Kathmandu, August 21
When Grade 9 student Pranav Paudel decided to give his sister a birthday present, he chose something she did not expect: a website.
The static website featured different birthday-themed designs. What surprised his sister was not just the gift itself, but the fact that her younger brother had built it without having taken any IT-related course.
Pranav, a student at Manakamana School in Narayantar, Kathmandu, had taught himself what is now popularly known as ‘vibe coding’.
He built the website in his spare time after developing an interest in technology. It was a simple project, but for his family, it was evidence of something more important—his curiosity and willingness to teach himself.
“We were happy to see that he is curious. Even afterwards, whenever he is interested in something, he tries to accomplish it through vibe coding and enjoys doing it himself,” Pranav’s sister said. “He has started dreaming of becoming a chess player as well as pursuing a career in IT. He is learning accordingly.”
Pranav is just 14, and he does not have deep technical knowledge. Yet, using AI, he is able to turn an idea into something tangible.
So, what exactly is ‘vibe coding’?
Vibe coding refers to a way of developing software in which a person describes in natural language what they want a computer programme or website to do, and an artificial intelligence tool generates the necessary code.
Instead of manually writing every line of code, users can give instructions such as, “I want a website with this design” or “I want this button to perform this action.” The AI tool then generates the code.
The process is iterative. Users run the generated code, identify errors or shortcomings, tell the AI what needs to be fixed, and repeat the process.
In other words, the person does not necessarily need to understand or manually write every line of code. They work with AI to turn an idea into a functioning product.
The term ‘vibe coding’ gained wider attention after AI researcher and OpenAI founding member Andrej Karpathy used it in a post on X on February 2, 2025.
Karpathy described it as a new kind of coding in which a person essentially gives themselves over to the process and focuses less on the code itself.
The phrase quickly entered the technology mainstream. Collins English Dictionary subsequently named ‘vibe coding’ its Word of the Year in 2025.
From students to government offices
What began gaining popularity among technology enthusiasts and developers around the world is now finding its way into Nepal’s software development ecosystem.
The practice is being used not only by students and individual developers, but increasingly in the development of internal software and government-related systems.
Nepal Telecom, for instance, has started developing several internal software systems using AI-assisted coding practices.
The company is developing a Procurement Management System to digitally manage its purchasing process, a Duty Chart Management System to manage employee shifts and responsibilities, and a Meeting Management System to record meeting agendas, participants and decisions.
It is also developing a Workflow Management System to make office files and decision-making processes paperless, as well as a Unified Ticketing System to manage network equipment and customer complaints in an integrated manner.
An engineer at Nepal Telecom told Onlinekhabar that the organisation is following the practice of vibe coding, using AI to create an initial version of the code based on requirements before technical teams test and modify it.
“We are following vibe coding and then modifying it in our own way. Security and other aspects are taken into consideration. It has made the process much easier,” the engineer said. “This is also the way software development is increasingly being done.”
The practice is also being adopted within the government.
An official at the Office of the Prime Minister and Council of Ministers said the government has been giving priority to AI-assisted coding in the development of internal software, while its use is increasing in software being developed in other areas as well.
A member of the Nepal Association for Software and IT Services Companies (NAS-IT) said the use of vibe coding could also be observed in some newer government tools and websites.
According to the member, AI-assisted coding is being used to add new features, fix common user problems, modify forms and dashboards, and generate reports in systems related to driving licences and the Inland Revenue Department’s electronic lottery system.
“These things can be understood clearly by looking at the websites and systems. Anyone with software knowledge can identify the use of this process,” the member said.
But the growing use of AI does not mean the government is simply taking AI-generated code and putting it into operation.
The Prime Minister’s Office official said the government was not blindly deploying AI-generated software without understanding its implications.
Systems involving driving licences, tax information, payments and personal data are sensitive, the official said. Code generated through AI is therefore not deployed directly without testing.
According to the official, security testing, code reviews and access controls are carried out before AI-generated code is incorporated into critical systems.
“Previously, even for a minor change, a large portion of the code had to be typed manually. Now, we can give clear instructions about the requirement and use AI to create an initial structure of the code,” the official said. “The technical team and engineers then modify and test it according to the requirements of the government system. We do not build systems blindly.”
Faster development, bigger security concerns
The appeal of vibe coding is straightforward: it can make software development faster and allow people with limited programming knowledge to build things that would previously have required considerably more technical expertise.
But that speed comes with concerns.
As the popularity of vibe coding has grown worldwide, cybersecurity experts have raised questions about the risks associated with AI-generated code.
Reports published by technology publications have warned that vibe coding can create significant security vulnerabilities, including the risk of data leaks. AI-generated applications can potentially contain problems such as command injection vulnerabilities and exposed API keys.
An analysis published by Code by Melendez has suggested that evidence from 2026 indicates security vulnerabilities can be found in a significant proportion of AI-generated code when it is produced without adequate human supervision.
Other analyses have also pointed to problems such as insecure databases, weak authentication, exposed API keys and a lack of proper code auditing in applications created through vibe coding.
That does not necessarily mean the technology should be rejected.
Instead, experts recommend a hybrid approach: let AI generate the code, but have human developers review it, conduct security tests and make the necessary changes before deployment.
A growing global market
The rise of vibe coding is not limited to individual developers experimenting with AI. It is also becoming a growing global market.
According to a report by Future Market Insights, the global vibe coding market was valued at around $7.1 billion in 2026. The market is projected to grow at an annual rate of 17.1 percent over the next decade and reach $34.4 billion by 2036.
Technology companies including Microsoft, Google, OpenAI, Anthropic, Replit, Lovable and Vercel are developing tools and platforms designed to make AI-assisted software development easier.
Tools such as Cursor, Claude Code, Replit, Bolt and Lovable are allowing people with limited programming knowledge to create websites and applications.
The trend is becoming increasingly visible in Nepal as well.
But its adoption among Nepal’s established technology companies remains limited.
Gauravraj Pandey, president of NAS-IT, said vibe coding has yet to become widespread among large and established technology companies in Nepal.
According to Pandey, concerns about quality and long-term maintainability have prevented professional software companies from adopting it extensively.
“It is used less in professional work because of concerns about quality and long-term maintainability. It is more common in startups and freelancing, where quick results are needed,” he said.
Pandey identified three major challenges associated with vibe coding: security, system architecture and long-term maintainability.
Developers, he said, need to ensure that systems are built on a strong architecture capable of being expanded in the future, while also ensuring that the code can be maintained over the long term.
That does not make vibe coding inherently problematic, he said. If the process is properly managed and the code is reviewed and tested by experienced developers, it can still be useful.
‘Vibe coding cannot be rejected’
IT expert Vedvyas Pokharel believes vibe coding should not be viewed negatively as software development evolves.
The technology, he said, can significantly speed up developers’ work. He also uses it himself to develop software and provide services to clients.
“I use it myself and provide services to clients through vibe coding,” Pokharel said. “It has become very convenient recently. If you know how to use it carefully while paying attention to security, it makes the work easier and faster.”
Cybersecurity expert Naresh Lamgade also sees value in vibe coding, particularly in the speed it brings to software development.
He said it can make it easier for organisations to develop software according to their own requirements rather than relying entirely on external developers.
“It is very good because it increases the speed of work. But the main goal of AI is to complete the task. So when working at speed, less attention can be paid to security,” Lamgade said.
That trade-off is where much of the concern lies.
According to Lamgade, AI-generated applications can sometimes expose sensitive information such as API keys and secrets or leave connection mechanisms in plain text.
He said some platforms, including Lovable, have introduced security assessment as an additional paid feature, reflecting the growing concern around security in AI-assisted development.
The desire to build software quickly, he warned, can increase risks if security is treated as an afterthought.
“Secret keys should be stored securely before hosting and launching. At least one round of security testing should be conducted,” Lamgade said. “Depending on the commercial risk, manual testing of the code is also necessary. If these things are done, vibe coding cannot simply be rejected.”
For a student like Pranav, however, the technology represents something much simpler: an opportunity.
He did not need years of formal programming training to turn an idea into a birthday present for his sister. Curiosity was enough to get him started, and AI helped him build.
That same technology is now moving beyond personal projects and into Nepal’s businesses and government offices.
