How a product from Meta launches products without being able to program (AI column)

I looked back six months ago. release Lenny's Podcast with Zevi Arnovitz, a product from banned in Russia Meta and former PM in Wix, but forgot to write about it. Then it seemed to me that this is a new filling of the role of the product, but if it works in a conditional Meta, then it is not a fact that it is easily transferred to other companies:) Now I wrote a post about what awaits products and remembered this video and decided to share it with you.

Zevi himself has no technical background and admits that he can hardly read code. At the same time, according to the participants of the issue, during the year he independently collected and monetized StudyMate, a service that turns educational materials into interactive tests.

Its working process is little like the classic vibe coding:

The idea becomes Linear’s task through MCP. Claude studies the codebase, asks questions and prepares a plan. Gemini takes the interface, Composer - fast execution, Codex - search for complex errors; The result passes manual QA, several AI-review and user testing; After errors, the instructions and documentation of the agent are updated.

In fact, Zevi built a small engineering organization around the models. It owns the problem and what the user needs to feel, and delegates the implementation to the agents. A product can now not only write a PRD and wait for a command, but go from an idea to a working product and feedback from users.

Hence his forecast: job titles and boundaries of responsibility will begin to collapse, and everyone will become “builders”. For beginners, it is also a way to get practice immediately in strategy, UX, marketing and development.

But there's an important caveat. At a large company, Zevi does not advise products to roll out base migrations or other risky changes on their own. The AI-native codebase must first be prepared by engineers: add context, documentation, and constraints. A smart start for PM is local UI feature, PR and final developer validation.

In general, now a product does not need to become a programmer, but it is no longer enough to remain only a task manager. You need to be able to assemble the system around the agents, check the results and be responsible for them. The phrase “AI did it” does not remove responsibility.

#AI #ProductManagement #AI4SDLC #Engineering #Management #VibeCoding