I was good at my job. For fifteen years, I worked as a freelance translator and interpreter, trilingual in French, English, and Spanish. I translated technical manuals for Syntegon, automotive documentation for Aston Martin, cloud platform guides for Google, industrial specifications for Siemens. I did simultaneous conference interpreting for the Wikimedia Foundation. I was fast, precise, and reliable. My clients came back. I earned a decent living.
Then the floor started moving.
How a career dies
Translation did not disappear overnight. It eroded. The first sign was rate compression: clients who had paid 0.12 euros per word started asking for 0.08, then 0.06. The justification was always some version of “we use machine translation now, we just need you to clean it up.” Post-editing, they called it. Same output expectations, half the pay, and a fundamentally different job. Instead of translating, you were proofreading a machine. The skill that had taken years to develop, the ability to reconstruct meaning across conceptual frameworks, was being bypassed.
The second sign was volume collapse. Not for all translation, but for the kinds that paid well: technical, IT, documentation. These are precisely the domains where large language models perform best because the language is structured, the terminology is consistent, and the context is bounded. Literary translation survived longer because it requires cultural judgment, but the market for literary translation was always tiny. The money was in technical work, and that is where the machines arrived first.
The part nobody talks about
By 2024, the trajectory was clear. But “clear” is a word that makes the process sound rational and controlled. It was not.
There is a specific kind of grief that comes with watching your profession become obsolete. It is not the same as being fired, because nobody fires you. There is no single moment of loss. Instead, there is a slow dawning, a period of months or years where you oscillate between denial and lucidity, where you tell yourself that the good clients will stay, that quality still matters, that machines cannot really do what you do. And then one day you open a file that was “pre-translated” by an LLM and you realize that it is 85 percent correct. Not perfect. Not elegant. But functional enough that the client no longer needs you to start from scratch. They need you to fix the last 15 percent, and they will pay you accordingly.
That moment is devastating, and it is devastating precisely because you cannot argue with it. The machine is not better than you. But it is good enough, and it is infinitely cheaper, and “good enough at lower cost” is how every market in history has worked. You spent a decade becoming excellent at something, and excellence turned out not to be the relevant variable.
What follows is a period of identity crisis that I think most people who have been through it do not talk about honestly. When your work is something you trained for, something you are genuinely skilled at, something that forms part of how you understand yourself, losing it is not just an economic event. It is a loss of self. You wake up and the thing that made you useful, the thing that justified your place in the professional world, is worth less today than it was yesterday, and it will be worth less tomorrow than it is today. The curve only goes one direction.
I went through about six months of what I can only describe as functional depression. I kept working. I kept delivering projects. But I knew, with a certainty that I could not argue away, that I was operating inside a shrinking box. The walls were moving inward and there was no mechanism to push them back.
The decision
Rather than competing with tools that would only get better, I chose to pivot toward a field where physical presence, hardware understanding, and human judgment cannot be automated away: industrial cybersecurity.
The decision was not random. I spent months analyzing which fields had structural barriers to automation, not just current ones but barriers rooted in physics. Three things made operational technology (OT) cybersecurity compelling.
First, the physical barrier. You cannot remotely audit a factory’s control systems with the same confidence that you can remotely audit an IT network. OT security requires physical presence: someone who can open an electrical cabinet, trace a wiring diagram, connect to a PLC via serial interface, and understand what they are looking at. This is not a job that AI can do from a data center.
Second, the regulatory pressure. The European Union’s NIS 2 directive, which came into force in October 2024, requires thousands of organizations across critical sectors to meet new cybersecurity standards. The compliance deadline is creating enormous demand for people who understand both IT security and industrial systems. The problem is that these people barely exist: IT security professionals rarely understand industrial protocols like Modbus or Profibus, and industrial engineers rarely think about cybersecurity. The intersection is almost empty.
Third, the linguistic advantage. Industrial cybersecurity in Europe is an inherently multilingual field. Standards are written in English (IEC 62443), regulations in the language of each member state, equipment manuals come from German, French, and American manufacturers, and the audit reports go to management teams who may speak none of these. A trilingual professional who can read a standard, debug a protocol, and write a report in three languages is not competing with fresh graduates. That person is competing with almost nobody.
The Mike Rowe question
Mike Rowe, the American television host who has spent two decades arguing that skilled trades are undervalued, recently warned of a “massive workforce shakeup” as millions of jobs are disrupted by AI. His core thesis is correct: there is a genuine skilled labor crisis, and the cultural bias toward white-collar office work has left critical infrastructure understaffed.
But Rowe’s framing is deeply American, and it does not translate directly to the French context. The 250,000 dollar electrician he cites is a data-center-boom-in-Texas story. French wage ceilings are lower, but so are the floors, and the social safety net changes the entire calculus. More importantly, France already has a mechanism for exactly what Rowe advocates: the alternance system, where students split their time between school and paid work in a company. It is precisely designed to bridge the gap between education and skilled employment, and it works better than the American system for most people.
The deeper issue with Rowe’s argument is the binary it creates: trades versus college. What I am building is neither. It is a hybrid technical profile that sits at the intersection of electronics, cybersecurity, and systems thinking, anchored by real projects (an interactive art installation built on microcontrollers, a self-hosted AI agent with a vector database and RAG architecture) that demonstrate the ability to design systems end-to-end.
What 37 actually means
Every career-change guide tells you that your age is an asset. Most of them are exaggerating. In most fields, a 37-year-old career changer is at a disadvantage against a 22-year-old with a fresh degree and no salary expectations.
But in regulated industrial environments, the dynamics reverse. Nuclear plants, pharmaceutical facilities, and energy infrastructure operate under strict safety cultures where maturity, precision, and the ability to communicate across hierarchies matter more than raw technical speed. A 37-year-old who has spent fifteen years working under tight deadlines with demanding international clients, who can read a standard in three languages, and who has demonstrated the discipline to retrain from scratch brings something that a 22-year-old with a pure computer science degree cannot: proof that they can function in a professional environment where mistakes have physical consequences.
The translation background is also more directly relevant than it appears. Auditing code and systems is, at its core, a form of reverse translation: you are reading something (a configuration, a network topology, a firmware binary) and determining whether the expressed implementation matches the intended specification. The skill of noticing the gap between what was meant and what was said is exactly what a good translator does, thousands of times a day, for years.
What comes next
The path I have chosen is not the fastest route to money and it is not the safest. What I am doing is harder: studying for a BTS while searching for an alternance placement, building projects to prove competence, and betting that the demand for OT cybersecurity professionals will continue growing faster than the supply.
Is it going to work? I do not know. But I know that staying where I was meant watching the value of my skills decline every year while pretending it was not happening. The grief of losing a profession is real, and I do not think it ever fully goes away. But there is a difference between mourning and paralysis.
At least now I am moving toward something that gets harder to automate, not easier. And the thing I lost, the ability to move between languages and frameworks and reconstruct meaning across them, turns out to be more transferable than I thought. It just needed a different surface to work on.
Adapt or die is not a philosophy. It is a description of what happens when you do not.
Written collaboratively with AI. See the blog page for more on my process.
Sources: EU NIS 2 Directive (2022/2555); IEC 62443 standard for industrial automation security; ANSSI (French National Cybersecurity Agency) qualification frameworks; Mike Rowe, Fox Business interview on workforce disruption (2025-2026); French Ministry of Labor data on alternance programs and salary scales for cybersecurity professionals.