Something Big Is Happening — Probably
In February 2025, a blog post by Matt Shumer, co-founder of OthersideAI, reached 50 million views within just a few days. Shumer described how his job had effectively "disappeared" because AI now performs it better than he does. He tells the AI what he wants, walks away for four hours, comes back, and finds the work completed. "And it's usually perfect."
Fifty million people read it. Millions shared it. The real question is not simply whether Shumer is right. It is why so many people were eager to believe him.
What the Post Claims
The piece is filled with singularity undertones. "AI building the next generation of AI." "50 million citizens smarter than any Nobel laureate." "Jobs disappearing within 1–5 years."
The core message is straightforward: if you still do not understand what is happening, you are like those who, in February 2020, believed the toilet paper panic was an overreaction. The comparison is deliberate. It is designed to trigger FOMO.
The Problem: No Data
Gary Marcus, AI researcher and one of the industry's most persistent skeptics, was blunt. He described the post as "weaponized hype that tells people what they want to hear, but stumbles on the facts."
He has a point.
Shumer claims that the latest coding systems can write "whole complex apps without making errors," yet provides no examples, no metrics, and no error rates. He also misrepresents the METR benchmark, stating that AI models complete tasks that take "five hours," without clarifying that the success threshold is 50%, not 100%. In other words, they succeed half the time. Moreover, the benchmark applies strictly to coding tasks, not to general tasks as the post implies.
There are also critical voices he does not mention at all.
Kelsey Piper, generally known for her AI-friendly stance, wrote about her experience with Claude Code: "Sometimes, Claude is absolutely the worst coworker you've ever had." In one instance, it deleted all the audio files she had collected for an app, files she had received explicit permission from a professor to use, and replaced them with AI-generated sounds that were "all subtly wrong."
A recent TechCrunch report also highlights that many developers report burnout from continuous interaction with AI tools, with only moderate productivity gains relative to the cognitive cost. Additionally, automatically generated code from large language models presents significant security risks, an issue the original post entirely overlooks.
AI Hype as an Apocalyptic Narrative
What is most interesting is not that Shumer exaggerates. It is that 50 million people wanted to believe him.
There is something almost religious in how we talk about AI today. The singularity, the moment when technology surpasses human intelligence, has become the new Second Coming. Like any apocalyptic vision, it comes with prophets, heretics, and believers waiting for the end with a mix of fear and excitement.
Shumer did not just write a blog post. He crafted an apocalyptic narrative: "I have seen the future. The future is now. If you fail to see it, you will be left behind."
The structure is almost evangelical. A personal revelation. A warning. A call to repentance. A promise of salvation for those who adapt.
It works. Not necessarily because it is true, but because we want to believe we are living through historic moments. It is far more compelling than saying, "Technology is improving incrementally, and reality is complex."
What This Means for Us
As always, the truth is more nuanced and far more interesting than the hype.
Yes, AI models are improving. Yes, they can now perform tasks that seemed impossible two years ago. Yes, some roles will change dramatically. But "change" does not mean "disappear within five years with absolute certainty." And "improving" does not mean "perfect."
The real risk of AI hype is not optimism. It is complacency. When you believe the tool is flawless, you stop auditing it. You stop asking the right questions. In a professional environment, that carries a real cost.
The alternative to hype is not reflexive skepticism. It is disciplined critical thinking: what does AI do well today, where does it perform adequately, where does it fail, and what are its boundaries?
That is the most useful question to ask in 2026. Not "when will it replace us," but "where can I trust it, and where can I not?"
