How a Meta PM Ships Products Without Writing Code - The Vibe Coding Workflow

The PM Who Builds Production Apps Without Knowing How to Code

Zevi Arnovitz has zero technical background. He studied music in high school. He wasn't in a tech unit in the Israeli army. Yet he's a PM at Meta who builds and ships production apps as weekend projects - using AI as his entire engineering team.

The moment everything changed: "When Sonnet 3.5 came out, I was watching a YouTube video of people building apps with Bolt or Lovable. It felt like someone came up to me and said, 'Hey, you have superpowers now.'"

On why this matters: "If you're nontechnical like me, code is terrifying. It's the scariest thing in the world to look at. I look at it as exposure therapy."

His engineers at Meta ask him to teach them what he's figured out. This conversation reveals the complete workflow.

The Slash Command System: A Complete PM-to-Shipped-Product Workflow

Zevi has built a series of reusable prompts (slash commands) in Cursor that form a complete development workflow:

1. /create-issue - Quickly capture a bug or feature idea mid-development, creates a Linear ticket
2. /exploration-phase - Claude analyzes the codebase and asks clarifying questions before any coding
3. /create-plan - Generates a markdown file with clear steps, status tracking, and technical decisions
4. /execute-plan - Actually builds the thing
5. /review - Claude reviews its own code
6. /peer-review - Multiple AI models review each other's code
7. /update-docs - Updates documentation for future agents

The key insight: "The big difference between just vibe coding and building serious apps is I spend a lot of time going back and forth and understanding. The exploration phase is critical."

How to Make AI Models Review Each Other's Work

The most innovative part of Zevi's workflow is the peer review system. Since he can't review code himself, he has different AI models review each other:

How Zevi describes each model (his words):

Claude: "She would be the perfect CTO. She's very communicative. She's very smart. She doesn't just go with the flow and do whatever you tell her. She's very opinionated, but also super collaborative."

Codex (GPT): "I always imagine it as the best coder within the company who comes to the office with a hoodie and sandals and sits in a dark room. You basically only bother him when you have the worst bugs. He'll close the door for two hours and come out and say 'I fixed it.' Really not communicative but it solves all the worst problems."

Gemini: "Gemini is like a crazy scientist who's super artsy, super talented at designing, but if you sit next to it and watch it work, it's terrifying. You would fire that person instantly... But at the end of the day, Gemini is very good at design."

"I have each model review the code, then use /peer-review which tells Claude: 'You're the dev lead on this project. Other team leads have found these issues. Don't take what they said at face value - you have more context. Either explain why they're wrong or fix the issues.'"

The models argue. Claude gets "sassy" sometimes: "This has been raised for the third time and for the third time I'm telling you this is not an issue. This is by design."

The CTO in Your Pocket: Why Starting with a ChatGPT Project Matters

Before Cursor, Zevi created a "CTO" project in ChatGPT - a custom prompt that acts as the complete technical owner:

"I told it: I own the problem. I own how we want users to feel. You're the complete owner of how this is built. I want you to challenge me. I don't want you to be a people pleaser."

Why this matters: "Regular ChatGPT would be the worst CTO because it's such a people pleaser. I asked it if Bun JavaScript is similar to Zustand and it said 'oh yeah, exactly the same.' Then it said, 'I thought you were making this up and I was riffing with you.' That's terrifying."

The Learning Opportunity Hack

One of Zevi's most powerful slash commands is /learning-opportunity:

"Every time something is difficult for me to understand, I do /learning-opportunity. It primes Claude: 'I am a technical PM in the making. I have mid-level engineering knowledge. Explain what we're working on using the 80/20 rule.'"

On the atrophy fear: "I have a very strong disagree to the idea that AI makes your skills atrophy. The misconception is that PMs should always have the right answers. My job is harnessing anything that gets us as quickly as possible to delivering the right solution. If you're using AI to just create outputs, that's AI slop. But if you use it intentionally, it's a game changer."

The Post-Mortem Habit That Compounds

"When Claude fails to do something or creates a really bad bug, I ask it: 'What in your system prompt or tooling made you make this mistake?' Claude goes introspective, then I say, 'Let's update your tooling and documentation so this never happens again.'"

This is how the prompts keep getting better. The whole system improves every time something goes wrong.

7 Takeaways for Non-Technical Builders Using AI

• Start with a ChatGPT project - Create your "CTO" before touching code. Learn in a safe environment.
• Graduate tools slowly - GPT → Bolt/Lovable → Cursor. Code is exposure therapy.
• The exploration phase is critical - Don't let AI start coding immediately. Planning matters.
• Have models review each other - Claude, Codex, and Gemini catch different things.
• Treat failures as system upgrades - Every bug is a chance to improve your prompts.
• Use /learning-opportunity - Turn every confusing moment into education.
• Own your outputs - "If you say 'sorry, that was built by AI' - that's your mistake."

What This Means for the Future of Knowledge Work

Zevi's prediction: "Titles are going to collapse and responsibilities are going to collapse. Everyone's going to become a builder. It's the best time to be a junior. When else in history could you get out of school and build a startup on your own?"

This isn't theoretical. He built a complete quiz app (Studymate) as a weekend project. He fully localized it from Hebrew to English in two days. He went from zero domain to live website in 90 minutes.

The skill that matters now isn't coding - it's being "a 10x learner, not a 10x PM." The prompts, slash commands, and workflow he's sharing are downloadable. The only barrier left is starting.

Try This Workflow in TeamDay

Zevi's entire workflow runs on Claude Code - which means it works directly in TeamDay with zero local setup.

What TeamDay adds:

• No installation - Claude Code runs in the cloud, ready to use
• Skills - Save Zevi's slash commands as reusable TeamDay skills
• Scheduling - Run automated tasks daily/weekly (like code reviews)
• Always-on - Unlike a laptop, TeamDay runs 24/7

Quick start:

1. Create a skill for /exploration-phase with Zevi's prompt template
2. Create a skill for /create-plan with the markdown output format
3. Use the peer-review pattern with Claude reviewing its own work

The slash command system Zevi built is exactly how TeamDay skills work. His workflow is TeamDay-native.