Nepal Telecom, the state-owned telecom operator, has built roughly half a dozen internal software systems using AI tools that generate applications from simple instructions. The approach, called vibe coding, is spreading through government offices and private companies across Nepal, moving AI-assisted development beyond startups and into public infrastructure.
The company's new software wing developed a demand management portal that centralises requests from local offices up to ministries, and a meeting management portal designed to ease pressure on limited office space. Engineers also automated the process of assigning IP addresses - unique numerical identifiers for network devices - which previously required manual processing. A unified ticketing system and a customer care portal are under development, according to senior business officer Prakash Chandra Sigdel.
"In some areas, we also take support from external vendors, as we cannot compromise on quality and customer experience," Sigdel said at a recent event in Kathmandu. "But we are moving towards building small and specific software in-house." Nepal Telecom is also studying a new billing system using similar technology; its existing billing system has been linked to recurring problems, including incorrect balance deductions.
How vibe coding works
Vibe coding lets users generate and refine software through written instructions, or prompts, rather than relying solely on manual programming. People with limited programming knowledge can direct AI tools to write, test, and refine code. This falls under the broader category of Generative Code automation, where AI models produce functional code from natural language descriptions. "In-house coding has proven effective," Sigdel said, adding that companies understand their own requirements better than outside vendors.
Private sector pushes into agentic engineering
Private technology firms are adopting similar methods at speed. Nischal Shrestha, chief technology officer at Amnil Technologies, said the field is shifting toward what he calls agentic engineering - where multiple AI agents handle the entire software development cycle. "AI does not just write code. It runs and tests it as well," Shrestha said. "These tools can help plan projects, design software, check for problems and even suggest features a company may need."
Shrestha said this shift lets Nepali engineers compete with counterparts in larger markets such as India. Tasks that once took weeks can now be completed in a few hours. But he warned that the model carries risks, particularly around security. Without a solid understanding of programming fundamentals, reliance on AI can introduce vulnerabilities. Effective use requires strong prompting skills and the ability to guide AI step by step.
Security risks and the skills gap
Rojesh Man Shikhrakar, director of AI education and talent development at US-based firm Fusemachines, said the main concern is cybersecurity. "People without basic programming knowledge relying entirely on AI can create serious security gaps," he said. Shikhrakar pointed to past incidents, such as municipal data leaks, as examples of how weak systems can be exposed when security practices are not properly followed.
Some industry experts argue that the bigger issue in Nepal is not job losses from AI, but a widening skills mismatch. Gaurav Pandey, president of NAS-IT, said many graduates enter the job market without the skills employers need. "Students and their families spend between Rs1.4 million and Rs2 million on higher education, yet many struggle to meet even the basic requirements for entry-level technology jobs," Pandey said. "The gap between university education and industry demand remains significant."
Tools driving the shift
Developers in Nepal are increasingly relying on tools such as VS Code, GitHub Copilot, ChatGPT Codex, Claude, Gemini, and Cursor as coding assistants. Platforms like Lovable are used for website development, while Microsoft Power Apps and Copilot serve enterprise applications. Free tools such as Ollama are gaining popularity, alongside AI app builders like Base44, which lets non-coders create applications. Beyond software development, AI is also used for research, documentation, and analysis. Naveen Jaiswal, chief product officer at Codewing Solution, said tasks that once took hours can now be completed in minutes, and AI's improving capability in the Nepali language is likely to drive further adoption in Nepal.
Global signals
AI-driven development is advancing globally. An AI startup founder, Ziwen Xu, is reportedly experimenting with building a version of the game Grand Theft Auto VI using a large AI model, Claude Max 20x, through prompts alone. Reports have also suggested that Elon Musk's space company, SpaceX, explored acquiring the vibe coding startup Cursor for about $60 billion. Microsoft and Google have both acknowledged that AI now assists in generating a significant share of their code.
These developments offer a glimpse of where software development may be heading. "If the next generation cannot work with AI agents and manage complex systems, they will struggle in the market," Shrestha said. For developers looking to build those skills, resources like the AI Learning Path for Software Developers provide structured training on AI coding tools and agent-based workflows.
Why this matters for IT and development professionals
The shift toward AI-assisted coding is not theoretical - it is already changing how state-owned enterprises and private firms build software. Engineers who can guide AI agents, spot security flaws in generated code, and integrate these tools into full development cycles will have an edge. Those who treat AI as a replacement for fundamentals risk creating the very vulnerabilities that Shikhrakar described. The skills employers need are shifting, and the gap between traditional computer science education and on-the-job requirements is widening.
Your membership also unlocks:
