Every few months a new AI strategy document appears somewhere in the world, and the pressure grows for Nepal to produce its own. That instinct is right. The risk is that we end up with a document that reads well at a conference and changes nothing on the ground.
Having studied AI awareness among Nepali professionals and written about AI policy in the national press, I keep returning to one conclusion. Policy fails here not because the ideas are wrong but because the plumbing is missing.
Start with what AI actually needs
AI runs on three inputs: electricity, connectivity and people. Nepal has made real progress on energy, but data centres and reliable power for digital services still lag. Internet access is widespread in cities and thin in the districts where public services need the most help.
The people question is the hardest. Our study of professionals found high curiosity about AI but shallow practical skill. Most respondents had tried a chatbot. Very few had used AI in a structured way at work. A policy that funds awareness campaigns but not hands-on training will not move that needle.
Three priorities that would matter
First, put AI literacy inside existing institutions rather than building new ones. Universities, civil service training centres and professional associations already reach the people who will use these tools. Give them curricula and trainers.
Second, make government data usable. Most ministries publish PDFs. AI systems, and researchers, need structured data. A modest open data mandate would do more for Nepal's AI ecosystem than any number of task forces.
Third, regulate uses, not the technology. Rules should target concrete harms such as fraud, discrimination and misinformation, and leave room for experimentation everywhere else.
The evidence habit
Whatever Nepal adopts, the policy should commit to measuring itself. Set baselines for AI skills and adoption, fund independent evaluations, and revise the strategy on a fixed cycle. That is how policy learns.
AI will reshape how Nepal works, learns and governs. The question is whether we shape that process deliberately or let it happen to us. A workable policy, honestly evaluated, is how we choose the first path.