Meta’s ambitious drive toward advanced artificial intelligence faces unexpected hurdles, according to comments made by CEO Mark Zuckerberg during a recent internal town hall meeting. Zuckerberg acknowledged that progress on AI agent development has been slower than anticipated, despite the company’s massive investment in AI research and infrastructure.
He emphasized that reaching “superintelligence” remains the company’s goal but insisted it will require sustained effort over time. Employees were told to expect some tangible AI benefits in the coming months, though the path forward demands balancing rapid innovation with workplace morale and trust.
This internal update reflects growing tensions within Meta’s AI strategy. The company has invested tens of billions in talent and technology to accelerate AI projects but is also grappling with the challenges that come with pushing teams aggressively. Recently, Meta reversed several controversial measures in response to employee concerns, signaling a recalibration of its approach.
One such reversal involves the AI training program that collected employees’ keystrokes and mouse movements to improve AI models. Initially mandatory, this program drew criticism and resistance after an internal leak exposed sensitive employee data. Meta has since paused the initiative and declared it will continue only on an opt-in basis if restarted. The chief technology officer admitted this rollout had negatively affected employee morale and trust, though it generated more useful data than expected.
Meta also eased demands on staff when it allowed engineers to leave the Applied AI task force—a unit where many had been reassigned amid the company’s push for AI advancements. This move came after warnings about some of the lowest morale levels in Meta’s history, following significant layoffs earlier this year.
These developments underscore the complex challenge Meta faces in advancing cutting-edge AI while maintaining workforce cohesion and addressing employee concerns about privacy and workload.