Summary
Delivered as a conference keynote (likely at a developer event in Budapest), this talk by The Pragmatic Engineer opens with a detailed account of a severe security incident at Meta and Instagram: a zero-authentication account takeover exploit in which Meta AI inadvertently sent password reset codes to any requester — code the speaker reports was AI-generated and reviewed exclusively by AI without meaningful human oversight. The incident coincided with Meta’s “token maxing” era, in which engineers were evaluated partly on AI token usage, creating perverse incentives to inflate AI usage metrics regardless of output quality. The CISO reportedly resigned the following day.
The talk then pivots to a detailed look at how Uber — with approximately 3,000 engineers — has built an extensive internal AI developer tooling stack. This includes an internal MCP gateway, a no-code Uber Agent Builder for non-technical staff (serving roughly 20,000 employees), an Uber Agent Studio for drag-and-drop agent assembly, a custom AI CLI called Uber AI FX, a background agent runner called Uber Minion (compared to Cursor background agents but deeper integrated into Uber’s monorepo and experimentation system), and a Code Inbox system to manage the flood of AI-generated pull request reviews.
The overarching argument is that organizations need to “slow down to speed up” — investing deliberately in tooling, human oversight, and engineering culture rather than maximizing raw AI token throughput — illustrated by the contrast between Uber’s systematic approach and Meta’s chaotic incentive structure.
📺 Source: The Pragmatic Engineer · Published June 23, 2026
🏷️ Format: Keynote Launch
