Is translation actually a dying career, or is that overstated?

It is overstated as "dying" and understated as "repriced". The profession is not disappearing: certified and sworn translation, court and conference interpreting, patent and medical work, literary translation and transcreation are all real and will stay real, because somebody must be accountable for the accuracy or because voice is the product. What has genuinely gone is the large commodity middle, the competent general per-word work that made up most of the jobs. If you are in the surviving tier, you are fine. If you are being paid per word to clean up machine output at a rate that falls every year, the trend line has been in one direction for a long time and it is not going to turn.

What do translators actually bring to software that other switchers do not?

Two things, and the second one is rare enough to be worth money. First, you hold two formal systems in your head at once and notice the exact point where meaning fails to carry between them, which is most of what debugging is. Second, you have been professionally reviewing machine output for years. Every other career switcher is learning to use AI for the first time and learning, painfully, that it is confidently wrong. You already know that in your bones. The habit of trusting nothing until you have checked it is the single most valuable thing a new developer can bring, and it is the habit that takes everyone else the longest to build.

Do I have to abandon languages entirely?

No, and the people who do usually leave value on the table. Localisation engineering is a real discipline that sits exactly between the two: managing string extraction, translation pipelines, locale handling, right-to-left layouts and the tooling that moves content between systems. It is chronically understaffed because it needs someone who understands both sides, and almost nobody does. There is also a growing amount of language-adjacent AI work, evaluating model output, building and grading translation quality systems, designing the review workflows, where subject expertise is the scarce half rather than the code.

How long does the switch take, and can I do it while still translating?

Realistically 6 to 12 months of consistent part-time study to a first offer, and yes, freelance translation is one of the better jobs to do this alongside, because you control your own schedule in a way salaried workers do not. The honest requirement is 10 to 15 protected hours a week held for most of a year, plus two or three genuinely deployed projects. Build them from your own world: a terminology manager, a translation-memory tool, a QA checker for machine output. Projects that come from real working knowledge interview far better than another generic to-do app.

What does the pay difference actually look like in Malaysia?

Freelance translation income here is highly variable and has been compressing, particularly on per-word general work where you are effectively bidding against machine output. The developer destination enters at roughly RM 6,000 to RM 9,000 a month at the AI-capable junior level per our State of AI Hiring report, with RM 3,500 to RM 6,500 typical at smaller local employers and RM 7,500 to RM 9,000 at multinationals. The more important difference is not the starting number but the direction: one of these has a rate that falls as the models improve, and the other has a rate that rises with them.

All posts

Career switch · From translation · Malaysia

Translator to tech: the switch that needs the least defending.

Every other career-switch guide has to first convince you that AI is coming for your job. You already know. You have watched your rate per word fall for years while the machine output you are paid to clean up kept getting better. So this guide skips the persuading and goes straight to the part nobody writes for translators: what actually transfers, which lanes are open, and why you arrive with an advantage most switchers spend a year acquiring.

Deric Yee Updated 7 September 2026 9 min read

A shelf of well-used reference books - the craft that machine translation repriced

Say the honest part first

We have written role-by-role assessments for twenty-one professions, and translation is the one where the answer came out closest to yes. Not for all of it, but for the large commodity middle, which was most of the jobs. The full assessment is in will AI replace translators, but the short version is that machine translation became good enough for everyday commercial, technical and web text, and the per-word market repriced accordingly. What survives is the end where somebody carries legal or reputational liability, and the end where voice is the product: certified and sworn work, court and conference interpreting, patent and medical translation, literary work and transcreation.

If you are in that surviving tier, this guide is optional reading. If you are doing general translation or post-editing machine output at a rate that has been falling every year, read the trend line honestly. Post-editing pays a fraction of the equivalent translation rate, and the better the machine gets, the less you are paid for the same document. That is not a career direction. It is income while you build something else, which is a perfectly good thing for it to be, as long as you are actually building something else.

What translation actually trained

Translators consistently undersell themselves as “good with languages”, which is like describing a structural engineer as good with shapes. The actual training is more specific and more transferable than that.

You hold two formal systems in your head at once and move meaning between them under constraints, which is a structurally similar problem to most of programming. More usefully, you have spent years noticing the exact point where meaning fails to carry. That is what debugging is: not knowing the answer, but locating precisely where the thing stopped being true. Most beginners are terrible at this and it takes them months. You have been doing it professionally.

You are precise about ambiguity. Translators are trained to notice that a sentence has three possible readings and to resolve which one is meant before committing. Requirements documents, API contracts and bug reports are all full of exactly that problem, and the industry is full of people who charge ahead and build the wrong reading.

And the one that matters most: you have been professionally reviewing AI output for longer than almost anyone in the industry. Every other career switcher is discovering for the first time that a language model is confidently, plausibly wrong, and learning the hard way not to trust fluent output. You learned that years ago, for money. In a market where the failure mode of new developers is trusting generated code, arriving with a professional review reflex already installed is a genuine head start.

Three lanes, not one

Lane one: localisation engineering. The discipline that sits between the two worlds. String extraction, translation pipelines, locale and encoding handling, right-to-left layouts, pluralisation rules, and the tooling that moves content between systems without breaking it. It is chronically understaffed for one reason: it needs somebody who genuinely understands both sides, and developers who understand localisation properly are rare. This is the lane where your existing career is a qualification rather than a thing you left behind.

Lane two: language-adjacent AI work. Somebody has to evaluate model output, design the quality frameworks, build the grading systems and run the review workflows. That work is growing and it is bottlenecked on subject expertise, not on code. A translator who can also build is close to the ideal candidate, and it is worth reading what jobs AI creates alongside this, because this category barely existed three years ago.

Lane three: the full switch. Straight into general software development, where the language background becomes a personality trait rather than a job requirement. This is the highest-ceiling option and the one with the most open roles. Pay enters at roughly RM 6,000 to RM 9,000 a month at the AI-capable junior level (the data), with the full bands by employer type in the Malaysian salary guide.

You do not have to pick now, and picking now would be a mistake. All three lanes need the same foundation, and you will choose between them with much better information six months in than you can today.

The plan, and the one thing to build

Six to twelve months of consistent part-time study, 10 to 15 protected hours a week, alongside whatever translation work you are still doing. Freelancers have a real advantage here: you control your own calendar in a way salaried workers do not, and the scheduling problem that defeats most people is covered in detail here.

The portfolio advice is where translators get an unfair advantage, so use it. Do not build a generic to-do app. Build the tools your own trade needed and never got. A terminology manager that actually handles context. A translation-memory tool that does not fight you. A QA checker that flags the specific ways machine output goes wrong, which you can enumerate from memory and a developer cannot. These projects interview extraordinarily well, because they demonstrate two things at once: that you can build, and that you know a domain well enough to have opinions about it. That combination is what employers are short of.

Underneath every lane is the same foundation: being able to structure a problem, build, debug, ship and operate real software, with AI fluency and the judgement to catch what the model got wrong. That last clause is the part you already have. None of that learning is wasted whichever lane you end up in, which is exactly why it is worth starting before you have decided. The zero-cost way to find out whether it suits you is the free trial: a week of real projects, a live instructor session, no card.

FAQ

Is translation actually a dying career, or is that overstated?
It is overstated as "dying" and understated as "repriced". The profession is not disappearing: certified and sworn translation, court and conference interpreting, patent and medical work, literary translation and transcreation are all real and will stay real, because somebody must be accountable for the accuracy or because voice is the product. What has genuinely gone is the large commodity middle, the competent general per-word work that made up most of the jobs. If you are in the surviving tier, you are fine. If you are being paid per word to clean up machine output at a rate that falls every year, the trend line has been in one direction for a long time and it is not going to turn.
What do translators actually bring to software that other switchers do not?
Two things, and the second one is rare enough to be worth money. First, you hold two formal systems in your head at once and notice the exact point where meaning fails to carry between them, which is most of what debugging is. Second, you have been professionally reviewing machine output for years. Every other career switcher is learning to use AI for the first time and learning, painfully, that it is confidently wrong. You already know that in your bones. The habit of trusting nothing until you have checked it is the single most valuable thing a new developer can bring, and it is the habit that takes everyone else the longest to build.
Do I have to abandon languages entirely?
No, and the people who do usually leave value on the table. Localisation engineering is a real discipline that sits exactly between the two: managing string extraction, translation pipelines, locale handling, right-to-left layouts and the tooling that moves content between systems. It is chronically understaffed because it needs someone who understands both sides, and almost nobody does. There is also a growing amount of language-adjacent AI work, evaluating model output, building and grading translation quality systems, designing the review workflows, where subject expertise is the scarce half rather than the code.
How long does the switch take, and can I do it while still translating?
Realistically 6 to 12 months of consistent part-time study to a first offer, and yes, freelance translation is one of the better jobs to do this alongside, because you control your own schedule in a way salaried workers do not. The honest requirement is 10 to 15 protected hours a week held for most of a year, plus two or three genuinely deployed projects. Build them from your own world: a terminology manager, a translation-memory tool, a QA checker for machine output. Projects that come from real working knowledge interview far better than another generic to-do app.
What does the pay difference actually look like in Malaysia?
Freelance translation income here is highly variable and has been compressing, particularly on per-word general work where you are effectively bidding against machine output. The developer destination enters at roughly RM 6,000 to RM 9,000 a month at the AI-capable junior level per our State of AI Hiring report, with RM 3,500 to RM 6,500 typical at smaller local employers and RM 7,500 to RM 9,000 at multinationals. The more important difference is not the starting number but the direction: one of these has a rate that falls as the models improve, and the other has a rate that rises with them.

Back to all posts The full role assessment →

You have been reviewing AI output for years.
That is the habit everyone else takes a year to learn.

Six real projects in six days, a live instructor session, no card. The week that tells you whether building is the lane you want, using the one advantage you already have.

Start the free trial See the curriculum