a disciplined slash-command workflow.
He uses Claude Code running inside Cursor, having graduated from Bolt when connecting payments broke down. He also keeps Codex and Cursor's Composer open, and uses Gemini 3 for front-end work. Code is just files, he says, so a project carries from app to app.
Seven reusable prompts: create issue, which files a Linear ticket via MCP; exploration phase, where Claude reads the code base and asks clarifying questions; create plan, producing a markdown plan with status trackers; execute plan; review; peer review; and finally update the docs so agents write better code later.
After Claude reviews its own code, he has Codex and Cursor's Composer review the same branch, then runs slash peer review, pasting each as 'dev lead one' and 'dev lead two'. Claude is told not to take them at face value, but explain why they're wrong or fix them.
Bolt and Lovable were built to write code immediately, which caused gnarly bugs once things got complex. So he created a CTO project with a custom prompt: it owns how things get built, he owns the problem, and it must challenge him rather than please him.
Start slowly with a GPT project, beautiful UI and super simple, then graduate to Bolt or Lovable, then Cursor in light mode, and only gradually open a terminal and go full dark mode. He calls it exposure therapy, because code is terrifying to a non-technical person.
The models are the same; Claude underlies Bolt and Lovable too. The difference is the harness: Bolt, Lovable, Replit, Base44 and v0 add layers that take hard decisions out for you, while Claude Code hands you the raw model, full tools, and every decision.
Slash learning opportunity. When something is hard to grasp, he runs it and describes what he wants to learn. It primes Claude with 'I am a technical PM in the making, I have mid-level engineering knowledge', and asks it to explain the current work using the 80-20 rule.
He treats documentation and tooling as the biggest productivity hack. When Claude fails or ships a bad bug, he asks what in its system prompt or tooling caused the mistake, then has it update the tooling, docs or slash commands so that mistake never happens again.
First make the code base AI native, with markdown files that explain to agents how to work in each area, and technical people should do that. He would not have PMs shipping heavy database migrations, but contained UI projects, with a PR handed to a dev for finishing, are realistic.
When Meta reached out he opened a Claude project as his coach, filled with frameworks from Ben Erez, and mock interviewed against it. He also built a segmentation quiz game in Base44 to play on the bus. The biggest change, though, was human mocks from cold LinkedIn outreach.
“I have zero technical background.”
“So let's say Claude, she would be the perfect CTO.”
“And I think the only way that AI makes you worse at your job is if you're using it wrong.”
“And I think, contrary to what a lot of people think, it's the best time to be a junior.”
“How a Meta PM ships products without ever writing code | Zevi Arnovitz.” https://www.youtube.com/watch?v=1em64iUFt3U. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/how-a-meta-pm-ships-products-without-ever-writing-code-zevi-.html.