The cursor was flickering with a rhythmic, taunting consistency over the Henderson project metadata. I was deep in the weeds, tagging the 44th set of architectural bridge joints for our latest generative model, trying to ignore the way the fading Friday light hit the dust motes in my home office. It was exactly 5:04 PM. My fingers were still hovering over the keyboard, mid-sentence in a Slack reply to my lead, Sarah, explaining why the reinforcement learning feedback from the 124th batch was showing a bias toward Victorian-era structural logic. I hit enter. The message didn’t send. A small, gray circle appeared next to my words-the universal symbol for a connection that had been severed before it could be completed.

At first, I thought it was just my Wi-Fi. It’s a 44-year-old house, and the router likes to throw tantrums whenever I’m processing large datasets. But then my phone, sitting face-up on the desk, lit up with a notification that didn’t come from Slack. It was an automated email from Workday. The subject line was as sterile as a surgical suite: ‘Important Information Regarding Your Separation from the Company.’ My heart didn’t just sink; it felt like it had been physically uninstalled. I’d spent 64 months at this company. I had curated 10,004 individual datasets. I had been on the Henderson project for 24 weeks, working until 8:04 PM most nights to ensure our training data was as authentic as the human experience it was meant to mimic.

And just like that, at 5:04 PM on a Friday, I was no longer an ‘AI training data curator.’ I was a ‘separated unit.’

[The Algorithm of Human Disposal]

There is a specific kind of cruelty in the timing. Choosing Friday at 5:04 PM isn’t a mistake; it’s a calculation. It’s the ultimate expression of ‘efficiency.’ By doing it then, the organization ensures that the ‘impacted individuals’ cannot immediately call HR, cannot disrupt the workflow of the remaining staff, and have an entire weekend to process the trauma in isolation before the markets open on Monday. It’s a way of cleaning the house while everyone is already walking out the door.

I remember thinking about my office setup. I had spent $444 of my own money on a high-back ergonomic chair and another $204 on lighting because, as a curator, I spend 14 hours a day looking for nuances in pixels. I even bought an Air Purifier Radar because the air in this windowless corner gets heavy when the GPU fans start screaming and the dust starts dancing. Now, looking at that purifier, I realized the irony. It was filtering out the impurities in the air, but I was the impurity being filtered out of the corporate organism. The organization didn’t want the messiness of a face-to-face conversation. They didn’t want the ‘noise’ of a six-year veteran asking why the Henderson project was being prioritized over people’s mortgages.

Companies often talk about ‘culture’ and ‘family,’ but culture is what happens when things go wrong, not when the quarterly earnings are up. When an organization chooses a Friday evening mass-deletion of access, they are admitting that their ‘values’ are just metadata-labels applied to a product to make it more sellable, but containing no actual weight. They treat layoffs as administrative events to be executed with the same precision as a server migration. But a server doesn’t have a mortgage. A server doesn’t have a child who needs 14 new pairs of socks for the school year. A server doesn’t feel the sudden, cold vacuum of being told their six years of loyalty was worth a 504 gateway error on their login screen.

I’m a person of strong opinions, and one of them is that we are currently over-optimizing for the wrong metrics. We measure ‘efficiency’ in terms of how quickly we can remove a cost, but we never measure the ‘cost’ of a broken spirit. I’ll admit, I’ve made my share of mistakes. Just last week, I accidentally joined a high-level strategy video call with my camera on while I was wearing a stained hoodie and eating a bowl of cereal at 10:04 AM. I was mortified. I watched 44 executives blink at me in silence before I could fumble for the ‘stop video’ button. It was a moment of raw, unpolished humanity. I felt exposed. But at least in that moment, I was seen. This Friday exit felt like the opposite-it was the act of being made invisible by design.

Actually, maybe I’m being too harsh. There’s a part of me-the part that curates logic for a living-that understands the cold math. If you have to cut 1,204 people, doing it via a mass-automation is ‘cleaner’ for the balance sheet. It prevents the 44-minute-long crying sessions in a glass-walled conference room. It stops the ‘contagion’ of grief. But that’s exactly the problem. Grief is a human metric. If you’re not willing to sit in the room with the grief of the people who built your Henderson project, you don’t deserve the success that project might eventually bring.

The contradiction of my own life is that while I hate this system, I’m the one who spent 84 months teaching the machines how to navigate it. I taught them what a ‘happy’ face looks like versus a ‘satisfied’ face. I taught them to recognize the difference between ‘urgent’ and ‘important.’ And yet, the machine that fired me couldn’t tell the difference between ‘terminating a process’ and ‘destroying a livelihood.’ It’s the ultimate failure of AI training: we’ve taught machines to mimic us, but we’ve taught corporations to act like machines.

I sat there for 44 minutes after the email arrived, just watching the purifiers’ light change from blue to amber. It was sensing something in the air-maybe just the sudden spike in my own cortisol. I thought about the 64 colleagues I’d had lunch with, the 444 threads we’d started on Slack about the best way to categorize ‘ambiguous’ data, and the 14 times I’d stayed up until dawn to meet a deadline. All of that data was now behind a wall I could no longer scale. My ‘separation’ wasn’t just from a paycheck; it was from a narrative I had helped write.

We believe in treating consumers with respect by being honest about products. We demand that labels are accurate, that ingredients are listed, and that claims are backed by data. Why don’t we demand the same from our organizations? If a company’s ‘ingredient list’ includes the potential for a 5:04 PM Friday deletion, they should have to list that on their career page. ‘Warning: This role contains 0% job security and may be terminated via a Workday notification if the efficiency algorithm demands it.’

I spent 34 minutes trying to find a way to contact Sarah, my lead. I wanted to tell her about the 124th batch-that the bias wasn’t in the data, it was in the way we were asking the question. But I realized she was probably staring at the same ‘access revoked’ screen I was. Or maybe she was the one who had to click the final button that sent the email, her own camera turned off so nobody could see the 44 tears she wasn’t supposed to shed in a professional environment.

In the end, the Henderson project will continue. Someone else-perhaps a junior curator making 34% less than I did-will pick up the metadata tags for the bridge joints. They will see my notes and wonder why I stopped at batch 44. They won’t know that I was in the middle of a thought when the world went dark. They won’t know that for 64 months, I cared about the integrity of those bridges more than the people who owned the company did.

And that’s the real tragedy. We are a society of bridge-builders being managed by people who only care about the toll booth tolls.

I stood up and walked to the window. The sun was almost down. I had 24 unread texts from friends who had heard the news through the grapevine. I didn’t answer them yet. I just watched the air purifier hum in the corner, doing its job, filtering out the invisible particles, making the room a little more bearable while I figured out how to breathe in a world that had suddenly become very thin.