AI took my job. Here is what actually happened.
"AI took my job" is the version that travels. It fits on a thumbnail, it makes people click, and I made that thumbnail myself.
Here is the version that is true. A French MedTech with a shrinking subscriber base ran a second wave of licenciements économiques — redundancies for economic reasons, a specific legal category here, where the company has to prove the cuts are about its finances and not about you. A list had to be drawn up. I asked to be on it.
AI did not take my job. AI made it impossible to keep pretending the job was still interesting.
Those are different sentences, and the second one is the one worth eleven hundred words.
What the job actually was
I was the mobile engineer on a connected medical device: a bracelet that relieves chronic pain, osteoarthritis among others, paired with an app on iOS and Android.
That last sentence sounds ordinary and it is not. A medical device app is not a normal app. It lives under IEC 62304 for software lifecycle and ISO 13485 for quality management, it carries a CE marking, and every consequence of that lands on whoever ships it. You do not "just push a fix". Traceability, documented risk analysis, change control — and at the end of the chain, an actual person wearing the thing, in actual pain.
I owned development, maintenance and release for both platforms.
At the start it was the best job I could have asked for. Real constraints, real users, a product that does something to a body rather than to a funnel. Nobody has to invent meaning for that job.
Then everyone my age left
The first redundancy wave took the colleagues I had grown with. Not a reorganisation — an evacuation of a generation. I was twenty-four and I came out of it as the single person responsible for the mobile applications of a medical device.
Written down, that reads like a promotion. Lived, it is something else. A job stops being a craft and becomes a duty when there is nobody left to argue with about it. No one to review a tricky decision, no one to disagree with an architecture, no one who cares as much as you about whether a screen is right. You are not senior. You are alone, which is a different thing that happens to look similar on a CV.
The codebase I could not fix
The iOS app was legacy Swift built on SnapKit, laying out views in code, on an MVI architecture.
I want to be fair to MVI, because the pattern is not the villain. It is coherent, it is predictable, and on a large team it is genuinely good: everyone knows where an intent goes, where the state lives, what reduces into what. The structure is the point.
Solo, it is overkill, and the cost compounds. Shipping one small feature meant opening ten thousand files' worth of ceremony — an intent, a reducer, a state, a view, the wiring between all four, spread across a navigation layer I had inherited and could not simplify. The pattern that protects a team from each other protects a solo developer from nothing, and charges full price for it.
I wanted to migrate to SwiftUI. I never did, and the reason matters for the rest of this article: at the time, it would have meant rebuilding every screen by hand. This was the early days of AI assistance, before it could carry that kind of work. The migration was correct and unaffordable, so it stayed on a list.
That is the state I was in. Responsible for everything, able to improve nothing.
What AI actually changed
Early 2025, everyone around me got on Claude, me included. And the honest result, the one this whole article exists to say, is this:
the job got more boring, not less.
The part of that job that was genuinely hard was the detective work. A customer calls support: the bracelet stopped mid-session. Somewhere between the hardware, the Bluetooth service, the app's state machine and the user's actual behaviour, there is a cause. Finding it took days and felt like being good at something. I worked hand in hand with the support team, because they had the only ground truth that mattered — what the person on the phone actually did.
With Claude, correlating a customer report to a Bluetooth service to a piece of app state collapsed from days into an afternoon.
That is a real gain and I would never give it back. But look at what it left. Remove the hard investigation from a job where you cannot modernise the codebase, cannot ship ambitious features, and have nobody to build with, and what remains is maintenance. Competent, well-documented, regulated maintenance.
AI did not replace me. It removed the last part of the job that needed me awake.
A tool that makes the interesting 20% of your work trivial does not free you up. It exposes exactly how much of the other 80% you were tolerating because the 20% was worth it.
I asked to be on the list
So when the company's numbers kept going the wrong way and a new wave became unavoidable, I put myself forward.
In France that is not a dramatic gesture. A licenciement économique comes with a notice period, severance, and unemployment insurance calculated on your previous salary — roughly a year of stable income if you do not waste it. Volunteering for a list that has to exist anyway means one colleague who wanted to stay gets to stay, and I get a runway I could not have bought.
It is the best professional decision I have made.
What a year of runway actually bought
Not rest. Three things.
A freelance project for a ski resort near where I live, in Savoie — which put me back in front of a client, a deadline and a scope, the three things a salaried job had been quietly deciding for me.
My own products. Luteal, built for one person and downloaded twenty thousand times in three months. Lexlane, which failed in ways I learned more from. Both of them exist because there was finally time to be wrong on my own budget.
And preparation. I have already been through Google's interview loops once. I intend to go back, this time without a day job quietly consuming the part of my brain the process needs. That is its own article and I will write it.
What I would not give back
I am not going to pretend those years were wasted. Being the only mobile engineer on a regulated medical device taught me things no comfortable team would have.
Total ownership of production bugs, with nowhere to pass them. The habit of working directly with support instead of treating tickets as noise, because they held the only reliable account of what users did. What IEC 62304 and ISO 13485 actually require once you stop treating them as paperwork. And the knowledge, which I now carry into everything, that software attached to a human body is a different discipline from software attached to a metric.
The job was not taken from me. It quietly finished, some time before I admitted it, and the tooling of 2025 is simply what made the ending legible.