Workflow integration

Three modes of use, three different effects

A useful framing. Generative AI ends up serving you in roughly three modes, and each has a different effect on you.

Mode 1 — Amplifier. You bring a skill; the model speeds up the parts of the work that don’t require your judgement. A senior writer uses it to draft routine sections; a senior programmer uses it for boilerplate. The skill stays sharp because the interesting parts of the work remain yours.

Mode 2 — Apprentice. You’re learning a domain; the model is a always-available teacher. You ask it to explain, to walk you through, to critique your attempts. Used this way, generative AI is an accelerator for skill acquisition.

Mode 3 — Substitute. You don’t have the skill, and you ask the model to do the work for you. The output is, on the surface, similar to what an expert would produce. But you cannot verify it, cannot extend it, cannot maintain it. Over time, this mode makes you dependent and brittle.

The same tool produces all three effects. The difference is how you use it and what you do with the output. The same conversation with ChatGPT can be Mode 1 for a senior employee, Mode 2 for a junior, and Mode 3 for someone trying to skip a step they shouldn’t.

The skill question

A working principle: use generative AI to do faster what you already know how to do; use it to learn what you want to know; do not use it to skip skills you need to acquire.

This is harder than it sounds. The temptation when the model can produce a passable version of something is to not learn that thing — why bother? But there are categories where the underlying skill matters more than the immediate output:

• Writing in your own voice. If you outsource your writing, your voice never develops.
• Understanding code you ship. If you don’t read it, you can’t fix it when it breaks.
• The judgement that makes your work valuable. AI can imitate craft; it cannot replace the judgement that decides which craft to apply.

For these, the right pattern is: do the work yourself first; then use the model to refine, polish, or speed up. The model should not be the first draft of skills you are still building.

Building workflows that don’t decay

A working pattern from people who have used these tools productively for several years.

Habit 1 — Maintain a prompt library. A simple notes file (or any text store) of prompts that worked well for tasks you do often. The library is a personal asset that compounds. Each new task is faster because the template is already there.

Habit 2 — Read the output as if a junior wrote it. Treat AI output as a first draft from a competent junior. Read critically. Edit. Push back. Reject when the draft is wrong. This habit preserves both your skill and the quality of your output.

Habit 3 — Keep some work AI-free. A specific category of work you do without the tool. Pick something that matters to you — writing in your voice, working through a problem from scratch, learning a new domain by struggling with it. Protect that practice. Skills that aren’t exercised, weaken.

Habit 4 — Track when AI helped and when it hurt. A monthly check-in: which tasks did I use AI for, and was the result better or worse than I would have produced alone? Honest answers reveal where the tool earns its keep and where it doesn’t.

Habit 5 — Re-evaluate the tools every six months. The model landscape changes fast. The right tool in January is often not the right tool in July. Keep light habits of trying alternatives.

The Nepali context

A specific observation for the audience of this course.

The Nepali knowledge economy in 2026 is at an unusual moment. Generative AI is available — anyone with a phone and a few rupees can use frontier models. The cost of producing draft text, draft images, draft code is collapsing. This is mostly good for Nepali professionals: tasks that used to require senior expertise can now be partly done by anyone who can prompt well.

But the same change creates new ways to fall behind. Professionals who use generative AI to amplify their existing skill will move faster. Professionals who use it to substitute for skills they haven’t built will produce work that is fluent but cannot be defended. Both will look the same to the casual observer. Over time, the difference compounds.

The advice that applies to Nepal specifically: invest in the skills that AI doesn’t replicate. Deep local knowledge — how things actually work in Nepali contexts. Long-form judgement — what should the right answer here be, given everything I know about this client / this market / this community. Relationships and reputation — the trust that lets others rely on your work. These are the durable assets in the AI era, and they are precisely the assets that Nepali professionals have, often more than foreign competitors.

A closing paragraph

If this course has worked, you should now be able to say something like this in your own words:

Generative AI is a fast, fluent, frequently-wrong writing partner. It is excellent at producing first drafts in modalities where data is plentiful, particularly text and images. It is unreliable for facts, biased toward whatever was common in its training data, and confidently wrong in ways that look right at a glance. To use it well, I prompt clearly, verify factual claims, specify when I want my own context, watch for cultural defaults that miss Nepal, protect confidential data, disclose use where it matters, and keep working on the skills I want to preserve.

That paragraph is what this course has been building toward. If you can hold it steady, the rest is practice. The tools will change; the principles compress to that paragraph.

What comes after

This course covered use. If you want to build with these tools — wire them into your own software, fine-tune them for your own data, deploy them on your own infrastructure — that is a different course, with a Python prerequisite. We will be releasing it.

If you want to think harder about AI in Nepal — sectoral case studies, opportunities, the policy questions — the AI for Nepal course will cover that ground.

For now, take what you have learned. Start small. Practice the prompts you’ve seen here on tasks you do this week. Notice what works and what doesn’t. Build a quiet, durable competence with these tools. The country needs more of that.

Check your understanding

Quick check

—

Which use of generative AI is most likely to preserve and grow your professional skill over time?

1. AAsking the model to do work in skills you have not yet learned, so you can submit it as your own
2. BRefusing to use the model entirely so your skills cannot atrophy
3. CUsing the model as an *amplifier* — doing faster what you already know how to do, and as an *apprentice* — learning new domains with it as a teacher — while keeping some work AI-free to maintain practice
4. DUsing only image generation and never text

Check answer→

End of course

You’ve reached the end. The course has tried to give you a practical toolkit: how to prompt, what each modality does, where the failures are, how to use these tools honestly and over the long term.

If you finish the final exam with 80% or better, you can issue yourself a certificate of completion. Take it as evidence that you can talk about generative AI sensibly, use it competently, and recognise the situations where it’s wrong for the job.

Then keep practicing. The tools change. The principles don’t.