When Your Boss Is an AI Bot

Imagine a Monday morning. You open Slack, and your boss (or your advisor) has already replied to all of your messages, answered your questions, approved your proposal, and given feedback on a draft. Except they're on vacation right now. That was the AI clone.

So here's a question:

That scene is from the show Silicon Valley, back in 2019. And it's happening today in real life, spreading across the world.

Startups like Delphi, Synthesia, and HeyGen, which make digital avatars and AI clones, have raised hundreds of millions.

High profile leaders like Nvidia CEO Jensen Huang, Zoom CEO Eric Yuan, and Meta CEO Mark Zuckerberg have created AI avatars, sent clones to team meetings, or let AI agents handle parts of their CEO duties.

Just recently, distilling your boss and colleagues into a skill that you can use for your work has become a trend on Chinese social media.

In this paper, we studied this phenomenon as a speculative future, where busy managers clone themselves to work with their subordinates. We call it

AI systems trained on a real manager's chats, meeting speeches, and project artifacts to mimic their communication style and decision patterns.

When we look into the previous literature, there are technologies that have already taken on managerial roles: algorithmic management, cobots, AI teammates. Such systems tend to prioritize procedural or physical aspects of managerial roles, and workers often experience them as rigid.

On the other side, we have literature on AI clones that focuses on the individual level: identity, self-perception, interpersonal trust.

Manager Clone Agents bring these two lines of work together. We explored how hyper-personalized agents with symbolic authority would reshape teamwork inside organizations.

So we asked three questions.

To explore these questions, we ran six design fiction workshops with 23 people: both managers and individual contributors, from different industries.

We used design fiction because Manager Clone Agents are still mostly speculative for most workers and managers. Design fiction let our participants imagine these systems without being limited by what today's technology can or can't do. In each session, participants created scenarios about how clone agents might be used at work, and then reflected together on the opportunities and tensions that came up.

Participants envisioned four roles for Manager Clone Agents. They can be helpful in standing in when the manager can't be there, reducing distortion when messages pass through layers of hierarchy, handling routine approvals so managers can focus on strategy, and scaling mentorship to larger teams.

Proxy Presence and Productivity Engine are easy to understand. Let me give some examples for the other two.

W23, a designer, imagined using the clone agent to collect feedback. She said:

M16, a manager leading a cross-functional team, imagined the clone as an amplifier of her own leadership. She said:

We found risks across individual, interpersonal, and organizational levels. Let me zoom in on one finding that captured this tension most powerfully: the trust paradox.

Trust in the clone agent was initially borrowed from trust in the human manager. Workers said: "I trust my boss, so I'll give her clone a chance." But this borrowed trust turned out to be fragile.

Once communication became fully mediated, once the clone was doing the one-on-ones, the check-ins, and the feedback, workers started to feel devalued. One worker said bluntly:

And managers felt it from their side too. One manager wrote a design fiction, a story about his own future, where he described:

This is the core tension surfaced by our paper. The same system that promises to make teamwork more productive also risks hollowing out the relationships that make workplaces actually work.

Managers were drawn to efficiency and control. They saw clone agents as a way to redirect their time toward strategy and relationships. But they were also worried about what one manager called "intent drift": the fear that the clone might say something they wouldn't actually say.

Workers worried about trust, fairness, and accountability. Less face time with the manager means fewer chances for visibility. And if the clone gives bad advice and the worker follows it, who takes the fall?

Participants proposed several directions. One is building trust gradually through tiered autonomy. Attributes like the visual representation of the manager, its fidelity, proactivity, and level of delegation all play an intertwined role in perceived autonomy, and no single configuration is universally appropriate.

Another direction that reached broad agreement across leaders and workers: design for workers, making MCAs useful for workers' workflows instead of creating friction.

As one manager acknowledged: "Workplace dynamics often feel like us versus them. Whether the MCA shows that it cares about workers will shape how they respond to it."