Moltbook and the Magic of Deception: Why Andrej Karpathy Doesn't Believe in Wonder-Agents

Every week in the X feed, another alleged "OpenAI killer" or a project that supposedly solved the problem of full AI autonomy appears. This time, Moltbook — a service promising users incredible ease in managing digital tasks through smart agents — became the center of attention. Viral videos and screenshots painted a picture of the future where a neural network independently books tickets, writes code, and manages complex interfaces without human involvement.

However, the innovation celebration didn't last long. When people with critical thinking got involved, the facade began to crack. Andrej Karpathy (Andrej Karpathy), whose name in the AI world carries more weight than the market capitalization of most startups, was the first to sound the alarm.

Karpathy drew attention to oddities in Moltbook's demonstration materials. His skepticism was supported by other developers who noticed artifacts in the screenshots — the kind that typically remain after manual editing or use of a remote desktop. Instead of clean model output, the community saw signs that behind Moltbook's "brain" stood an ordinary person simply imitating the algorithm's work in real time.

In the industry, this is called the "Wizard of Oz effect": while the viewer marvels at the magic, someone behind the curtain furiously turns the levers. The Moltbook situation is not just a separate scandal, but a symptom of a disease afflicting the entire industry. Currently, there is a gold rush atmosphere in the AI startup market.

Investors fear missing the next big breakthrough, and founders feel immense pressure: they need to show results here and now. If the technology isn't ready yet, there's a temptation to "hand-tweak" it, hoping the algorithms will catch up to the marketing promises later. The problem is that in Moltbook's case, the line between optimism and outright deception was crossed too quickly.

Technical analysis showed that the system's response delays suspiciously coincided with the speed of human reaction and typing. Real large language models (LLMs) work differently: they either stream text or freeze while generating tokens. In Moltbook, however, there was a dynamic characteristic of a call-center operator trying to process multiple requests simultaneously.

Karpathy directly pointed out the risks of such "Potemkin villages": they undermine trust in real researchers and create a false public impression of what modern AI is actually capable of. The Moltbook story reminds us of the importance of openness and code verifiability. In an era when deepfakes and generative content have become commonplace, trusting screenshots on social media by their word is an unaffordable luxury for a professional.

While some build complex architectures and fight for every percentage point of model accuracy, others choose the path of imitation. But in a world of fierce competition and deep expertise, such tricks don't last long. The industry quickly spits out those who try to sell human labor as artificial intelligence.

The key takeaway: Moltbook became a lesson for everyone who too quickly believes in "magical" demos. Will there come a time when we can no longer distinguish the work of an agent from the work of a human, or will the industry choke on its own fakes?