The straight answer first
Everyone is asking this question, and two sides are giving two exaggerated answers.
One side says most jobs will be gone in a few years. The other says it is all hype and nothing will change. Both are incomplete, and both sides have something to sell.
The reality is in the middle. It is less dramatic, but more useful. Your job will probably still be there, but some of the tasks you do inside it every day will change. How much they change depends on how many hours of your day you sit at a computer making the same kind of thing again and again.
The job does not go, the tasks do
Think of any job as a collection of small tasks.
An office assistant's day goes on writing emails, typing reports, taking meeting notes, finding files, answering the phone and handling people. Of these, AI is good at three: email drafts, tidying up notes and the first draft of a report. It is not good at the rest.
Now the question changes. It is no longer "will I lose my job?" but "what percentage of my day goes on those three kinds of work?" If eighty percent of your day goes there, your pressure is high. If twenty percent goes there, your work will only get faster.
Do this sum once for yourself. For one week, write in a notebook what you spent time on. When you look at the list you will know where you stand. You do not need any expert's prediction.
Measure your own risk, in seven steps
It is better to measure your own work than to waste time reading predictions. You can do it in one week.
- Write in a notebook for seven days. Each day, note which task took how many minutes. Write the small tasks too, such as "email to client, 25 minutes".
- Group the tasks. At the end of the week, put similar tasks together and work out the total time for each group.
- Next to each group, answer four questions. Is the task done fully on a computer? Is it done in almost the same pattern every time? If it goes wrong, is the mistake caught at once? Can it be done without any information that lives inside your organisation?
- If all four answers are "yes", that is a high-risk group. The fewer the "yes" answers, the lower the risk. This simple count is truer for you than any report.
- In the high-risk group, pick the biggest task and try doing it with AI this very week. See with your own eyes how much worked and how much did not.
- Decide where the saved time will go. If the saved time goes into more of the same kind of work, your position will not change. If it goes into lower-risk work, such as talking to clients or verifying quality, it will.
- Measure again after three months. How the list has changed is the real news about your progress.
A format for measuring risk
| Feature of the work | Raises the risk | Lowers the risk |
| Where the work happens | Entirely on a computer | You need a body, a machine or to stand in front of people |
| Kind of work | Almost the same mould every time | The situation is different every time |
| Source of information | What is on the internet is enough | You need information from inside the organisation or the area |
| Cost of a mistake | At worst you write it again | Money, health or legal liability is at stake |
| Who sets the price | The buyer looks for the cheapest person | The rate depends on relationship and trust |
| Accountability | Nobody needs to sign | Someone must take responsibility and sign |
The more ticks you get in the right-hand column, the stronger your position. And here is the interesting part: the features in the right-hand column cannot be bought. They are built with time and relationships.
Which tasks are really under pressure
To be honest, we have to say that for some tasks the pressure is not imagined. People who do online work for foreign clients are already feeling it.
- Writing plain blogs or product descriptions at a per-word rate
- Simple translation, especially into or out of English
- Turning audio into text and making subtitles
- Pulling text out of images and doing data entry
- Very plain logo, thumbnail and social media post design
- Setting up template websites in the same mould
- First-level customer support replies that follow a script
Notice what they have in common. Each is done on a computer, the result is in hand at once, the work goes round and round in the same pattern, and the buyer usually looks for the cheapest person.
There is another risk that is talked about less. Many organisations now hire fewer juniors, because one experienced person can push through more work with AI than before. That does not mean jobs are being cut. It means fewer new jobs are being created. For people trying to get in this year, that is the real competition.
There is an uncomfortable side to this too, which gets little discussion. Junior tasks were the ladder for learning. People understood the subject by making small reports, and then moved up to bigger work. If the bottom rungs of the ladder go, what is the way up? Nobody has a good answer yet. A person entering now has to build the ladder themselves, which means making something with their own hands, outside paid work, and showing it.
Which tasks will not go easily
Now the other side, and for Bangladesh it is good news.
- Work done with the hands and the body. Electrician, plumber, mason, tailor, cook, driver, construction worker.
- Direct care of people. Nurse, physiotherapist, looking after children and the elderly.
- Work that stands on local trust and relationships. The neighbourhood shopkeeper, the local contractor, field sales, dealer management.
- Work where someone takes responsibility and signs. Doctor, lawyer, engineer, auditor. AI can draft, but no Model will take the blame for a mistake.
- Classroom teaching. AI will write questions and make notes, but managing forty teenagers and reading in their faces who does not understand is a human job.
- Information that is not on the internet. The market price in your area, the habits of the machines in your factory, what your buyer likes. AI does not know what it has never read.
Garments need a separate word, because this question comes up most in Bangladesh. An AI like ChatGPT will not run a sewing machine. The risk there comes from automation and robotics. That is a different thing, and it needs huge investment, so the change comes slowly. But the office work of a garments company, such as merchandising emails and reports, is certainly within AI's reach.
Profession by profession
| Profession | What AI can do now | What it still cannot do | What you should do |
| Content writer | Drafts, headlines, outlines | Your own experience, interviews, credibility | Leave plain writing and go deep in one field |
| Graphic designer | Fast variations, backgrounds, drafts | Brand thinking, understanding the client, print reality | Make AI a machine that produces your drafts |
| Data entry operator | Pulling data from images and PDFs | Messy paper, local handwriting, verification | Move toward verification and reporting |
| Junior developer | Plain code, finding bugs, tests | System design, responsibility, understanding real problems | Learn fast with AI and move up to senior work |
| School teacher | Questions, notes, first-draft marking of answer scripts | The classroom, motivation, handling parents | Cut preparation time and give it to teaching |
| Accountant | Formulas, voucher summaries, report wording | Tax law liability, local rules, signing | Get expert in the rules, not in typing |
| Call centre agent | Answers to common questions, call summaries | Handling angry customers, exceptions, decisions | Build skill in complex and sensitive cases |
| Electrician or mason | Nothing, at most advice | All of the work | Stay calm, and use AI for costing and marketing |
Six months of a freelance writer in Chattogram
Suppose a freelance writer in Chattogram writes plain blog posts in English. At the start, the work was simple. Two articles a day, each of a thousand words, at whatever per-word rate was going.
Two things happened in six months. One, new job ads started to carry terms like "lower rate if drafted with AI". Two, an old client said they were now making the drafts themselves and only needed editing.
The writer's first reaction was to cut the rate. That did not work, because you cannot compete with the person who goes lowest. Someone will always go lower.
On the second try the writer changed the kind of work itself. They picked one field, garment exports, because they already know that subject from their own circle. Now they have AI do the draft, then add two things by hand that AI does not have: quotes from talking to a real buyer, and a description of a real process in a factory.
What was the result? Articles a day rose to four. The per-word rate fell, but the total rate per article rose, because what is sold now is not writing but the knowledge wrapped around the writing. And the biggest change: earlier they had to look for new clients every month, and now three or four regular clients carry the month.
The thing to notice here is that they did not fight AI. They moved their work to a place where AI cannot go alone.
Walking the wrong road in a Dhaka office
We need to see the opposite picture too, because there is a lot of wrong use as well.
Suppose an employee in a mid-sized Dhaka office makes a weekly report. They start having AI write the report, and the time drops from 40 minutes to 10 minutes. It goes well for three weeks.
In the fourth week it comes out that they put in two numbers without checking them, and one department name was wrong. To the manager it looks like this: this person used to work slowly but could be trusted, and now works fast but cannot be trusted.
The lesson is bitter but simple. A part of the saved time has to go back into verification. It is tempting to cut a 40 minute job to 10 minutes and use the other 30 minutes elsewhere, but cutting it to 15 minutes and spending 5 minutes on verification is what lasts.
And this shows which skill has gone up in price. The price of making a draft has fallen. The price of catching a mistake has risen. A person who can catch mistakes will not be replaced by AI, because to catch mistakes you really have to know the subject.
Which skills are gaining value and which are losing it
This format will help when you decide where to invest your time.
| Skill | Which way the price is going | Why |
| Fast typing and tidy writing | Falling | Drafts are cheap now |
| Catching mistakes and verifying quality | Rising | The mistake now arrives sounding confident |
| Plain translation | Falling | The machine does a fair job |
| Translation that understands terms and context | Rising | Nuance is out of the machine's reach |
| Bargaining and handling clients | Rising | You cannot bargain with a machine |
| Deep knowledge of one field | Rising | It is the only basis for catching mistakes |
| Fitting in stock templates | Falling | Anyone can do it themselves |
| Taking responsibility and signing | Rising | No Model takes the blame |
What the people who are holding up are doing
One pattern stands out. The people who are not under pressure do not think of AI as a rival. They think of it as an employee.
A designer used to give the client two options. Now they give six. The client is happy, and the designer still gets the work, because a client cannot bargain with AI, but can with a person.
An accountant has AI do the voucher summaries and the wording of reports, and spends the saved time reading the rules. They understood that their value is not in typing but in being able to take responsibility.
A call centre agent leaves the call summaries and the common replies to the machine and takes only the complex and angry customers themselves. Those are the cases where the organisation loses or gains the most, so this agent's position is stronger than before.
One thing is common to all of them, and it is the habit of verifying. A person who can look over what AI gives and catch the mistakes is valuable. A person who hands it in without looking will quickly lose trust.
There is another common point. None of them moved to a new profession. Each changed the split of work inside their old profession: they gave the machine the repetitive part and kept the part that needs judgement. It was not a big leap. Changing the split was enough.
What you can do in 30 days
You do not need an expensive course. A small plan is enough.
- Week one, measure your work. Finish the first four of the seven steps above this week.
- Week two, have AI do one task completely. Pick the most boring, repetitive task. Start by assuming the first try will be bad.
- Week three, start collecting your own instructions. Keep the instructions that gave good results in a note on your phone. After a month this note will be your most valuable asset, more than any list you could buy.
- Week four, build the habit of verifying. In every result, check the numbers, dates and names separately. This habit is what will protect you from being replaced.
- At the end of the month, show it at the office or to the client. "This task used to take an hour and a half, now it takes 25 minutes, and the quality has not dropped." An employee who can say this sentence is usually last on the layoff list.
Along with this, hold on to four things, because their value has risen, not fallen. Deep knowledge of your subject, because you cannot catch mistakes without it. Skill in working with people, meaning understanding clients, setting the price, delivering on time. Knowledge of your area, which is not in the internet's English data. And the willingness to take responsibility, because when there is a mistake a person has to stand up.
The mistakes being made most often now
| Mistake | What happens | Fix |
| Competing by cutting your rate | Someone cheaper will always exist | Pick a field, go deep, sell the result |
| Handing in AI's work without verifying | Fast work, but trust is lost | Give part of the saved time to verification |
| Buying a "prompt course" hoping to change profession | Money goes, skill does not come | Use it every day in the work you already do |
| Giving the office's confidential files to an outside service | Breach of policy and risk to data | Ask permission first, change the details |
| Putting all the saved time into more of the same work | Your position does not change | Move the time to lower-risk work |
| Waiting without learning anything new | Someone else makes the decision | Try changing one task this month |
When not to try using AI at all
If you try to push it into every task, you lose time instead of saving it. There are places where the honest answer is: leave it.
Where the task is small and you are quick by hand. Sitting down to arrange instructions to write a three-line message is foolish. The task would have been done in the time it takes to write the instruction.
Where the organisation's policy does not allow it. In many offices it is forbidden to give customer information or internal documents to an outside service. Starting without knowing the rule is a bigger risk to your job than any AI mistake.
Where the responsibility will be written in your name. An audit opinion, an engineering calculation, a legal reply, a medical decision. You can take a draft, but you must understand every line in it yourself, or it is not your work.
Where you are skipping the step of learning. When you learn a new skill, if you get past the first hard step with AI, the skill does not form. Later, when you need to catch a mistake, there will be nothing in your head to catch it with.
Where the relationship is the point of the work. A personal message to an old client, a reply to a colleague's news of a bereavement, scolding or praising someone on the team. In places like these, a real voice is worth much more than tidy language.
Claims that are exaggerated
You will keep hearing some claims, and it is best to doubt them.
Ads like "learn AI prompts and earn a lakh taka a month". Writing prompts is a useful skill, but on its own it is not a profession. The person selling you the course is making their income from exactly that.
Talk like "AI will run the whole business by itself". AI makes drafts. It does not make decisions. And when there is a mistake, it does not pay compensation.
Predictions tied to a particular year. Who will lose what by which year is mostly a guess, and the more confidently a guess is told, the more it gets shared.
Denial from the other side is not right either. A person who says "nothing will happen, it is all hype" is not looking at the facts either. People whose work has already become cheap are not lying.
Questions and answers
I am over 40. Is it possible to learn something new? What you are being asked to learn here is not programming. It is instructing a tool well, and that can be done in Bangla. In fact you have more of an advantage, because the experience needed to catch AI's mistakes is something newcomers do not have and you do.
What should I have my children study? Instead of assuming some one subject is "safe", ask this question: is the subject just memorising facts, or does it involve hand skills, working with people or taking responsibility? Health, engineering, teaching and skilled technical trades are still firm ground. Along with any subject, teach them to use AI.
Is starting freelancing foolish now? Not foolish, but the road of seven or eight years ago is no longer there. The trick of grabbing plain work at the lowest rate will not work now. Pick one field, go deep in it, and sell the client a result, not hours.
If my office brings in AI, should I learn it beforehand? Yes, and it is the lowest-risk investment there is. An employee who can show one way to save time with AI in their own department is usually last on the layoff list. But get permission before you give the office's confidential information to any outside AI.
AI knows Bangla well, so will Bangla work also come under pressure? AI's hold on Bangla is growing, so plain Bangla writing will come under pressure too. But it is still clearly weak on the fine points of the language, regional feel and local context. A person who writes good Bangla and can fix AI's draft will have more work, not less.
I have just graduated and junior posts are shrinking. What do I do? Do not wait for paid work. Make something with your own hands and show it. A small project, a site, an analysis, where you can explain every decision behind it. In an interview this ability to explain now helps more than a certificate, because the bigger question is not whether AI did the work but whether you understand it.
If everyone in the office uses AI, what is my separate value? When everyone has the machine, the difference is made in the input and in the verifying. Your organisation's internal information, knowing your customers, the reality of your area: others do not have these. Being able to give good input and being able to send back bad output are now the real skills.
I am working with AI. Should I tell the client? It depends on the nature of the work. If the client is buying a result, such as a design or a report, then which machine was used is usually not their question. But if the contract has a term about it, or if they ask directly, tell them plainly. Being found out for hiding it does more harm than having said it.
In short
- AI will not take the whole job of most people, but it will take some specific tasks inside it, so do the sum by task, not by job title.
- The most pressure is on computer work that repeats the same pattern at a cheap rate, such as plain content, simple translation and data entry.
- Hand skills, direct services, local relationships and jobs that take responsibility are safe in the near future, because they need a body, trust and accountability.
- In Bangladesh the bigger risk is not job cuts but fewer new junior posts and falling rates for online work.
- The price of making a draft has fallen and the price of catching mistakes has risen, so give part of the saved time back to verification and keep deep knowledge of your subject.
- Treat ads like "learn prompts and earn a lakh" and predictions tied to a particular year both with suspicion.