We will go through the entire process, from a completely empty repository to creating a full-fledged responsive WordPress site, where most of the code will be written not by hand, but through text commands to OpenAI's Codex AI assistant. And in the end, we'll try to answer the main question: should engineers be worried about their future?
What is the Philosophy of Vibe Coding?
Vibe coding (or vibe programming) is a development approach where the programmer shifts from the role of an executor to that of an architect or creative director.
Instead of meticulously writing every line, handling every error, and remembering function syntax, the developer sets the general direction, the “vibe,” and the high-level functional requirements for the project. They describe the desired outcome in natural language, and the AI assistant transforms this vision into a concrete implementation.
If you imagine traditional development as building a wall where the programmer is a bricklayer carefully placing each brick (line of code), then vibe coding is when the architect tells a team of robots: “I need a sturdy wall in a Gothic style, with a pattern here and an arched passageway.”
This approach is a logical extension of the evolution from imperative programming (“do step A, then step B”) to declarative programming (“I want to achieve result X”). AI-assisted coding takes this idea to its absolute conclusion.
The term emerged organically within the developer community as they realized that interacting with powerful large language models (LLMs) is less about writing code and more about conveying the right atmosphere, context, and intent. It's true “code vibing,” where the key skill is not knowledge of syntax, but the ability to clearly and vividly communicate one's vision to the machine.
Vibe Coding: A Future Opportunity or an Existential Threat?
The emergence of AI-assisted coding will inevitably transform the role of an engineer and the demands on their skills. It's a two-edged sword: while it presents remarkable possibilities for productivity, it also poses significant risks that we will point out.
New Horizons and Opportunities:
- Faster and faster: Routine tasks that previously required hours (such as writing boilerplate code, setting up configuration files, and conducting basic tests) can now be finished in minutes with “vibe code”.
- Unlocking Creative Potential: Developers doesn't expend mental energy on syntax and routine – they instead can fully focus on business logic, system architecture and abstract high-level solutions.
- Democratization of Development: People from non-technical backgrounds who will be able to create simple applications and prototypes – simply by describing what the business needs.
- Improved Code Quality: AI assistants can identify potential vulnerabilities and ensure a consistent coding style. They are tireless code reviewers.
Risks and Challenges to Be Aware Of:
- The Illusion of Control: Abstract commands can lead to unexpected results. An AI can generate code that appears to work, but contains hidden issues: with performance, security, or scalability.
- De-skilling: Over-reliance on AI could lead to a new generation of developers who lack a deep understanding of the fundamental development process. They risk becoming “prompt operators,” incapable of solving a problem when the AI fails.
- The “Black Box”: Sometimes, the generated code can be so complex that even an experienced developer will find it difficult to understand its logic. It could complicate debugging and maintenance significantly.
- Transformation of the Job Market: Although the complete disappearance of the profession is unlikely, the developer's role is transforming. The demand for engineers who can only code will decrease, and, instead, the demand will grow for system architects.
Will AI vibe coding become a trend? Absolutely. But it won't replace developers – AI assisted coding rather become part of their toolkit, like an IDE or Git.
Real-World Example: Creating a WordPress Theme from Scratch with Codex
But enough with that, let's go to practice. The work began by setting up a local development environment (Local WP) and an empty GitHub repository, which was connected to Codex. And this is where the first indicative situation occurred.
When the AI suggested standard tasks like “explain the codebase” or “find and fix a bug,” it concluded on its own that there was nothing to analyze—the repository was empty. This is the first and most important lesson: artificial intelligence needs either context for analysis or a clear task to create something new.
Then the first specific request emerged: create a basic "skeleton" of a WordPress theme with all the necessary files (style.css, index.php, functions.php, etc.). The AI successfully handled this, but new problems arose here:
- An attempt to create a Pull Request (PR) failed because the system did not support binary files, which included a temporary screenshot.png image.
- When the author asked the AI to delete this file, it also deleted the critically important index.php, breaking the entire theme.
This demonstrated that the AI does not always understand file dependencies and importance, and therefore requires careful supervision. Moreover, the correction process was complicated by the fact that Codex created a new PR for each change, cluttering the commit history.
From Chaos to Style: Iterative Development of the Homepage
The creation of the homepage became the heart of the experiment and best illustrated the “human-AI” interaction model. The process went through several stages:
- Iteration 1: The First Draft. First, it was necessary to provide a detailed request to create a custom front-page.php template.
An important point was to specify all the sections: header, banner, service cards, gallery slider, contact form, and footer.
The result was structurally functional but visually "raw" and had errors: the navigation looked bad, and the slider controls were misaligned with the content.
- Iteration 2: Fixes and Styling. The next request was aimed at solving these problems:
- Implement a "burger menu" for mobile devices.
- Apply a specific color palette (#FFF9F4, #FB8500, #383838).
- Replace the static form with a shortcode from the Contact Form 7 plugin.
- Fix the positioning of the slider elements.
- Add rounded corners to soften the design.
The result was much better, but again not perfect: the footer was left without the new color, and the burger menu did not work correctly.
- Iteration 3 and 4: Final Polishing. The subsequent requests became even more specific. Step by step, each element was brought to order: the mobile menu logic was fixed, the image sizes in the slider were limited, and autoplay was enabled. The final request was aimed at giving the site a modern, "trendy" look and included:
- Changing the fonts to Poppins from Google Fonts.
- Adding soft gradients and patterns to the background.
- Redesigning the cards with shadows and hover effects.
- Implementing a "sticky" header.
- Animating the burger icon and enabling smooth scrolling to sections.
- Embedding a Google Maps map.
After these steps, the homepage finally acquired a finished look.
Expanding the Theme and Brilliant Automation
With the design ready, the AI was tasked with creating a blog catalog page and a single post page, adhering to the already established style. The AI successfully coped, demonstrating its ability to follow a given pattern.
The final chord, which showed the true power of the AI, was the automation of deployment.
Codex was asked to write a GitHub Actions script that would automatically upload the theme files to the server via SFTP every time the main branch was updated. The request was very precise: use repository secrets for access and exclude unnecessary files from the upload. The AI performed this task excellently, confirming its effectiveness in working with the backend and automation.
Conclusions: Will Codex Replace Engineers?
My experiment clearly shows that the answer is a categorical no. AI assisted programming is not a profession killer but it's most powerful evolutionary catalyst.
Strengths of Codex:
- Boilerplate Code Creation: Ideal for quickly generating the basic structure of a project, saving hours of work.
- Complex Logic and Automation: Excellently handles writing scripts, configuration files, and other non-visual logic (e.g., CI/CD scripts).
- Refactoring and Routine: Effective for performing bulk, repetitive changes in the codebase.
Weaknesses and Limitations:
- Fine-tuning UI/UX: The AI requires several iterations and very detailed instructions to achieve the desired visual result. It is often faster and more effective to make the final adjustments manually.
- Speed for Simple Tasks: For minor changes (changing a color, text), the interaction process with the AI (request, wait, check) is too slow.
- Working with Files: There are limitations (binary files) and the risk of unpredictable actions, such as deleting necessary files.
I don't think LLMs will replace engineers. This approach has clear advantages, but also disadvantages. It is a tool that accelerates, but does not replace thinking.
The developer's main task is becoming not writing code, but the ability to ask the right questions, decompose complex tasks, critically evaluate the result, and conduct an effective dialogue with the machine.
We are entering the era of the “centaur developer” – a hybrid of human and artificial intelligence, where the human acts as the strategist and the AI as the tactical executor. Coding rocks – tools like these make development less static and more dynamic.
The future of development is already here. If you want to prepare your team for this new reality, I, Alex Gurbуch, would be happy to provide a consultation or conduct a training session.
