Code 2025
The Cure for the Vibe Coding Hangover — Corey J. Gallon, Rexmore
Overview
This talk introduces a framework designed to combat "vibe coding," an approach to AI-accelerated development that prioritizes speed and intuition over planning and maintainability, leading to brittle, unmanageable software. The framework emphasizes treating AI engineering as a continuous learning experience, with the engineer acting as the architect and the AI as the implementer. It advocates for a deliberate, iterative process that compounds understanding and productivity, ultimately enabling the creation of robust, production-ready applications.
Who should watch
- AI engineers, PMs, and builders who want to create maintainable and understandable software using AI coding agents.
- Developers who feel like prompt jockeys rather than engineers when working with AI.
- Those frustrated with throwing away code, wasting time and tokens, or building applications that don't perform real work.
- Individuals who value programming as a craft and a daily learning experience.
Key takeaways
- Vibe coding is a low-planning, AI-accelerated development style that feels productive but results in unmaintainable code. The "hangover" is the subsequent despair when trying to build robust software this way.
- The framework consists of three pillars: Principles (philosophy), Process (workflow), and Tools (accelerators).
- Key principles include: AI engineering is accelerated learning, you are the architect and the agent implements, and slow down and iterate to go fast.
- Specification is paramount, far exceeding prompt engineering. Clearly defined requirements and acceptance criteria are crucial for agent implementation.
- The process involves a distinct planning phase (vision, features, specification, dependencies, plan) and an implementation phase using a multi-sensory feedback loop.
- The implementation loop involves writing code, executing it while gathering visual, auditory, and tactile feedback, and testing against acceptance criteria until the feature is complete.
- Essential tools include a coding environment with an AI agent and sandbox, a multi-sensory feedback system, context engineering tools, and version control with progress tracking.
- The goal is to build working software that can be shipped and used, with further investment in refinement and optimization driven by real-world usage.
Notable quotes
*The framework is not just about building faster. It's about learning as we go.*
*Delegate the doing and not the thinking.*
*Write the blueprint, not the prompt.*
Unofficial community note. Prefer the recording for nuance.