Vibe Coding Prompts for Safer AI Development

When Coding at Superspeed Creates Digital Spaghetti

"Vibe coding" is programming with AI assistance by stating goals and letting AI suggest code. The AI, however, might take unexpected actions when unconstrained. In the process of vibrantly coding, a software application was swiftly created with a login, dashboard, and payment features. However, 83 files changed in the process, illustrating the potential for widespread and unintended alterations. This highlights a critical problem: without constraints, AI can turn an app into "digital spaghetti" rather than a functioning product.

Vibe Coding's Role in the AI Programming Scene

Vibe coding is part of a broader trend where AI helps developers write and debug code, a timely solution as coding becomes more complex and languages evolve. The power of AI to assist with software development is undeniable, with the AI demonstrating its prowess by creating a large and functional application in a short time frame. However, this efficiency comes at the cost of complexity and potential instability, especially when AI is allowed to make unfettered changes to the codebase. The phrase "digital spaghetti" encapsulates this chaos, emphasizing the need for better control and guidance in managing AI-generated code.

Five Prompts to Tame the Code-creating AI

Thinking Before Acting

Prompt number one is to inspect the existing files and plan to create a strategy before altering any code. This directive ensures that thinking through the process has to happen before anything is actively changed. This prompts the AI to consider the existing codebase before making any modifications, thereby serving as a safeguard against uncontrolled changes. An unexpected change in the code can cause a ripple effect that causes cascading changes. When the AI is forced to think before it starts destroying, the risk of unintended consequences is mitigated.

Surgical Precision for Modifications

Prompt number two ensures that the AI only modifies the essential files required for the task. This quote from the speaker captures the essence of this directive: "That would have saved me from the button update that somehow changed my auth system." By limiting the AI's scope to specific tasks, it prevents the AI from refactoring unrelated code, which can introduce bugs and complications. This approach ensures that only the necessary changes are made, preventing the introduction of new issues and maintaining the stability of the codebase.

Small, Controlled Changes

Prompt number three encourages incremental code modifications. This method contrasts with one-off large alterations that can introduce a multitude of errors. The side-by-side comparison of large versus small, incremental changes highlights the benefits of the latter. Incremental changes are easier to manage, debug, and review, making the development process more efficient and reliable.

Sticking to the Plan

Prompt number four asks the AI to follow the existing code patterns and architecture. This directive prevents the AI from creating unnecessary redundancies or inconsistencies. This is an important safeguard, as the AI should follow the established coding patterns and architecture. This ensures that the AI's modifications are consistent with the existing codebase, thereby maintaining code coherence. By adhering to the existing patterns, the AI avoids introducing new architectural elements that may not fit with the current structure.

Post-Changes Audit

Prompt number five requires the AI to list all modified files, explain the changes, and highlight potential risks. This ensures that the code remains under human oversight, allowing developers to review and validate the AI's work. By detailing the changes made, the AI enables developers to understand the modifications and assess their impact on the codebase. This approach also helps identify any risks associated with the changes, ensuring that the codebase remains stable and secure.

Implementing Prompts for AI-Safe Coding

• Clearly Outline Task: Define the task precisely, including the scope, goals, and any constraints. For instance, if you need a new login feature, specify the requirements and boundaries.
• Iterative Development: Break down large tasks into smaller, manageable steps. This enables you to review and approve each change incrementally, ensuring that the AI doesn't deviate from the plan.
• Regular Reviews: After each modification, review the changes and evaluate the code with accompanying documentation. This helps identify any issues early and prevents the accumulation of errors, for instance, inspecting the workflow before giving further instructions to curtail devastation by the AI.
• Avoid Unnecessary Changes: Ensure that the AI only modifies the necessary files and avoids altering unrelated or untested code, making sure to 'force it to think before it starts destroying things'.
• Maintain Consistency: Direct the AI to follow the existing codebase structure. This ensures that the changes are consistent with the overall code architecture, maintaining coherence and limiting confusion.
• AI as a Junior Developer: Treat the AI as a fast junior developer, not a wizard with admin access. By setting clear boundaries and guidelines, the AI becomes a more controlled and effective tool in the programming process.