Tech Executives Deploy AI Clones for Work

The use of executive replicas extends beyond holographic displays. Axios reported in June 2025 that executives were using platforms such as Delphi and Tavus to create text, voice, and video agents from their writing, interviews, speeches, and meetings. The report named LinkedIn co-founder Reid Hoffman, Zoom CEO Eric Yuan, and Khosla Ventures managing director Keith Rabois among executives who had created AI replicas.

Axios also reported that Otter CEO Sam Liang had asked whether a "Sam Bot" could attend meetings for him. The company's Meeting Avatar was a prototype at the time, according to Axios, with a proposed capability to answer questions, participate in discussions, record meetings, and produce notes and summaries.

Fast Company Middle East separately reported that Meta was building a photorealistic, AI-powered 3D avatar of CEO Mark Zuckerberg for employee interaction, citing the Financial Times. The publication reported that the avatar would be trained on Zuckerberg's image, mannerisms, tone, speaking style, and public statements. It also noted that Meta had not responded to its request for comment.

The available reporting documents practical benefits, but it also identifies unresolved reliability and trust questions. Axios described an interaction with Delphi co-founder and CEO Dara Ladjevardian's chatbot in which the bot said chats were inaccessible to Ladjevardian, while Ladjevardian told Axios that he could access them and attributed the response to a probable bug.

That example matters because executive avatars combine several high-risk properties in one interface: an authoritative identity, potentially sensitive source material, and generative responses that may be delivered in a familiar voice or visual likeness. In comparable enterprise deployments, clear disclosure of whether an interaction is with a human or an agent, defined access controls, and auditability of agent outputs are common governance concerns.

The workforce dimension is also unsettled. Axios reported that some executive clones have been used for investor calls, employee discussions about company culture, and video meetings. When digital doubles move from scripted content into live conversation, organizations commonly face harder questions about who is accountable for an inaccurate statement, what information the system can access, and how employees should treat its responses.

For ML and product teams, the reporting points to a distinction between a polished avatar layer and dependable agent behavior. A realistic face, voice, and multilingual delivery can improve reach, but those features do not establish factual accuracy, privacy guarantees, or authority to act. The strongest early evaluation criteria are therefore likely to be task-bounded performance, traceable source grounding, permissioning, and explicit human escalation paths.