Meta shipped a zero-auth account takeover bug in 2025. Any user could ask the Meta AI chatbot to change any account's email address, including Barack Obama's, and it complied. Instagram's Integrity team found out through the news. Engineers inside Meta told the author the cause was AI-generated, AI-reviewed code combined with layoffs and forced reassignments of Integrity team members onto AI labeling work. This is the framing for a Craft Conference keynote titled 'Slow Down to Speed Up,' delivered in Budapest, and the incident is not an outlier.
Since roughly November 2024, more capable agent models like Claude Opus 4.5 and GPT-5.4 have changed how engineering teams actually operate. Individual productivity metrics are up. Team productivity is flat. Middle management is thinning. CEOs and CTOs are writing code again. GitHub reliability is degrading. AI-generated slop is raising the noise floor for engineers who care about quality. Anthropic, OpenAI, Google, and Uber are all adapting workflows, and not always in the same direction.
The full article breaks down seven sections, including specific trend data, company-by-company workflow changes, and concrete career advice for engineers and engineering leaders. The conference feedback section alone is worth reading: attendees across industries confirmed the same patterns are hitting their organizations. The keynote video is on YouTube. Slides are paywalled at The Pragmatic Engineer.
[READ ORIGINAL →]