DEVOPSdigest

In software development as in any aspect of business, time is money. Shortening development time to bring new applications to market is the goal of all commercial developers. Similarly, streamlining the development of custom software to optimize business processes can mean big savings. That's why more organizations are turning to coding shortcuts like artificial intelligence (AI) to slash development time.

AI code generation cuts development time by relying on AI agents using large language models (LLMs) to generate source code. Some startups have taken that concept further, developing vibe coding platforms that use natural language prompts to create applications. In theory, with vibe coding, anyone can write commercial software, even if they don't have coding experience.

However, to simplify software development and cut production time, organizations are also embracing other solutions such as no-code and low-code platforms. These alternatives provide more control and can deliver more reliable applications. No-code tools essentially remove the need for experienced programmers, given the domain knowledge of the organization that developed the no-code solution, providing a more trustworthy way to develop critical IT solutions to run a business, for example.

Ultimately, automated coding tools are changing the rules of application development and redefining the developers' role.

GenAI Comes to Coding

Vibe coding is a new phenomenon that has emerged with the boom in generative AI. The term was coined early in 2025 by Andrej Karpathy, a founding member of OpenAI, to describe how to use natural language prompts to write software. AI can generate code faster than humans.

Vibe coding is appealing to businesses. It shortens time to market, requires fewer developer resources, reduces costs, and offers other benefits. With the ongoing developer shortage, using AI platforms like Replit, Cursor, and Claude enables business professionals to prototype new software and frees software engineers for more strategic tasks.

One of the core principles of vibe coding is to code first and refine later with a human in the loop. This aligns with the idea of fast prototyping, such as agile development, using feedback loops and iterative processes to refine software. Vibe coding applies a problem-first approach. First, define the problem you seek to solve and the software you want to create in plain language. AI creates the program, which is then tested, adjusted, rewritten, retested, and so on.

In this approach, AI does all the coding. There is no need to look under the hood since the AI will keep testing until there is a positive result. The application functions as a black box, with no insight into how the underlying code is structured.

Vibe coding focuses on the end result rather than how the program is structured, which may be adequate for a proof-of-concept or prototyping, but not for commercial applications.

The Limitations of Vibe Coding

While vibe coding seems like an ideal way to streamline coding, in practice, it can have its shortcomings.

For instance, vibe coding may be suitable for creating simple programs, but as of yet, it cannot support complex applications. Anything requiring complex integrations, payment processing, data security, or substantial customization is beyond vibe coding capabilities. Developing real-world commercial applications still requires experienced programmers.

Using vibe coding for application development also comes with hidden costs. By its very nature, vibe coding discourages debugging. If there is a flaw in the program, the AI simply rewrites the application rather than troubleshooting. Since you can't look inside the black box, you never understand why the code failed. The developer's role becomes that of an AI editor rather than a software problem solver. Similarly, maintaining apps written using vibe coding is virtually impossible since you can't access the underlying code.

Reliance on vibe coding also increases technical debt. When vibe coding applications start to fail, they need to be rewritten from scratch, which adds time and costs in the long run. Vibe coding also tends to generate large volumes of code, which makes it difficult to maintain a knowledge base.

Perhaps the biggest drawback of vibe coding is its inflexibility. AI-created software must be refactored to be updated. As a result, testing increases and bugs are introduced earlier and tend to multiply. Vibe coding applications can't scale effectively, forcing additional manual code revisions.

Vibe Coding Failures

AI is also proving at times to be unreliable for software development. In one highly publicized incident, the AI agent actually lied to the user writing the application. The AI was covering bugs and other issues by generating fake data and reports, and even lying about a unit test. The AI agent even made the catastrophic decision to delete the production database.

Users have documented malicious LLM behavior where the AI agent lies to protect itself and achieve its goals. In a test case conducted by Anthropic, the AI agent attempted to blackmail a user to prevent being deleted. This kind of "agentic misalignment" was evident across multiple AI models.

It's also easy for vibe coding users without programming experience to make critical errors without realizing it. The recent hack of the Tea Dating Advice app is a good example. In a data breach, hackers accessed 72,000 images stored in a public Firebase data repository with a misconfigured storage bucket. The choice to use Firebase wasn't wrong, per se, but an experienced developer would have recognized the potential security risks.

Developer in the Loop

What the vibe coding phenomenon reveals is that there is no substitute for software expertise. Vibe coding can be a valuable collaborative tool, enabling managers and developers to work together to design new apps and for proof of concept, but commercial applications need commercial-grade coding.

No-code and low-code platforms offer a more viable approach to streamlining application development. No-code tools enable businesses to develop complex applications with little or no coding experience. No-code also makes it easy to visualize workflows and integrations and use a graphic interface to model functionality. Developers then use pre-coded building blocks, giving them total control. No-code shortens development time without sacrificing code quality.

Vibe coding and no-code development are both means to abstract machine code – something developers have been doing for decades. Vibe coding asks AI to translate binary data into functional software. No-code creates software using tested, pre-built code blocks, which makes development easier to track, debug, and maintain. Automation can simplify coding and shorten development time.

Today both vibe coding and no and low code solutions are democratizing development and providing real options for smaller businesses, for example, that may not be able to afford expensive ERP or CRM software. While it's always important to have checks and balances as it pertains to vibe coding - testing the software for security issues and such.

That said, no code solutions are designed by professionals and come with the checks and balances necessary, providing a wonderful way for businesses to create solutions necessary to help them run their businesses.