Here’s Google’s Replacement for Vibe Coding

Quick Overview

The video details the creation and execution of a complex, context-driven development workflow using the Gemini CLI extension called Conductor, demonstrating how it scaffolds a new Next.js project named Sprint Cycle, defines product requirements, establishes a tech stack, and generates initial tracks for development, despite encountering and manually correcting errors related to database connectivity and missing configuration steps.

Key Points: The video showcases the Conductor extension for Gemini CLI, a structured workflow tool for agentic development, contrasting it with standard chat sessions by using Markdown files to persist and track project progress across sessions. The demonstration involves setting up a new project called 'Sprint Cycle,' a project tracker designed for small to medium-sized software development teams, emphasizing speed and simplicity. Conductor guides the user through four setup phases: Project Discovery, Product Definition (using Q&A to refine product.md and product-guidelines.md), Configuration (selecting a tech stack, confirming Next.js/TypeScript/Supabase/Drizzle stack), and Track Generation. Initial track generation for scaffolding failed because the AI incorrectly tried to use 'npm' instead of 'pnpm' (as specified later) and failed to find temporary backup files, requiring manual intervention to delete backups and reconstruct the directory structure. The AI initially proposed an overly broad single track for building the MVP, which the user corrected, guiding the AI to break it down into smaller, manageable tracks, starting with project scaffolding and core schema. Despite successful scaffolding and configuration of code style guides and workflow, a subsequent attempt to push the database schema failed due to a missing 'postgres' role, requiring the user to manually check and correct the environment configuration (.env.local file). The process highlights that while Conductor automates much of the setup, complex projects still require user guidance (like correcting package manager choice and fixing environment variables) to navigate agent errors and ensure correctness.