In this paper, Shekhar and Saurombe (2026) built an entire study around a single post titled “Hey people who lost their jobs to AI, what happened?” They analyzed 1,454 Reddit comments, and what they found is a portrait of fear, grief, and dark humor.
The authors name the thing they’re studying “algorithmic anxiety.” They’re careful to say it’s not plain fear of unemployment. It’s a compound feeling that mixes economic dread with something deeper: the worry that human work itself is becoming obsolete, that the skills you spent fifteen years building can evaporate overnight. For students sitting in our classrooms right now, that’s the labor market we’re preparing them for.
The Finding That Reframes the Whole Study
The cleverest move in the paper is methodological, and it changed how I read everything else. Shekhar and Saurombe ran two sentiment tools over the same comments. VADER, which reads surface words, scored 52% of comments as positive. BERT, which reads context, scored 51% as negative. That reversal isn’t an error. The authors argue it reveals how workers mask real distress behind irony, a comment like “great, I’m free thanks to our AI overlords” that a shallow tool hears as cheerful and a contextual tool hears as a scream.
I find that genuinely useful. We spend a lot of energy measuring sentiment about AI through surveys and polls, and this is a reminder that the surface number can be the opposite of the truth underneath. People joke when they’re hurting. Any read of public feeling about AI that misses the joke misses the point.
Algorithmic Anxiety as a Breach of Trust
Where the paper earns its theoretical weight is in how Shekhar and Saurombe extend the idea of the psychological contract, the unwritten deal between worker and employer. When an algorithm makes the decision a manager used to make, they argue, you get new kinds of betrayal. They name three: technological betrayal, algorithmic abandonment, and digital dehumanization. A worker trained their own AI replacement and was kept on until the last day. A graphic designer watched their job shift from creating to correcting machine output.
The authors are sharp on how this damage accumulates. They describe a “thousand cuts” pattern, observing that “workers describe a ‘thousand cuts’ phenomenon where no single change seems breach-worthy, but cumulative impact devastates trust” (p. 12). No single automation step feels like the betrayal. The pile of them does. This connects directly to what Hartzog and Silbey (2025) argue about AI eroding institutions from the inside, where trust dissolves not in one dramatic act but through a slow draining of the human element.
What lands hardest is the authors’ read on where the stress actually comes from. They conclude that “the stressor is not technology itself but technology’s implications for human worth” (p. 12). The anger in the data wasn’t aimed at the models. It was aimed at executives and “shareholder capitalism,” at the human decisions behind the automation. That tracks with what Claessens and colleagues (2026) found about people’s negative perceptions of outsourcing work to AI: the resistance is moral, not technophobic.
Where I’d Be Careful
I want to be fair about what this study can and can’t say. The thread asked people who lost jobs to AI to come forward. By design, it pulled negative stories. Shekhar and Saurombe are upfront about this, repeating that they’re describing the phenomenology of algorithmic anxiety, not its prevalence. The 51% negativity is not a population statistic, and they say so plainly. Reddit also skews younger, more male, more Western, and more technologically literate than the workforce as a whole.
So this is a study of the wound, not the whole body. It tells us what AI displacement feels like from the inside for people who experienced it as a threat. It can’t tell us how common that experience is, or how many workers found AI genuinely augmenting. I’d hold the depth and the limit at the same time. Both are real.
Why a Workplace Study Belongs on an Education Blog
Here’s my real reason for covering a paper that never mentions schools. The students in our classrooms are walking into exactly this. If algorithmic anxiety is the emotional texture of the AI labor market, then AI literacy can’t only mean knowing how to prompt a chatbot. It has to include the human and economic dimension, the questions of dignity, identity, and what work means when machines do the tasks that once defined a profession.
I’ve argued before, reading Yee and colleagues’ (2025) work on AI fluency, that the durable skills are the human ones: judgment, collaboration, the things models can’t replicate. Shekhar and Saurombe’s data backs that up from the other direction. Their workers, even the bitter ones, kept affirming creativity, empathy, and ethical judgment as the qualities no algorithm touches. That’s not nostalgia. That’s a curriculum signal.
References
- Claessens, S., Veitch, P., & Everett, J. A. C. (2026). Negative perceptions of outsourcing to artificial intelligence. _Computers in Human Behavior_, 177, 108894. https://doi.org/10.1016/j.chb.2025.108894
- Hartzog, W., & Silbey, J. (2025). How AI destroys institutions [Draft]. Boston University School of Law. : https://scholarship.law.bu.edu/faculty_scholarship/4179
- Shekhar, A., & Saurombe, M. D. (2026). Algorithmic anxiety: AI, work, and the evolving psychological contract in digital discourse. Frontiers in Psychology, 17, Article 1745164. https://doi.org/10.3389/fpsyg.2026.1745164