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

Product managers who can't code are about to become extinct. That's the stark reality Zevi Arnovitz delivers from his experience shipping AI-powered products at Meta. While most PMs debate whether to learn Python, Arnovitz has moved past coding entirely , he's orchestrating multiple AI models to build complex applications faster than traditional engineering teams.

Zevi Arnovitz has discovered something counterintuitive about AI coding tools: the eager ones that immediately start writing code are actually the worst for serious product development. "These products were built in a way where they were super eager to write code. So their system prompt was your coding agent. So when you write something, they'd straight away start coding. So at the beginning of a project, this was super fun and exciting because they just go and start building your app. But later on, when things got more complex, this created much more problems because planning is really important when you're implementing something technical."

This insight flips the conventional wisdom about AI development tools. Tools like Bolt and Lovable that promise instant gratification by jumping straight into code creation actually create technical debt and architectural problems. The secret is forcing AI to plan first, then execute. Arnovitz has developed a system that treats AI models like specialized team members, each with distinct strengths and weaknesses.

His approach centers on comprehensive planning before any code gets written. Using a template he discovered on Twitter, Arnovitz forces Claude to create detailed markdown plans with status tracking for every task. "It has the critical decisions we've made and the tasks broken down. And this is a perfect plan, and it's also a really good way to write this because a lot of times, I'll use different models to execute certain stuff."

Once the plan exists, Arnovitz deploys different AI models like a technical manager assigning specialists. Cursor's Composer handles straightforward implementation tasks with speed. The new Gemini model takes ownership of UI work because of its superior design capabilities. Claude manages complex backend logic. This isn't random assignment , it's strategic deployment based on each model's proven strengths.

The most sophisticated element of Arnovitz's system is AI peer review. He has different models critique each other's code, creating a quality control process that mirrors human engineering teams. "I'll do peer review a bunch of times, and I'll have other models review other models code and kind of have them fight it out. Basically, like sometimes, Claude Code will get really sassy and be like, 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."

This inter-model tension generates better code than any single AI working alone. When models disagree, they're forced to justify their approaches, leading to more robust solutions. Arnovitz describes it as having them "fight it out" , a deliberate friction that prevents groupthink and catches errors that would slip past a single reviewer.

The implications extend far beyond individual productivity. Traditional product development assumes a clear division between strategic thinking (PM work) and technical implementation (engineering work). Arnovitz's approach collapses this distinction. When a PM can orchestrate AI models to handle implementation while maintaining strategic oversight, the role fundamentally changes.

This shift demands new skills from product managers. Instead of learning to code, they need to learn to manage AI teams. That means understanding each model's capabilities, designing effective prompts, and creating systems that prevent AI from making expensive mistakes early in development. The PMs who adapt will gain unprecedented speed and autonomy. Those who don't will find themselves competing with colleagues who can ship products without waiting for engineering resources.

The technical landscape is moving toward AI-augmented development whether traditional teams embrace it or not. Product managers should start experimenting with multi-model orchestration now, beginning with simple projects where mistakes don't matter. Focus on building planning templates, understanding model strengths, and creating review processes. The goal isn't to replace engineers , it's to move fast enough that speed itself becomes a competitive advantage.