Meta built an AI clone of Mark Zuckerberg. Staff can now ask it questions and get answers that sound like the real CEO.
The system probably runs on Llama models with a memory system pulling from town hall transcripts, policy memos, internal documents. Training data includes thousands of hours of Zuckerberg's public appearances, from keynotes to congressional hearings, plus proprietary material like executive emails and strategy presentations. Meta's AudioBox technology likely handles voice synthesis.
This is where corporate AI gets strange.
Zuckerberg can't personally answer every employee question. Scaling executive communication through AI makes practical sense on paper. But staff querying a synthetic version of their boss instead of the real person? It's leadership on demand. Blurred lines between human direction and algorithmic output.
And it raises questions that feel new. When a bot trained on your CEO's voice tells you to prioritize something, do you treat that like a direct order? What happens when the AI-Zuck says something the real Zuck disagrees with? Who's accountable then, particularly given the dangers of reduced human scrutiny?
Meta isn't the first to experiment with AI personas internally. But this is the highest-profile example by far. If it works, expect other large companies to build similar tools.
The tech is ready. The real question isn't whether employees want this. It's whether leadership on demand still counts as leadership at all.