V
vchitai
Guest
AI is making businesses more productive than ever. I'm starting to wonder who will actually benefit from that productivity.
Most conversations about AI and jobs start with a layoff. I think the more important story is quieter. It's about the jobs that never get created, the knowledge that never gets passed on, and countries like Vietnam, whose way into the global economy was simply being needed.
Let me start with a made-up story. It's simplified on purpose.
Linh is 24. She graduated in IT two years ago and works as a junior developer at an outsourcing company in Quang Trung Software City, on the edge of Ho Chi Minh City. Her team of twelve builds and maintains a claims system for an American insurance company.
Her work isn't glamorous. She writes tests, fixes small bugs, and picks up the features nobody senior wants. Every pull request goes through a senior engineer who leaves more comments than she'd like. She's learning fast.
Her salary would look tiny in San Francisco. Here, it covers a shared room in Go Vap, the payments on her motorbike, and the money she sends home to her parents in Can Tho every month.
One quarter, the client rolls out AI coding tools to its own in-house engineers. Nobody at the vendor worries much. The client says the tools are there to make everyone faster.
Six months later, the contract comes up for renewal. The client asks for five engineers instead of twelve. Its in-house team, with AI, can cover the rest.
Nobody did anything wrong. The client cuts costs and ships just as fast. The AI vendor, most likely American, gains a paying customer. The outsourcing company stays profitable with a leaner team. On every dashboard, this is a success.
Linh isn't fired, exactly. She's moved to the bench to wait for a new project. A few months later, her company quietly cancels next year's fresher intake.
Part of the money that used to pay twelve salaries in Ho Chi Minh City now pays for a subscription in California. Linh's landlord, the cơm tấm stall outside the office, and her parents in Can Tho all feel it a little.
This isn't a small corner of the economy. Vietnam has around 530,000 IT engineers and more than 57,000 new IT graduates every year. Next door, the Philippine IT-BPM industry employs 1.82 million people. A lot of those jobs look like Linh's.
Not every AI rollout ends like this, and not every saved dollar goes to tech companies. Some gets reinvested. Some shows up as lower prices. But the mechanism is plausible enough that I can't stop thinking about it.
The most important part of Linh's story isn't Linh. It's the freshers who never got an offer letter.
When people talk about AI taking jobs, they picture termination notices. I think the bigger change is the job that never gets posted. A company that planned to hire 200 people next year decides 50 is enough. Nobody is laid off. There's no headline. 150 opportunities just don't exist.
In software, this is already the easy move. If five engineers with AI can ship what used to take ten, you don't need to fire anyone. You just stop opening roles.
The data is starting to show this shape. Stanford's Digital Economy Lab tracks US payroll data from ADP. Their August 2026 update finds no sign of widespread job displacement overall. But employment of 22-to-25-year-olds in AI-exposed occupations is now 19% below where it would be if it had kept pace with their less-exposed peers, up from 13% in the first version of the paper a year earlier.
Experienced workers show no comparable gap. And the gap comes mainly from companies hiring fewer young people, not from firing them.
Vietnam doesn't have data that clean yet, but the early signals point the same way. In the first half of 2026, IT job postings on Saramin rose just 2.58% year over year, while applications rose 21.3%. Postings asking for AI skills nearly doubled.
More people are lining up at a door that's barely getting wider. Nobody gets fired. There are just fewer ways in. For a country with a young workforce and a new wave of graduates every year, that matters more than any layoff.
For most of modern economic history, growing a business meant hiring people. For Vietnam, that link turned into an industry. Companies in Japan, the US, and Europe needed software built. Vietnam had engineers who were good and affordable. The work flowed here.
It wasn't a perfect deal. Outsourcing often meant long hours, thin margins, and little say in what got built. But it brought foreign income in and gave hundreds of thousands of young people a way into the global economy.
And the salary wasn't even the most valuable part. The experience was.
Most first jobs in software are full of small, unglamorous work. Fixing minor bugs. Writing tests. Maintaining code someone else wrote and nobody wants to touch. None of it makes it onto a slide. But that's where people learn how systems actually behave.
Here's the uncomfortable part: that unglamorous work is exactly what AI is best at.
The obvious worry is that juniors lose those tasks. I think the deeper problem is what happens to the learning that used to come with them.
Think about how Linh actually learns. Not from the tickets themselves, but from the friction around them. She gets a strange bug, spends an afternoon in the logs, guesses wrong twice, and finally asks her senior. He doesn't just fix it. He explains why the retry logic was built that way, which incident caused it, and what will break if she changes it.
None of that is written down anywhere. It lives in his head, and it moves to hers because the work forces them to talk.
That's how most real expertise in software travels. Not through documentation, but through people struggling with the same problems side by side.
AI shortcuts that loop from both ends.
A junior with an AI assistant can now produce working code without going through the struggle that used to build understanding. The output looks right. The tests pass. But the part that would have learned why it works never got exercised. She gets the result without the process.
A senior with an AI assistant no longer needs to hand the small task to a junior at all. It's faster to prompt than to explain. So the conversation where the knowledge used to move simply doesn't happen.
Nobody decides to stop training the next generation. It just becomes a little less convenient, every single day.
For a few years, nothing looks wrong. The seniors are still around, and they're more productive than ever. The problem shows up later, when they move on and the people replacing them can operate the tools but never built the judgment to tell when the tools are wrong. By then, there's nobody left to learn it from.
This matters even more in Vietnam, because the country's problem was never a shortage of juniors. A supervisor at a company in Quang Trung Software City put it bluntly to VnExpress: the market has few experienced developers and an excess of freshers.
The only known way to make a senior engineer is to let a junior do junior work, with someone experienced nearby, for a few years. If AI absorbs both the work and the conversations around it, the senior shortage doesn't get solved. It gets locked in.
The same shortcut shows up in the money.
Say a US company spends $5 million a year on an outsourcing contract in Vietnam. That money pays salaries, rents offices, and keeps local businesses running. Some of those engineers eventually leave to start companies of their own.
Now the company automates much of that work. It cuts the contract to $2 million and adds $1 million a year in AI services. The numbers are made up. The mechanism isn't.
The company saves $2 million. Its shareholders benefit. The AI provider earns $1 million. Vietnam loses $3 million a year in service exports, along with the jobs attached to them.
Outsourcing moved work from one worker to another. The income still landed in someone's pocket, just in a different country. A subscription is different. Part of the work stops being a labor requirement at all, and the value moves toward whoever owns the technology. Right now, those owners sit in a handful of wealthy economies.
The usual reply is that AI creates new jobs too: GPUs, data centers, power plants. True. But a data center isn't an employment engine the way an outsourcing office is. Running one takes a fairly small permanent staff, with different skills, often in a different country. From where Linh is sitting, those jobs might as well be on another planet.
I want to be fair to the optimistic case, because I partly believe it.
AI might be one of the best things that has ever happened to small businesses in Vietnam. A three-person team can now build what used to take twenty. A solo founder can sell to customers in the US without an office there. The leverage is real.
But access to a tool isn't ownership of the value it creates. A Vietnamese startup built on an American model still needs customers, distribution, capital, and data. As software gets cheaper to build, the ability to build it matters less, and the advantage shifts to whoever owns those things. They aren't spread evenly.
The same is true at the level of a country. For twenty years, Vietnam built its tech capability the slow way: millions of hours of real project work, which turned juniors into seniors, seniors into team leads, and team leads into founders.
If productivity now comes mainly from imported models, the country can get more output without accumulating the same depth of capability at home. Using a tool well is not the same as being able to build it, or even to judge it.
Step back, and the learning story and the economic story turn out to be the same story at different scales.
It used to work like this. To produce a result, a person had to go through the process of learning how. To grow, a company had to hire people and develop them. To build a tech services industry, a country had to accumulate experience through millions of hours of real work.
AI is loosening all three links at once.
A person can produce a result without deeply understanding how it was made. A company can grow its output without growing its headcount in proportion. And a country can raise productivity with imported technology without building the same depth of capability at home.
Each of these, on its own, looks like efficiency. Together, they close the main doors ordinary people used to walk through to take part in creating value: learning on the job, getting hired, and growing with an industry.
That's also why "just learn AI" isn't an answer. For an individual, it's good advice. But if every worker becomes twice as productive, companies don't need twice as many workers. When total demand for a kind of work drops, not everyone can win by being better than everyone else. That's the difference between a career strategy and an economic solution.
I don't know whether AI will reduce total employment. The evidence is still coming in, and cheaper services may open up markets we can't picture yet.
But I don't think the most worrying outcome is a world without jobs. It's a world that keeps getting richer, while the path for an ordinary person to take part in creating that wealth keeps getting narrower.
I'm not arguing against AI. Progress like this isn't something we can realistically reverse. I'm arguing against one assumption: that more productivity automatically means more shared prosperity.
It might. It often has. But the mechanism matters. When businesses grow by hiring and training people, the gains spread almost by default, as wages and as skills. When they grow by buying technology, how the gains spread depends on ownership, competition, and policy. And the skills don't spread at all unless someone decides to pass them on.
Linh will probably be fine. She's good, and she'll land on another project. I'm less sure about the 22-year-olds graduating in Ho Chi Minh City, Hanoi, and Da Nang next summer.
For twenty years, the industry needed people like them, and being needed is what turned them into engineers, team leads, and founders. Nobody set out to train them. The work simply required it.
That requirement is what's going away. If we want the next generation to have the same path, someone will have to build it on purpose.
The economy has never been better at growing without us. That's exactly why we should ask who it's growing for.
Most conversations about AI and jobs start with a layoff. I think the more important story is quieter. It's about the jobs that never get created, the knowledge that never gets passed on, and countries like Vietnam, whose way into the global economy was simply being needed.
Linh's Story
Let me start with a made-up story. It's simplified on purpose.
Linh is 24. She graduated in IT two years ago and works as a junior developer at an outsourcing company in Quang Trung Software City, on the edge of Ho Chi Minh City. Her team of twelve builds and maintains a claims system for an American insurance company.
Her work isn't glamorous. She writes tests, fixes small bugs, and picks up the features nobody senior wants. Every pull request goes through a senior engineer who leaves more comments than she'd like. She's learning fast.
Her salary would look tiny in San Francisco. Here, it covers a shared room in Go Vap, the payments on her motorbike, and the money she sends home to her parents in Can Tho every month.
One quarter, the client rolls out AI coding tools to its own in-house engineers. Nobody at the vendor worries much. The client says the tools are there to make everyone faster.
Six months later, the contract comes up for renewal. The client asks for five engineers instead of twelve. Its in-house team, with AI, can cover the rest.
Nobody did anything wrong. The client cuts costs and ships just as fast. The AI vendor, most likely American, gains a paying customer. The outsourcing company stays profitable with a leaner team. On every dashboard, this is a success.
Linh isn't fired, exactly. She's moved to the bench to wait for a new project. A few months later, her company quietly cancels next year's fresher intake.
Part of the money that used to pay twelve salaries in Ho Chi Minh City now pays for a subscription in California. Linh's landlord, the cơm tấm stall outside the office, and her parents in Can Tho all feel it a little.
This isn't a small corner of the economy. Vietnam has around 530,000 IT engineers and more than 57,000 new IT graduates every year. Next door, the Philippine IT-BPM industry employs 1.82 million people. A lot of those jobs look like Linh's.
Not every AI rollout ends like this, and not every saved dollar goes to tech companies. Some gets reinvested. Some shows up as lower prices. But the mechanism is plausible enough that I can't stop thinking about it.
Hiring, Not Firing
The most important part of Linh's story isn't Linh. It's the freshers who never got an offer letter.
When people talk about AI taking jobs, they picture termination notices. I think the bigger change is the job that never gets posted. A company that planned to hire 200 people next year decides 50 is enough. Nobody is laid off. There's no headline. 150 opportunities just don't exist.
In software, this is already the easy move. If five engineers with AI can ship what used to take ten, you don't need to fire anyone. You just stop opening roles.
The data is starting to show this shape. Stanford's Digital Economy Lab tracks US payroll data from ADP. Their August 2026 update finds no sign of widespread job displacement overall. But employment of 22-to-25-year-olds in AI-exposed occupations is now 19% below where it would be if it had kept pace with their less-exposed peers, up from 13% in the first version of the paper a year earlier.
Experienced workers show no comparable gap. And the gap comes mainly from companies hiring fewer young people, not from firing them.
Vietnam doesn't have data that clean yet, but the early signals point the same way. In the first half of 2026, IT job postings on Saramin rose just 2.58% year over year, while applications rose 21.3%. Postings asking for AI skills nearly doubled.
More people are lining up at a door that's barely getting wider. Nobody gets fired. There are just fewer ways in. For a country with a young workforce and a new wave of graduates every year, that matters more than any layoff.
How Juniors Become Seniors
For most of modern economic history, growing a business meant hiring people. For Vietnam, that link turned into an industry. Companies in Japan, the US, and Europe needed software built. Vietnam had engineers who were good and affordable. The work flowed here.
It wasn't a perfect deal. Outsourcing often meant long hours, thin margins, and little say in what got built. But it brought foreign income in and gave hundreds of thousands of young people a way into the global economy.
And the salary wasn't even the most valuable part. The experience was.
Most first jobs in software are full of small, unglamorous work. Fixing minor bugs. Writing tests. Maintaining code someone else wrote and nobody wants to touch. None of it makes it onto a slide. But that's where people learn how systems actually behave.
Here's the uncomfortable part: that unglamorous work is exactly what AI is best at.
Who Teaches the Next Generation?
The obvious worry is that juniors lose those tasks. I think the deeper problem is what happens to the learning that used to come with them.
Think about how Linh actually learns. Not from the tickets themselves, but from the friction around them. She gets a strange bug, spends an afternoon in the logs, guesses wrong twice, and finally asks her senior. He doesn't just fix it. He explains why the retry logic was built that way, which incident caused it, and what will break if she changes it.
None of that is written down anywhere. It lives in his head, and it moves to hers because the work forces them to talk.
That's how most real expertise in software travels. Not through documentation, but through people struggling with the same problems side by side.
AI shortcuts that loop from both ends.
A junior with an AI assistant can now produce working code without going through the struggle that used to build understanding. The output looks right. The tests pass. But the part that would have learned why it works never got exercised. She gets the result without the process.
A senior with an AI assistant no longer needs to hand the small task to a junior at all. It's faster to prompt than to explain. So the conversation where the knowledge used to move simply doesn't happen.
Nobody decides to stop training the next generation. It just becomes a little less convenient, every single day.
For a few years, nothing looks wrong. The seniors are still around, and they're more productive than ever. The problem shows up later, when they move on and the people replacing them can operate the tools but never built the judgment to tell when the tools are wrong. By then, there's nobody left to learn it from.
This matters even more in Vietnam, because the country's problem was never a shortage of juniors. A supervisor at a company in Quang Trung Software City put it bluntly to VnExpress: the market has few experienced developers and an excess of freshers.
The only known way to make a senior engineer is to let a junior do junior work, with someone experienced nearby, for a few years. If AI absorbs both the work and the conversations around it, the senior shortage doesn't get solved. It gets locked in.
Where the Money Goes
The same shortcut shows up in the money.
Say a US company spends $5 million a year on an outsourcing contract in Vietnam. That money pays salaries, rents offices, and keeps local businesses running. Some of those engineers eventually leave to start companies of their own.
Now the company automates much of that work. It cuts the contract to $2 million and adds $1 million a year in AI services. The numbers are made up. The mechanism isn't.
The company saves $2 million. Its shareholders benefit. The AI provider earns $1 million. Vietnam loses $3 million a year in service exports, along with the jobs attached to them.
Outsourcing moved work from one worker to another. The income still landed in someone's pocket, just in a different country. A subscription is different. Part of the work stops being a labor requirement at all, and the value moves toward whoever owns the technology. Right now, those owners sit in a handful of wealthy economies.
The usual reply is that AI creates new jobs too: GPUs, data centers, power plants. True. But a data center isn't an employment engine the way an outsourcing office is. Running one takes a fairly small permanent staff, with different skills, often in a different country. From where Linh is sitting, those jobs might as well be on another planet.
Using AI Is Not Owning It
I want to be fair to the optimistic case, because I partly believe it.
AI might be one of the best things that has ever happened to small businesses in Vietnam. A three-person team can now build what used to take twenty. A solo founder can sell to customers in the US without an office there. The leverage is real.
But access to a tool isn't ownership of the value it creates. A Vietnamese startup built on an American model still needs customers, distribution, capital, and data. As software gets cheaper to build, the ability to build it matters less, and the advantage shifts to whoever owns those things. They aren't spread evenly.
The same is true at the level of a country. For twenty years, Vietnam built its tech capability the slow way: millions of hours of real project work, which turned juniors into seniors, seniors into team leads, and team leads into founders.
If productivity now comes mainly from imported models, the country can get more output without accumulating the same depth of capability at home. Using a tool well is not the same as being able to build it, or even to judge it.
One Pattern, Three Levels
Step back, and the learning story and the economic story turn out to be the same story at different scales.
It used to work like this. To produce a result, a person had to go through the process of learning how. To grow, a company had to hire people and develop them. To build a tech services industry, a country had to accumulate experience through millions of hours of real work.
AI is loosening all three links at once.
A person can produce a result without deeply understanding how it was made. A company can grow its output without growing its headcount in proportion. And a country can raise productivity with imported technology without building the same depth of capability at home.
Each of these, on its own, looks like efficiency. Together, they close the main doors ordinary people used to walk through to take part in creating value: learning on the job, getting hired, and growing with an industry.
That's also why "just learn AI" isn't an answer. For an individual, it's good advice. But if every worker becomes twice as productive, companies don't need twice as many workers. When total demand for a kind of work drops, not everyone can win by being better than everyone else. That's the difference between a career strategy and an economic solution.
What Actually Worries Me
I don't know whether AI will reduce total employment. The evidence is still coming in, and cheaper services may open up markets we can't picture yet.
But I don't think the most worrying outcome is a world without jobs. It's a world that keeps getting richer, while the path for an ordinary person to take part in creating that wealth keeps getting narrower.
I'm not arguing against AI. Progress like this isn't something we can realistically reverse. I'm arguing against one assumption: that more productivity automatically means more shared prosperity.
It might. It often has. But the mechanism matters. When businesses grow by hiring and training people, the gains spread almost by default, as wages and as skills. When they grow by buying technology, how the gains spread depends on ownership, competition, and policy. And the skills don't spread at all unless someone decides to pass them on.
Linh will probably be fine. She's good, and she'll land on another project. I'm less sure about the 22-year-olds graduating in Ho Chi Minh City, Hanoi, and Da Nang next summer.
For twenty years, the industry needed people like them, and being needed is what turned them into engineers, team leads, and founders. Nobody set out to train them. The work simply required it.
That requirement is what's going away. If we want the next generation to have the same path, someone will have to build it on purpose.
The economy has never been better at growing without us. That's exactly why we should ask who it's growing for.