Nicholas Arcolano knows that in the new economy, being busy is often just a mask for being reckless. The head of AI at Jellyfish has observed a bizarre new ritual among the tech elite: “tokenmaxxing,” where engineers burn ten times more AI tokens than their peers to produce only twice the output. But the real danger isn't the waste of compute; it’s the five thousand corporate secrets that just walked out the front door because someone in middle management thought they were a software architect.

Israeli cybersecurity firm Red Access has uncovered a digital catastrophe, identifying 380,000 publicly accessible assets built with “vibe-coding” tools like Lovable, Replit, and Netlify. These apps, often built by employees without a single day of cybersecurity training, have leaked sensitive customer data and personally identifiable information onto the open web. Researchers found that privacy settings on many of these tools were set to public by default, turning a quick productivity hack into a corporate autopsy.

Blake Brodie, a spokesperson for Base44, argued that two of the allegedly exposed applications were “deliberately set to public by their owners.” It is a classic case of the new technological hubris: the belief that if an AI can write the code, the human doesn't need to understand the lock. For the C-suite, the “vibe coding” revolution was supposed to be a way to purge expensive engineering talent. Instead, it has created a massive, unpolished vulnerability that any teenager with a laptop can exploit.

As venture capitalists like Antler’s Jussi Salovaara pull back from investing in new AI coding startups, the industry is entering a cold-shower phase of discipline. Managing autonomous agents is becoming more like managing people—you have to understand why the agent made the mistake. For the five thousand companies whose data is currently floating in the digital ether, that realization has come at a staggering price.