First decide what you actually want to learn
Inside the words "I will learn AI" there are at least three different things hidden, and people waste time because they do not separate them at the start.
- Using AI. Using ChatGPT or Gemini to do your study, writing, business and job work faster. This is what most people actually need.
- Building things with AI. Websites, chatbots, automation. Knowing code helps, but these days you can go a long way without it.
- Learning to build AI. Machine Learning, Model Training, research. This is a proper engineering career.
Most people who start with number three give up in three weeks. They see matrices and derivatives and feel this is not for them. Yet what they really needed was number one.
The path below is arranged from one towards three. Go as far as you need and stop there. There is no shame in that.
Sadia's calculation: start with time
Say Sadia is in the second year at a university in Dhaka. She cannot decide whether to learn to use AI or go straight towards ML.
First she made an honest calculation. After classes, travel, tuition and family work, she has 7 hours a week. Two hours on each of the two days off, and half an hour on each of the other five days. If she counted more than that, it would not be a plan, it would be wishful thinking.
With these 7 hours a year comes to about 350 hours. In that time she can learn to use AI very well, set up a working system for her own studies, and even reach the foundation of Python. But she cannot reach job level in ML in one year, and that is not her failure, it is arithmetic.
So Sadia started with Track A, and decided to look at the decision again after eight weeks. Her eight week routine is given below in this post.
You should start with the same task. Sit with a notebook and write down your real free time in a week. A plan that does not match your real time will fall apart in three weeks.
The first two weeks, just use it
No courses, no books, no videos. Half an hour a day, your own real work.
If you are a college student, paste today's hard paragraph from class and say: explain this in simple Bangla, then ask me three questions so I can tell whether I understood. If you are a shopkeeper, have it write Facebook posts for your five products. If you are a teacher, make ten MCQs on a chapter. If you are a freelancer, have it draft replies to a client's English email.
After two weeks you will feel on your own where it is brilliant and where it is poor. This feeling is the real learning, and no course can give it to you.
One useful habit: write down each day's uses in a notebook or a notes app. At the end of the month you will see three or four tasks coming round again and again. Those three or four are your real use, and the rest is curiosity.
How exactly to start on day one
If you sit down today, go in this order. The whole thing is one hour of work.
- Pick one tool, ChatGPT or Gemini, either one. Do not waste time signing up for both.
- Open a free account. There is no need to spend money in the first month.
- Pick one task you are stuck on right now. Not an imaginary task, a real task of today.
- Write it first the way it comes to your mind. Read the answer.
- Now ask for the same task again, but this time write who you are, who it is for, and how long the answer should be.
- Put the two answers side by side. Seeing the difference is the real lesson of day one.
- Open a note and write today's date and the task. This note is the proof of your progress after eight weeks.
The eight week routine, the way Sadia arranged it
This is not a course syllabus, it is an order of work. It is arranged on the basis of 7 hours a week.
| Week | What you will do | What you will have at the end of the week |
| 1 | Half an hour a day, your own real work, with no rules | A list of what you had it do over the seven days |
| 2 | The same work, but polish each answer at least twice | Seeing with your own eyes where it is good and where it is poor |
| 3 | Learn to add context, examples and structure | Three written Prompts for your own work that you will need again and again |
| 4 | Verification week, ten questions on the subject you know best | Your own list of where it makes mistakes |
| 5 | Images and files, getting it to read screenshots, summaries from PDFs | A question bank made from your own notes or a book |
| 6 | Understanding English material in Bangla, dictating with voice | One English chapter, noted down in your own language |
| 7 | Start making one small thing, small enough to finish | A half finished piece of work |
| 8 | Finish it and show it to at least three people | A working link or file that can be shown |
After eight weeks the decision will become clear on its own. If, at the making stage, you want to know more about what happens inside, Track B is for you. And if you are satisfied with getting the work done, stay in Track A. Both are honourable answers.
The real rules of writing a Prompt
What is sold in the market as Prompt Engineering has a working part that can be told in five minutes. There is no magic sentence.
- Give the context. I run a small clothes shop in Mirpur, Dhaka, and my buyers are mostly middle aged women. Adding this one line will make the quality of the answer jump.
- Say who it should write like. Explain like an experienced Bangla teacher.
- Give an example. Show an old post you like and say: write in this style.
- Say the structure. Five bullets, each one line.
- Do not stop after one try. The first answer is a draft. Say: the second point is too bookish, write it in spoken language.
The last point is the most valuable. The difference between people who get good results from AI and those who do not is not in the words of the Prompt. It is that one person polishes three times and the other gets disappointed at the first answer and leaves.
See the difference in a small example. Weak Prompt: write a Facebook post. Good Prompt: I sell cotton three piece sets in Mirpur, a new lot has come before Eid, my buyers are middle aged women who like to come to the shop, see and buy. Write a 3 line post, do not mention the price, and end with a line asking them to message the inbox. In the second one you spent an extra 40 seconds, and the result will be several times more useful.
The habit of verifying, which nobody teaches
AI says wrong things with confidence. This is not an occasional accident, it is part of the way it works. So the second step of learning is the rule of doubt.
The simple rule: any fact where a mistake would cost you money, health, results or reputation should not rest on an AI answer. For prices, dates, government rules, admission requirements, medicine doses and sections of law, use the official website or a direct phone call.
And do one test. Pick a subject you know best, your own area or your own profession. Ask AI ten questions about it. Once you see the mistakes with your own eyes, the caution will stay for the rest of your life.
Sadia did exactly this in the fourth week, on a subject from her own department. Of the ten answers, seven were useful, two were vague, and in one a book name came up that does not exist. That one mistake changed the way she uses it. Now when she gets a book name or a source, she looks it up first, then uses it.
Now two roads
| What you want | What you will learn | Time in practice |
| AI for daily work | Prompt, verification, your own 3 tasks | 2 to 4 weeks, 30 minutes a day |
| Doing twice the work in your own profession | Building a workflow for your own work | 2 to 3 months |
| A freelance service with AI | One specific service, portfolio, client contact | 4 to 8 months |
| Working with Python | Python, files and data, API | 6 to 12 months, 2 hours a day |
| A job in ML | The above, plus ML basics, projects, English | 1 to 2 years |
These times are counted for regular study. If you sit one day a week, the time will not double, it will become four times, because every time you will forget what came before.
Track A: your own work, without code
The biggest mistake on this track is signing up for twenty tools. Knowing one tool well is far more useful than knowing twenty tools by name only.
A three month plan could look like this:
- Month 1. Pick one chat tool, use it every day, and build three Prompts for your own work that you will need again and again.
- Month 2. Add images and documents. Making images, reading text from images, summaries from PDFs, dictating with voice.
- Month 3. Build one real thing. A shop catalogue, a coaching question bank, your own portfolio site. If you need a site without knowing code, there are tools now for building by chatting in Bangla, such as BanglaCodes, and if you learn code later you do not have to throw that site away.
The thing you build is your proof. In an interview or in front of a client, a working link does far more than a certificate.
There is only one rule for choosing the month three task: take something you can finish in eight weeks. A small thing built completely is worth much more than a big thing built halfway, because the experience of finishing is the real skill.
Track B: the technical path, keeping the order
If you keep to this order, far fewer people drop out.
- The basics of Python. Variables, loops, functions, lists, reading and writing files. Two to three months.
- Real small programs. Reading an Excel file and working out figures, renaming the files in a folder. The learning becomes solid here.
- Using an API. Sending a question to an AI Model from code and getting the answer. Here you will understand what Token, cost and limit really are.
- Handling data. Tables with pandas, graphs, cleaning. A lot of real jobs are exactly this.
- After that, the basics of ML. At this step you will learn the maths, because now you will understand why it is needed.
At each step it is worth saying separately what you will learn and what you will leave for now, because most time is wasted on things that are not needed.
| Step | What you will learn | What you will leave for now |
| Basics of Python | Loops, functions, lists, dictionaries, files | Classes, decorators, threading |
| Small programs | Working with files and folders, handling errors | Frameworks, database design |
| API | Sending requests, reading JSON, keeping keys hidden | Building your own server |
| Data | pandas, cleaning, simple graphs | Big data tools |
| Basics of ML | Splitting data, yardsticks, overfitting | Training a big Model yourself |
Those who advise picking up linear algebra and calculus at the very start are not wrong. They just forget that most people drop out at that step. Once you understand why it is needed, learning the maths is much easier.
Rakib's calculation: how long Track B really takes
Say Rakib works at a company in Chattogram, is 29 years old, and has never done coding. After office he can give 1 hour a day, 5 days a week.
The sum is simple. 5 hours a week, about 20 hours a month, 120 hours in six months. In these 120 hours he can finish the basics of Python and small programs, that is the first two steps above, properly. He will touch the API step too.
But in six months he will not reach job level in ML. Those who say you can become an ML engineer in six months are either counting four or five hours a day, or they mean something else by ML engineer.
There is one more thing in Rakib's calculation that most plans leave out. At least one week in every month will be lost, to illness, office pressure, family work. So he has counted the six months of work as eight months. If you keep this extra, you will not need to give up when the plan breaks.
The problem of English and its real fix
The truth is that almost all good learning material is in English. This is unfair, but it is real.
But there is a working fix for it now, and it is AI itself. Copy a part of an English tutorial and give it to AI, say: explain this in Bangla, then explain the hard words separately. For videos, take the subtitles or transcript and do the same.
One condition goes with it. Do not use this to avoid English, use it to understand. Because in a job or in freelancing you will have to read documentation in English and write to clients. Keep up the habit of reading one piece of English a day at the same time.
There is a trick that does two jobs at once. Read the English part yourself first, write down as much as you understood, then get the Bangla explanation from AI and compare it with your own understanding. This keeps the habit of reading English and also makes sure of the understanding. If you only get it translated, your English will be in exactly the same place after six months.
Free resources that will really help
- The free versions of the tools. ChatGPT, Gemini, Claude. The free version is enough for learning, there is no need to spend money in the first month.
- Google's Machine Learning Crash Course. Free, well organised, for the basics of ML.
- Kaggle Learn. Short free courses on Python and data, hands on.
- Google Colab. You can run Python in the browser, with nothing to install on your own computer. If your machine is old or weak, this is your place to start.
- Automate the Boring Stuff with Python. The book can be read free online, and the way it teaches is based on tasks, not on theory.
- Python's own documentation. It will feel hard at the start, but from the third month it is the most reliable place.
- Hugging Face's free course. If you want to work with language Models.
- freeCodeCamp. On YouTube and on the website. For the basics of programming.
- CS50. Harvard's free course, excellent for understanding how a computer works.
- Khan Academy. If your foundation in maths is weak, you can fix it here, and you will not need it before step five.
- Coursera. Many courses can be viewed free, and you pay only if you want the certificate.
- In Bangla. There are good channels on YouTube teaching Python in Bangla, and you will find some free content on platforms like 10 Minute School. Pick one series and finish it, which is better than starting ten and finishing none.
It is also easy to get confused about what to pick up when. Keep the table below in mind and it will be simple.
| Where you are now | What to pick up | What not to pick up yet |
| Just starting, only learning to use it | The free versions of the tools, and your own real work | No course at all |
| Decided to take up Python | Colab, Automate the Boring Stuff, a Bangla video series | ML, maths, frameworks |
| Python basics done | Kaggle Learn, Python documentation, small projects | Deep theory |
| Working with data | Kaggle datasets, ML Crash Course | Training a big Model yourself |
| Will work with language Models | Hugging Face's course | Reading research papers, that is later still |
What not to waste time and money on
- Prompt masterclasses costing thousands of taka. The working part of what they teach is in the one section above.
- A PDF collection of 500 Prompts. You will not even open them. Three Prompts for your own work are worth much more.
- Signing up for a new tool every week. Keeping up with news of new tools is not work, it is pretending to learn.
- The dream of training an LLM yourself. The cost and infrastructure are beyond an individual, and you do not need it.
- Videos made by AI at home earning thousands of taka a day. The people who make these earn from those videos, not from the system.
- Only collecting certificates. A certificate is not completely useless, but clients and employers will first look at what you built.
- Reading AI news every day. When a new Model comes out, nothing in your work will change, and if it matters the news will reach you anyway.
- The comparison pit. Spending a week arguing about which tool is better. If you work with the one you have in hand, you will get ahead.
Not every paid course is bad. The test is simple: does the course make you build something, or does it only show videos?
Common mistakes and how to fix them
- Mistake: making a perfect plan before starting to learn. Fix: use it for half an hour today. The plan will be settled as you work.
- Mistake: starting three things at once, Python and design and video editing. Fix: pick one for eight weeks, and write down the others and keep them aside.
- Mistake: thinking that watching videos is learning. Fix: after every hour of video, do something with your own hands for at least half an hour. If you do not, it will all be wiped in a week.
- Mistake: giving up when an error comes. Fix: copy the error message, give it to AI and ask what it is saying. Fixing errors is the real part of learning programming.
- Mistake: leaving your own subject aside and copying someone else's project. Fix: pick a real problem from your own study, shop or profession. Interest will pull you along.
- Mistake: stopping after four weeks because you think nothing has happened. Fix: open the note from the first day. Progress cannot be seen every day, it is seen when you look back.
- Mistake: waiting because you think you need the paid version to learn. Fix: start with the free version. Think about it when you keep hitting the daily limit.
When to stop learning with AI
AI is a great helper for learning, but in a few places it does direct harm.
- Where the struggle is the learning. Solving maths, finding a bug in code, writing a first draft. If you ask for the answer, that skill will not be built.
- To understand a new subject the night before an exam. You do not have the knowledge then to catch a wrong explanation, and no time to verify.
- Getting AI to write an assignment or thesis and handing it in. Apart from breaking the institution's rules, you will be caught the moment you are asked in a viva.
- On a subject where you have no way to verify. Your department's rules, admission dates, scholarship conditions, take these from the office or the website.
- When you are tired and just want to finish the work. In that state you will not verify, you will only copy. Better to stop and do it the next day.
How far you can go with a phone
All of Track A can be done on a phone. Chat, images, documents, voice, all of it runs on mobile. Most people in Bangladesh will start here, and that is natural.
To get into Track B you will need a computer. An old laptop will do, there is no need to buy a new one. Many have started with a college or university lab, a friend's machine, even a cyber cafe. The cost of internet is a real obstacle, so if you get into the habit of downloading videos and watching them offline, you will save data.
If your machine is weak, let me remind you about Colab. The code runs somewhere else, and your browser only shows the result. You do not have to install anything on your own computer, and the biggest obstacle at the start is gone.
Where AI helps and where it harms in studies is in the post on AI in studies.
Questions and answers
I am 35 and I have a job. Is it too late to start now? No. In Track A age is no obstacle at all. In fact, if you have work experience, you know which problem needs solving, which newcomers do not. In Track B too, age is not the main obstacle, time is. If you can give one hour a day for a whole year, you will go far.
I am weak in maths. Is AI not for me? To learn to use AI you do not need maths at all, not even a little. You will need maths only if you want to become an ML engineer, and even then not at the start but at the third step. And then the maths will be easier to learn because you will know why it is needed.
Which programming language should I start with? Python. This is the only answer, and there is nothing to waste time on in this decision. Almost everything in AI work is done in Python, and it has the most learning material too.
Will the free version do, or must I take the paid one? For the first month or two the free version will go well. When you see that you hit the limit every day and the work is earning you money or saving you time, think about spending money then. Not before.
How long each day will really get me somewhere? In Track A, 30 minutes a day is enough, but it must be six days. In Track B, one hour a day is the minimum, and if you give two hours the time is cut nearly in half. The most important thing is regularity, because when there are gaps the cost of forgetting what came before is the biggest.
Should I learn alone or join a group or a course? Learning alone is possible, but the risk of dropping out is higher. The fix that costs nothing is to find a partner, a friend or a colleague, and once a week show each other your progress. If you know you will have to show someone, the work gets done.
If I get AI to write the code, will I learn? Not in the first few months. It will do harm instead. Make the rule like this: try to write the code yourself first, and if you get stuck, ask AI why you are stuck, take the answer and fix it yourself. Having code written for you is a normal thing after the foundation is done, not before.
How soon can I start earning with AI? The honest answer: knowing AI is not by itself a way to earn. Income will come from selling the solution to a specific problem, and AI will help you do that work faster. Someone who already has a skill, such as writing, design, accounts or teaching, will see a difference in three or four months. Starting from zero will take longer.
Can I get a job or work without a certificate? Yes, if you have a portfolio. Employers and clients will first look at what you built and whether it works. A certificate can give an extra edge, but it will not take the place of a link to a working piece of work.
In short
- Learning AI can mean three different things: using it, building things with it, and learning to build it yourself, so decide at the start which one you need.
- For the first two weeks, no course. Have AI do your own real work for half an hour a day, and that will show you its strength and its weakness.
- There is no magic in Prompts: give context, give an example, say the structure, and do not stop at the first answer, polish it two or three times.
- One month is enough just to learn to use it, two or three months to apply it to your own profession, and to reach job level in Python and ML it will take one to two years even with regular study.
- Most good material is in English, so understand English text in Bangla with AI, but do not give up the habit of reading English.
- Costly Prompt courses, PDF collections of Prompts and signing up for a new tool every week are a waste of time. Instead, build one working thing that you can show people.