
The honest answer is that AI is far more likely to change most jobs than to delete them outright, but it will shrink some roles, and the change will not be even. Some people will find their work easier and better paid. Others will find that the tasks that made up their job are now done by software.
This guide looks at what the main research actually says, which kinds of work are most exposed, what is different about Nepal, and what workers, students and business owners can do now. It is written by a software team that builds and uses AI tools, so we have an interest in the topic; we have tried to stick to what the research says rather than to hype or fear.
First, a job is a bundle of tasks
Most arguments about 'AI taking jobs' go wrong because they treat a job as one thing. A job is a bundle of tasks. An accountant enters data, checks figures, explains results to a client and takes responsibility for the answer. AI can already do parts of the first two. It does much less of the last two. When a tool takes over some tasks, the job can get easier, can change shape, or, if the automated tasks were most of the work, can shrink.
That is why researchers talk about exposure rather than replacement. A job is exposed when AI can perform some of its tasks. Exposure is not the same as job loss.
What the research says
- IMF (January 2024): almost 40% of global employment is exposed to AI. The share is about 60% in advanced economies, 40% in emerging markets and 26% in low-income countries. In advanced economies, the IMF says roughly half of exposed jobs may benefit through higher productivity, while for the other half AI may take over key tasks, which could reduce demand for labour, wages and hiring, and in the most extreme cases make some jobs disappear.
- ILO (2023): the International Labour Organization's study of generative AI concluded that its main effect is likely to augment rather than automate jobs. It expected greater impacts in higher-income countries, and a larger effect on women because of their high share of clerical work.
- World Economic Forum (January 2025): its Future of Jobs Report projects that by 2030 about 170 million new jobs will be created, equal to 14% of today's employment, while 92 million will be displaced, equal to 8%, for a net gain of 78 million, or 7%. These are projections to 2030, not yearly figures, and they come from a survey of more than 1,000 large employers.
Read together, the studies point the same way: large parts of the workforce are touched, many jobs are changed rather than ended, and some are lost while others are created. None of them says that most jobs will vanish.
Caveats worth knowing
- These are forecasts. They measure exposure and employer expectations, not what has already happened. Forecasts about technology have often been wrong in both directions.
- Exposure is not outcome. What happens depends on what employers choose to do with the productivity: cut staff, serve more customers, or build new services.
- Poorer countries are exposed less, for now. The IMF notes that low-income countries often lack the infrastructure and skilled workforce to benefit from AI, which could widen the gap between countries. Less exposure is not the same as being better off.
Which work is most exposed
AI tools are strongest at tasks that involve information: reading, writing, summarising, sorting, translating, answering routine questions and producing first drafts. So the tasks most exposed tend to be:
- Routine information work: data entry, basic report writing, simple document handling, first-line customer replies.
- Templated creative and technical output: standard marketing copy, simple graphics, routine code and boilerplate.
- Entry-level analysis: pulling figures together and summarising them.
The work that tends to hold up better is work that depends on being physically present, on trust and relationships, on judgement in unclear situations, and on being accountable for the result. That covers many hands-on trades, care work, teaching, and roles where someone has to decide, negotiate or take responsibility. These are tendencies, not guarantees, and tools keep improving.
What is different about Nepal
- Services sold to overseas clients are in the front line. A lot of Nepal's IT and outsourcing work is done for clients abroad. Where the work is routine, such as templated websites, simple data processing or scripted support, clients will ask why it still costs what it did. Where the work is understanding a business problem, designing a solution and standing behind it, there is more room.
- Developers already use AI heavily. In the Stack Overflow 2025 survey, 84% of developers said they use or plan to use AI tools, yet 46% said they do not trust the accuracy of the output. In practice that means more work reviewing, testing and understanding code, not less need for engineers.
- Mixed Nepali and English matters. Tools perform differently across languages, so local knowledge of how customers actually write and speak is an advantage.
- Adoption will be uneven. Many small businesses are only starting to use even basic software, so the change will arrive gradually and at different speeds in different sectors.
If you work for a living
- Learn the tools in your own field. Someone who knows the work and the tools is more valuable than someone who knows only one of them.
- List your tasks, not your job title. Which are routine and could be automated? Which need judgement, trust or presence? Spend more of your time on the second kind.
- Get good at checking. Knowing when an AI answer is wrong, and being able to say why, is a skill that employers need.
- Build skills that tools do not replace: understanding customers, explaining clearly, framing a problem, taking responsibility.
- Keep the fundamentals. If you cannot do the work without the tool, you cannot tell when the tool is wrong.
- Review it every few months. Check how your own tasks are changing and adjust before you are forced to.
If you are a student or starting out
Learn the fundamentals first, and use AI as a tutor rather than a shortcut. Build real projects you can explain, because showing what you can do and why is what separates you from a tool. If you want a path into the field, read our guides to becoming a software engineer in Nepal and the AI engineering skills and roadmap.
If you run a business
- Start from tasks, not people. Look for repetitive tasks that are slow or error-prone, and automate those first.
- Keep people in the loop. Let the tool draft and a person approve, especially for anything that affects customers or money.
- Retrain before you replace. The people who know your customers and processes are the ones who can supervise the tools well.
- Be honest with your staff. Say what you are trying and why. Fear spreads faster than facts.
- Measure. Track time saved and errors before and after, and expand only what works.
Our guide to adding AI to business software in Nepal shows where to begin.
Common myths
- 'AI will take all the jobs by a certain year.' No serious forecast says this. The research above expects both jobs created and jobs lost.
- 'It is all hype and will pass.' The tools are widely used already, including by most developers, and they are improving.
- 'Only low-skilled jobs are at risk.' Exposure is highest in office and information work, including skilled roles, which is why the IMF finds advanced economies more exposed.
- 'There is nothing I can do.' Individuals and organisations that adapt early have a real advantage.
For the bigger picture
AI is the latest step in a long line of tools that changed how people work, from the abacus to the microprocessor. Our timeline from abacus to AI puts it in context, and how Nvidia's chips are made explains the hardware behind it.
Next steps
If you want to use AI in your business without losing control of quality, see our AI automation work. If you want to build the skills yourself, look at our AI and machine learning course, or contact us.
Sources and further reading
Have a project in mind?
Tell us what you're trying to solve. Or keep reading: AI Agents for Business: A Practical Guide for Nepal.



