AI Development Practices: Vibe Coding and Agentic Engineering

A new interview format called “vibe coding” is being used for AI product‑manager roles, where candidates prototype a feature—often on existing platforms such as LinkedIn—while demonstrating how they can leverage AI to streamline product design. The format blends product sense with rapid AI‑assisted coding, aiming to lower the barrier to prototyping but offering limited quality assurance.

At a Sequoia AI Ascent 2026 fireside chat, Andrej Karpathy distinguished this approach from “agentic engineering,” a discipline in which developers apply rigorous specification, diff review, testing and security oversight to AI‑generated code. Karpathy argued that agentic engineering can deliver performance gains far beyond 10× for teams that master its practices, and noted that the AI coding‑assistant market now includes dozens of companies with mature, series‑D‑plus products. The discussion highlighted four core skills—spec design, diff review, testing, and safety validation—that separate professional, high‑quality AI‑driven development from the more exploratory vibe‑coding style.