TATM #4: There will be a happy ending, by Julian Pintat
This article does something very exciting: it strikes a note of hope, and coming from someone like Julian Pintat, this means a lot. He has a wealth of experience in some of the translation industry’s most demanding sectors, and has watched his profession first get dismantled by AI and machine translation, and now, it seems, reassembled. Lots of us are wondering when the AI bubble will burst, but few are as well-positioned as Julian to hear it pop.
This article is an expansion on Julian’s contribution to the Brian Merchant’s Blood in the Machine newsletter back in August 2025, where we were also featured. The first act of this article was written in May 2025, and the second a few months later, in October. You can get in touch with Julian by email at j.pintat@pintat-fuchs.com, and we highly recommend subscribing to Brian’s newsletter as well.
This article is part of our ongoing Translators Against the Machine open call, where we call on fellow translators and other language professionals to tell us about their working lives. Our aim is to gather experiences and stories to paint a multifaceted, three-dimensional picture of what it’s really like to work in the translation industry. If you’re interested in submitting, you can read more about the open call here.
***
Act I: May 2025, Spring
I’ve been a technical translator for 15 years, self-employed all the way. I enjoy it, I am good at it.
I translate complicated, demanding material – mainly medical and pharmaceutical, like the UI and user guides for MRT imaging devices, or patient information and consent forms for clinical trials, or subtitles for a presentation on the side-effects of this or that new drug. I’ve translated documentation for the specialty filters you need in cooling loops for nuclear power plants and I’ve translated manuals for assembly systems for aircraft construction. I get to dive into obscure sub-specialties of technical fields and learn about stunning feats of engineering that few people have even heard of. It’s fun.
In a field where everybody seemed perpetually on the brink of starvation, I was able to make a good living. There were always ups and downs, but I managed to clear six figures in the good years and didn’t have to worry too much in the bad years. I worked long hours, I worked a lot of weekends, but I felt it all balanced out.
2025 has been, not to put too fine a point on it, absolute shit so far. Entire months went by with zero work. And the requests that are now coming in – almost all “PED”, post-editing.
Post-editing is when a client runs their text through machine translation, and then has it reviewed and edited by a human. It’s been around forever – since way before the current AI hype. It pays a quarter of what you’d get for translation work, and if you do it properly, it takes you just as long as translating from scratch.
So I would summarily reject PED requests. I’d take one or two per year just to take a look at the current state of the art, and invariably found, happily, that machine translation was still awful and I was going to be fine.
As of today, I’ve earned maybe €8000 this year. Requests are 90% PED. Unrelated calamities have drained the vast majority of my savings, and there is a very real possibility I’ll end up in personal bankruptcy.
Machine translation hasn’t even improved with the advent of AI – contrary to the online hype, our industry never got its big OpenAI moment. I’m starting to suspect that we’ve been forced down this path by an unhappy coincidence of sunk costs and economic downturn.
A little while ago, I started learning to code – I needed something to do after all.
ChatGPT and Claude started off as amazingly helpful tools.
Then at some point you’ve got the basics down and you’re trying to do marginally more complex things – and you notice how quickly they lose track and fall apart, how needlessly complicated their solutions are, how your entire architecture turns into a mess of barely-functional spaghetti.
Does this stuff work *anywhere*? My IT friends complain about being forced to use whatever hot new AI tool, and their companies stopped hiring junior positions. My own industry seems broken. After sending this mail, I’ll have to do some tedious, underpaid post-editing. I’ll hate it. Whoever will have to actually use the documents will hate it.
I believe this will pass. Sooner or later, the AI companies will have to stop losing money and adjust their pricing. And then it’ll turn out that using AI for everything gets you worse results than humans, at the same cost. And that will be that.
I just hope I can hang on until then.
Act II: November 2025, Autumn
In May, I wrote the piece above for Brian Merchant’s Substack, where it was published in August 2025. An interview with Time Magazine followed, and I ended up holding the dubious honor of having my name associated with LLM-induced white-collar misery.
I went back to work, what little work there was. AI was being forced into ever more tasks, being ever more useless: now there were requests to review AI proofreaders, who would inevitably ignore their own glossary and contravene their own style guides. A complete non-starter, if you take a second to think about it. If the AI was capable of finding and fixing errors with any semblance of consistency, you would just let it run multiple passes over its own output without any need for human intervention.
Then, among breathless pronouncements of “PhD-level abilities”, whispers of AGI and prophecies of doom, OpenAI released GPT-5. The most anticipated piece of software of the year, the decade even, fell flat on its face. The only difference users found was that they didn’t like it. The apex had been crossed.
And then, as everything AI started to slide down the long downwards slope of the hype cycle, my job came back.
A trickle, at first; then a bit more, and then the dam burst and I was working weekends again.
Real work. There were still PED requests coming in, but now they were negotiable. “No,” I could say, “the MT is awful. This needs to be billed as translation”, and more often than not, it was. Clients reappeared, many of them, no doubt, among the 95% of companies seeing no measurable return on AI investment. In the last two months, I’ve earned more money than in the entire first half of the year.
It’s still not what it used to be, but the recovery has begun. And I don’t think it’s just me. What about the other profession that was first in line to be consumed by the machine?
“Wow, this is really impressive,” you thought back in spring, as you watched hundreds of lines of code scroll by. So impressive, in fact, that it took a while to notice you could’ve written it faster from scratch – just like machine translation. And here, as well, eventually people figured it out, and the false economy was revealed. In the meantime, OpenAI is starting to engage in the kind of financial maneuvering that seems primarily designed to dress up the balance sheet and keep the party going for just a bit longer.
When I wrote the piece in May, I dreaded the work, the days of empty inboxes, the weeks without phone calls. Spring and summer were all silence and decay. This one, I’m writing sandwiched between two blocks of work, and another request has just come in. It’s Sunday. The music is back.
I told Time Magazine it would take 12-18 months for the hype to blow over. Just 8 weeks later, you can hear it, beneath the music, that noise of tortured steel, as the edifice they told us would be the great and terrible Machine God begins to sag under its own ridiculous weight. There’s a few ways this could play out; a drawn-out whimper as the entire thing slumps and settles; a hellish bang when a load-bearing member finally snaps. In either case, keep your ears open. It won’t be long now.
Produced under a Peer Production License
– Written by Julian Pintat
– Edited by Guerrilla Media Collective
– Lead image by Public Domain Review