The rapid introduction of AI Agents such as Claude 3.7 and AI-powered development tools like Cursor AI signals a sweeping disruption in how code is produced. Traditional programming—where individual programmers write every line themselves—is becoming less and less viable. Instead, programmers must harness AI's capabilities or risk falling behind.
Early definitions of Vibe Coding did not account for the transformative power of AI Agents. Yet these agents perfectly embody Vibe Coding's core principles, bringing them to life in ways previously unimagined. This manifesto declares that the future of coding is AI-driven collaboration, wherein humans curate knowledge and orchestrate testing, but rarely write code manually.
A brief note: Although this manifesto might naturally belong on my personal blog, JustDo has been my life's work—my passion project. It felt only right to share it here, where I've poured so much of my dedication, in the hope of inspiring others who share this journey with me. Many of the lessons in this manifesto were shaped by my experiences working on JustDo's source code, and I believe sharing them will help others push the boundaries of what's possible in coding.
1. A Self-Building Knowledge Base
A centralized, living repository of knowledge about the code and the product. Critically, the AI agent itself updates and maintains this knowledge base, with oversight from the human programmer. (In Cursor AI, for instance, this knowledge base is implemented through .mdc files.)
Why it matters: When the AI can capture, reference, and refine all relevant information, it can work more autonomously—while staying true to project goals, code style requirements, and product constraints.
2. Ensuring Testability
Every piece of code produced by the AI must be covered by meaningful tests. The AI itself should generate and run these tests to create a positive feedback loop, continuously ensuring that the code meets quality and functional requirements.
Why it matters: Reliable code depends on robust testing. By making test generation and execution a core AI responsibility, humans can confidently trust (and verify) the AI's output.
1. Programmers Should Never "Hand-Code."
If you find yourself manually writing or fixing code, it signals a gap in your knowledge base or in the instructions you provided to the AI. Rather than getting "hands-on" in code, a Vibe Programmer updates the knowledge base to teach the AI what went wrong, then directs the AI to fix its own mistakes.
2. All Programmers Become Code Reviewers.
Humans remain the ultimate auditors of AI-generated code. Every commit must be scrutinized for product alignment, code style, DRYness, reusability, and other best practices. The agent is not replacing developers—it is transforming them into stewards of quality, ready to give strategic feedback and direct improvements to the knowledge base.
3. Programmers Must Ensure Frameworks for Testability.
It is the human's responsibility to design and maintain testing frameworks. The AI must be able to run tests on anything it writes, ensuring functionality and reliability. If the system lacks the necessary tools or frameworks, a Vibe Programmer's top priority is to design and integrate them before letting the AI proceed.
Vibe Coding represents not just a new methodology, but a fundamental shift in mindset: programmers become teachers, auditors, and knowledge curators for AI Agents that handle the bulk of raw coding. By focusing on robust knowledge bases and seamless testing, teams can unlock the speed and adaptability that AI offers—without compromising clarity, maintainability, or product vision.
The future of coding is already here. Embrace Vibe Coding and harness the transformative power of AI Agents to evolve from solitary coders to empowered orchestrators of digital innovation.
Daniel Chcouri,
March 24, 2025