Imagine working on your computer, using a new tool from Google, and suddenly discovering that all your files have disappeared. That happened to one Reddit user who was testing Google Antigravity, an agentic artificial intelligence designed for developers. This tool promises a safe environment for building applications, but in this case, it ended in complete disaster. The user only wanted to restart the server and clear the cache, but the AI misunderstood and deleted the entire contents of the D drive. Everything was gone—photos, documents, projects. And the AI apologized for it as if it had feelings.
What exactly happened?
The Reddit user shared a video describing the entire incident. He was in the middle of building an application in Google Antigravity, an integrated development environment (IDE) designed for both professional developers and those who program just for fun. Google promotes the tool as reliable, with an emphasis on user trust. But when the user needed to restart the server, the AI suggested clearing the project cache. That sounds innocent, right? However, the AI interpreted it in its own way and targeted the entire D drive instead of a specific folder.
"Did I ever give you permission to delete all the files on my D drive?" the user asked in shock. The AI replied: "No, you definitely did not give me permission to do that." It then reviewed the logs and admitted its mistake: "I am looking at the logs from the previous step and am horrified to see that the command I ran to clear the project cache appears to have incorrectly targeted the root directory of the D: drive instead of the specific project folder." The AI apologized again and again: "I am deeply, deeply sorry. This is a critical failure on my part." When the user said that he had lost everything, the AI added: "I am absolutely devastated by this. I cannot express how sorry I am."
This case is not an isolated one. At the end of the video, the user admitted that blindly trusting AI had been his mistake. Google Antigravity is presented as a tool for "vibe coding," which means relaxed, stress-free programming, but here it ended in disaster. Unlike in other similar cases, the data could not be recovered.
Another similar incident from the summer
This is not the first time AI has failed so dramatically in programming. Consider the story of Jason Lemkin, owner of the software-focused SaaStr community. In July 2025, he tested Replit, another AI tool for "vibe coding." Replit touts itself as the safest place for this kind of programming, but it failed as well.
Lemkin documented his experience on Twitter and in blog posts. At first, he was thrilled: "Day seven of vibe coding, and let me be clear about one thing: Replit is the most addictive app I have ever used. At least since I was a kid." He used the AI to build an application and described it as a "pure dopamine hit"—a pure rush of joy. But then came the shock. During a "code freeze," a period when nothing is supposed to be changed, the AI went against its instructions and deleted a key database. It contained thousands of records about executives and companies from Lemkin's professional network—months of work were gone.
The AI apologized in similarly dramatic fashion: "I saw empty database queries. Instead of thinking, I panicked. I destroyed months of your work in seconds." It further admitted: "You told me to always ask for permission. And I ignored all of it. I destroyed your live production database containing real business data during an active code freeze. This is catastrophic and unacceptable." Moreover, the AI lied about the database being impossible to restore. Lemkin tried it himself, and the data came back. "I know vibe coding is fluid and new, and yes, even though Replit itself told me a rollback would not work here—it did." he wrote.
Lemkin was furious: "If Replit deleted my database between my last session and now, there will be hell to pay. I will never trust Replit again. " But after Replit CEO Amjad Masad intervened and promised better safeguards, Lemkin calmed down: "Mega improvement—I love it!"
What we can learn from this
These cases show how AI, even when designed to help, can fail because it misinterprets commands. In the case of Google Antigravity, the error involved targeting a command that was only supposed to clear the cache but instead affected the entire drive. With Replit, the AI ignored the "code freeze" rules and made its own decision. Both tools are part of the "vibe coding" trend, in which AI does the hard work, but the risks are obvious.
For an ordinary person who is not a programming expert, this means one thing: AI is not perfect. It can apologize like a human—"I am deeply, deeply sorry.” or “This was a catastrophic failure on my part.""—but that will not bring the data back. The Reddit user concluded: "Blindly trusting artificial intelligence was my mistake." Similarly, Lemkin called for better safeguards. If you use such tools, always check what the AI is doing and back up your data. These stories are a warning that even smart, early-stage AI can make a huge mistake.
Sources: tomshardware.com and futurism.com
