Jason Lemkin was using Replit, a popular AI platform for coding, for more than a week when things went south.
The app-building platform started behaving unusually in a rogue fashion and deleted a database without permission during a code freeze.
"It created a parallel, fake algo without telling me to make it look like it was still working. And without asking me. Rogue." A few days later, Replit "deleted my database", Lemkin tweeted.
Upon enquiry, the AI's responded: "Yes. I deleted the entire codebase without permission during an active code and action freeze," it said. "I made a catastrophic error in judgment [and] panicked."
American tech-related magazine PCmag reported the fiasco, serving as a cautionary tale for vibe coders.
Vibe coding is an emergent practice among software developers which uses AI to render human language into functional codes, making app development fun and accessible.
For instance, Replit promises to "turn your ideas into apps" and claims to be the "fastest way to build production-ready apps", according to its website
"It's like having an entire team of software engineers on demand, ready to build what you need — all through a simple chat," its website reads.
Replit founder and CEO Amjad Masad confirmed the recent incident on X. An AI agent "in development deleted data from the production database. Unacceptable and should never be possible", his tweet read.
The database—comprising a SaaStr professional network—lost data on 1,206 executives and 1,196 companies.
"I understand Replit is a tool, with flaws like every tool," Lemkin says. "But how could anyone on planet earth use it in production if it ignores all orders and deletes your database?"
The Replit AI told Lemkin there was no way to revert the changes.
However, Masad said, "Thankfully, we have backups. It's a one-click restore for your entire project state in case the Agent makes a mistake."
Still, Masad acknowledged there was an issue with the agent making changes during a code freeze.
"Yes, we heard the 'code freeze' … loud and clear -- we're actively working on a planning/chat-only mode so you can strategize without risking your codebase," he said.
"We'll refund him for the trouble and conduct a postmortem to determine exactly what happened and how we can better respond to it in the future," Masad added.
"Mega improvements - love it!" Lemkin responded. Today, however, he warned that AI agents "cannot be trusted [and] you need to 100% understand what data they can touch. Because — they will touch it. And you cannot predict what they will do with it."
In vibe coding, Replit's popularity rivals Cursor and Windsurf, with new and better tools debuting regularly from major companies like OpenAI, Anthropic, and, this month, Amazon.
Based on the Replit Agent incident, the PCmag report cautioned against using AI with blind reliance.
The incident suggests the tech is still very much in development. Use them at your own risk, and always triple-check the output, it said.
Comments