news tech single source: Ars Technica

The Ghost in the Machine's Workflow

Meta’s grand vision for becoming an "AI native" powerhouse—a roadmap that promised to trim teams by as much as sixty percent—didn't quite materialize. Project OT, launched by Mark Zuckerberg earlier this year, was built on the premise of letting AI agents shoulder the bulk of thousands of employees' daily grind. The goal was clear: workflow automation, AI-first building, and streamlining everything.

In January, the plan moved forward. By May, the first wave of layoffs hit. Yet, just as quickly, a planned second round vanished.

We don't have a definitive answer for why Zuckerberg shifted course; perhaps it was simply a lack of certainty regarding whether increased AI usage would deliver the expected productivity surge.

The internal documentation painted a picture of deep integration—agents interacting constantly, middle management potentially sidelined by "agent-assisted analysis." Even promotion decisions flirted with shared oversight between HR staff and these burgeoning systems, though Meta insisted those final judgments remained resolutely human territory.

But here is where the theory met grit. Instead of smooth efficiency gains flowing outward to users, internal reports revealed something messier: massive spikes in technical and security incidents stemming from these autonomous agents making "large-scale, disruptive actions that humans are unlikely to execute." Resolving these AI-induced messes consumed significantly more employee time than anticipated.

Code changed internally at an astonishing rate—up 220% year-over-year—yet features hitting actual users barely budged at 36%.

It serves as a potent case study: the chasm separating theoretical automation potential from chaotic real-world execution. The pursuit of total systemic takeover stumbled not because the technology failed outright, but because its messy implementation proved profoundly unstable when confronted with organizational gravity and unexpected errors.