PM hiring and an AI testing cautionary tale
A remote Principal Growth Product Manager role at Fyxer was posted this week, outlining metrics‑driven growth responsibilities, while a circulated CEO case study described a QA team fired after relying solely on AI testing—an approach that saved $1.2M but produced a $6M loss from a hallucinated discount. Those items together underscore hiring for metrics fluency and the limits of replacing human QA with AI. (x.com) (x.com)
Fyxer posted a Principal Growth Product Manager role this week as the company pushes for “$100M+ by EOY 2026” and asks for a product leader who can turn experiments into revenue. (jobs.ashbyhq.com) The listing says the job is based in Chancery Lane, London, on a hybrid schedule, not fully remote, and says the hire will own growth strategy through research, data, and experimentation. (jobs.ashbyhq.com) Fyxer’s own site says its software organizes inboxes, drafts replies, takes meeting notes, and schedules meetings, with professional plans priced at $37.50 per user a month when billed annually. (fyxer.com 1) (fyxer.com 2) The company says it is selling relief from administrative work, citing 5.6 hours a week lost to email admin, and says it now serves thousands of professionals. (fyxer.com) Outside material tied to Fyxer’s growth push points to a culture built around rapid testing: GrowthBook said last week that a team of four growth engineers ran 360 experiments in a year as Fyxer went from $1 million to $35 million in annual recurring revenue. (blog.growthbook.io) That emphasis on measurement landed next to a separate cautionary tale circulating online about replacing human quality assurance with artificial intelligence testing, including an unverified claim that a hallucinated discount caused a $6 million loss after $1.2 million in payroll savings. The specific case study in the cited social posts could not be independently verified from primary sources available on April 12, 2026. (x.com 1) (x.com 2) The underlying risk is well documented even without that anecdote. IBM says hallucinations are false outputs from foundation models that can mislead users and damage downstream systems when the output is treated as fact. (ibm.com) Stanford Human-Centered Artificial Intelligence reported in May 2024 that legal artificial intelligence models hallucinated in at least 1 out of 6 benchmarked queries, showing the problem persists even in tools built for narrow professional work. (hai.stanford.edu) The pairing of those two items is straightforward in practice: one side of the market is hiring product leaders to run faster experiments against revenue targets, while the other is still warning that automation without human checks can fail at basic validation. (jobs.ashbyhq.com) (ibm.com) As of Sunday, April 12, 2026, the confirmed facts are the Fyxer hiring brief, the company’s growth targets and pricing, and the broader record that artificial intelligence systems still invent facts often enough to make unattended testing a costly bet. (jobs.ashbyhq.com) (fyxer.com) (hai.stanford.edu)
