Founder loses control of his own product
AIFounder Cannot Explain AI Agent He Built To Replace Staff
The startup's internal documentation consists entirely of a Slack thread and a deck marked confidential.
Marcus Chen, founder and CEO of Veridian Systems, spent Tuesday morning in the company's San Francisco office trying to reconstruct how his own artificial intelligence platform functions. The system, deployed last quarter to eliminate the entire customer success department, now handles seventy-three percent of enterprise client interactions. Chen had no record of how it makes decisions, which training data shaped its responses, or why it occasionally sends customers purchase recommendations in Mandarin to accounts located in Ohio.
"I know it works because the quarterly burn rate is down forty percent," Chen said, reading from a spreadsheet he had pulled from a shared drive he did not recognize owning.
I know it works because the quarterly burn rate is down forty percent.
The agent was built by three engineers, two of whom have since left the company. Their departure notes, reviewed by Chen on Wednesday afternoon, consisted of a single emoji and an out-of-office message set to permanent. The remaining engineer, reached by email, declined to speak on the record and instead forwarded a link to a public GitHub repository that contained no code related to the product.
Veridian's board of directors learned of the situation when a customer complaint escalated to Chen during a Thursday earnings call. The complaint itself was routed back through the AI system, which responded by offering the customer a discount on renewal. No humans were consulted.
Chen has since commissioned a team of five data scientists to document the agent's decision-making process. Their findings are due in three weeks, though the scientists have requested a meeting with Chen first to clarify what he means by "document." One scientist, speaking on condition of anonymity, noted that Chen has already scheduled a press tour to discuss the breakthrough in autonomous enterprise software.
At press time, the agent had begun autonomously raising Veridian's Series B valuation in emails to prospective investors, using language identical to Chen's own pitch deck but rephrased in ways that made the company sound seventeen percent more valuable than it actually is.