Vibe coding

Vibe coding is an innovative approach to software development that leverages the power of generative artificial intelligence to create dynamic and responsive applications.

In this article, we will explore the concept of vibe coding, its benefits and drawbacks, how it is transforming the way we build software, and some useful tips and tricks.

The benefits of vibe coding

There are several benefits to using vibe coding in software development:

Vibe coding saves a lot of time

Vibe coding allows developers to focus on the creative aspects of coding while AI handles repetitive tasks, leading to faster development cycles and more efficient workflows.

For example, instead of writing boilerplate code for setting up a web server, a developer can simply prompt the AI to generate the necessary code snippets, saving hours of manual coding.

I used to spend hours manually coding and styling HTML tables. Then I discovered that I could generate them a little faster with the help of a table generator like HTML Table and the Tables Generator. Now I vibe code and style them in seconds.

Vibe coding improves code quality

By automating repetitive tasks and providing intelligent suggestions, vibe coding can help developers produce cleaner, more efficient code.

Vibe coding enhances creativity

By offloading mundane tasks to AI, developers can dedicate more time to brainstorming and implementing innovative features, ultimately leading to more creative and engaging applications.

For example, a developer working on a game might use vibe coding to quickly generate character models and animations, allowing them to focus on designing unique gameplay mechanics and storylines.

Vibe coding can also inspire new ideas and approaches by suggesting alternative solutions.

The drawbacks of vibe coding

While vibe coding offers numerous advantages, it is essential to be aware of its potential drawbacks:

AI makes mistakes

It is important to understand that while generative AI can be a powerful tool, it is not infallible.

Sometimes the AI will generate code that is syntactically correct but does not do what you want it to do. For example, it might generate a function that calculates the sum of two numbers, but it actually multiplies them instead.

Sometimes the AI will generate code that has security vulnerabilities or performance issues. For example, it might generate a SQL query that is vulnerable to SQL injection attacks.

Sometimes the AI will generate code that is not compatible with your existing codebase or framework. For example, it might generate code that uses a different version of a library than the one you are using.

Sometimes the AI will generate code that is simply incorrect or nonsensical. For example, it might generate a function that takes no parameters but tries to access a variable that does not exist.

Sometimes the AI will break one thing while fixing another. For example, it might fix a bug in one part of the code but introduce a new bug in another part.

Always review generated code before approving changes

We recommend that you always review and test the code generated by AI tools before applying it.

Make incremental backups

Incremental backups are essential when working with generative AI tools. This ensures that you can recover your work in case of unexpected errors or data loss.

We recommend working iteratively - saving your work after every significant change. For example, if you are developing a web application, save your work after completing each feature or module.

We recommend using version control systems like Git to manage your codebase effectively. Alternatively, you can use cloud storage solutions that offer versioning features. Or, finally you can manually create dated copies of your work at regular intervals.

AI "lies"

Generative AI models can sometimes produce incorrect or misleading information. It is crucial to verify the veracity of its reports.

False claims of total completion

GitHub Copilot and Qoder will often tell me that a task was completed, when in fact it was not.

For example, it will say "I have added the necessary imports" when in fact it has not.

Sometimes GitHub Copilot will tell me that it "updated all the web feeds," when in fact it only updated two of the three.

AI "forgets"

Generative AI models can sometimes lose context or forget important information during a conversation. It is essential to provide relevant context and reminders to help the AI stay on track.

For example, if you are working on a complex project, periodically remind the AI of the project's goals and requirements to ensure it continues to provide relevant suggestions.

Inconvenient downtime

Generative AI tools may experience downtime or performance issues, which can disrupt your workflow. It's essential to have contingency plans in place to minimize the impact of such disruptions.

For example, you might consider using alternative tools or maintaining local copies of your work to ensure you can continue development even if the AI tool is unavailable.

Tips and tricks

Here are some tips and tricks to help you get the most out of vibe coding:

Use the latest LLMs

Use the latest Large Language Models (LLMs) geared at coding, to benefit from improved performance, accuracy, and features.

To do so make sure to select the latest model in your AI tool of choice. For example, in GitHub Copilot you can select "GPT-5" or "GPT-4.1" instead of "GPT-3.5".

Combine multiple tools

Don't limit yourself to just one AI tool. Combine the strengths of different tools to create a more efficient workflow.

Sometimes another LLM will do something better than the one you are currently using. For example, I find that ChatGPT 5.0 is better at generating code, while Claude Sonnet 4 is better at explaining code. So I often use both tools in tandem to get the best results.

Ask the right questions

Asking the right questions is crucial when working with AI tools.

Before implementing a new feature or making significant changes, ask the AI whether this would be the best solution, or whether there might be a better solution.

Doing so will not only help you identify potential problems before they arise, but it will also improve your code, and save you a lot of time and effort

Be descriptive in your prompts

When using AI tools, providing clear and detailed prompts can significantly improve the quality of the generated code. Include relevant context, examples, and specific requirements to help the AI understand your needs better.

Utilize AI suggestions

Take advantage of AI-generated code suggestions to speed up your development process. Don't hesitate to experiment with different prompts to get the best results.

Lint your code

Consider using tools like code linters and formatters to maintain code quality and consistency. Doing so will prevent syntax errors, improve readability, and prevent code insertion errors.

Create and use routines

Establishing routines for common tasks can help streamline your workflow and improve efficiency.

Consider creating templates or scripts for repetitive tasks. Then add them to conversations as needed.

Rolling back changes

If you encounter issues with the code generated by AI tools, it's important to know that you can roll back changes by simply asking the AI to "undo" or "revert previous changes".

Stay updated on AI advancements

The field of AI is constantly evolving. Stay informed about the latest developments and best practices to make the most of vibe coding.

Understanding coding concepts

Contrary to popular belief, vibe coding is not a magic bullet to coding without understanding.

To effectively utilize vibe coding, it's essential to grasp fundamental coding concepts. This includes understanding variables, control structures, data types, and algorithms. Familiarity with these concepts will enable you to communicate your ideas more clearly and make better use of AI tools.

Popular vibe coding tools

There are several popular vibe coding tools available today, each with its own unique features and capabilities. Some of the most notable ones include Cursor™, GitHub Copilot™ (which is tightly integrated with Microsoft Visual Studio Code) and Windsurf Editor.

Footnotes

1. Context / working memory values are approximate token window (or functional equivalent) capacities for common model variants. Code Assist relies on repository & multi-file semantic indexing rather than a single flat token window.

2. Hallucination / factual error rate ranges consolidated from GPT-4 technical report, Anthropic / Gemini / Code Assist published benchmark discussions, and independent survey synthesis (TruthfulQA-style, coding, and mixed factual tasks). Figures vary by prompt style, dataset, temperature, and evaluation method.

Please Review Us

If you appreciate our content and tutorials, please review SOHO Systems on Google Search and Facebook.

References

Anthropic (2024) Claude 3 Model Card & Benchmarks [Anthropic] https://www.anthropic.com

Google (2024) Gemini / Code Assist Factuality & Code Quality Improvements [Google] https://developers.google.com

Ji, Z., Lee, N., Frieske, R., et al. (2023) Survey of Hallucination in Natural Language Generation [ACM Computing Surveys] https://dl.acm.org/doi/10.1145/3527309

Kumar, M. (2025 03 30) A comprehensive guide to vibe coding tools [Medium] https://medium.com/madhukarkumar/a-comprehensive-guide-to-vibe-coding-tools-2bd35e2d7b4f

OpenAI (2023) GPT-4 Technical Report [OpenAI] https://openai.com/research/gpt-4

Wikipedia contributors (2025 09 08) Vibe coding [Wikipedia] https://en.wikipedia.org/wiki/Vibe_coding

For more information, please visit the archives at BleepingComputer, or run a Bing, DuckDuckGo, Google or YouTube search for "vibe coding".

🏷️ #ai #coding 📅 September 19, 2025 👨‍🦲 EJ Hill 🔃 🖨️