Uber Engineers Built an AI Version of Their Boss

In a move that blurs the line between workplace automation and satire, Uber engineers have created an AI replica of their manager. The project, initially a joke, has sparked serious conversations about AI’s role in workplace hierarchy.

The AI boss can attend meetings via video call, respond to emails, and even approve time-off requests. It’s powered by a fine-tuned language model trained on thousands of Slack messages, meeting transcripts, and performance reviews.

How It Started

What began as a hackathon project quickly evolved into something more substantial. The engineering team trained the model on their manager’s communication patterns:

• Slack messages: 50,000+ messages over 2 years
• Meeting transcripts: 200+ hours of recorded meetings
• Email correspondence: 10,000+ emails
• Performance reviews: Feedback patterns and evaluation criteria

The result: an AI that sounds, writes, and decides remarkably like their actual boss.

What It Can Do

The AI boss handles several routine management tasks:

| Function | Capability | Accuracy |
|———-|————|———-|
| Meeting attendance | Joins via video, takes notes | 85% |
| Email responses | Replies to routine inquiries | 90% |
| Time-off approval | Approves based on team policies | 95% |
| Standup updates | Collects and summarizes progress | 88% |
| Performance feedback | Provides routine feedback | 75% |

For routine tasks, the AI performs comparably to a human manager.

The Manager’s Reaction

Surprisingly, the actual manager was amused rather than threatened. “I asked them to make it stricter than me on code review standards,” they joked. “If I’m going to be replaced by a bot, at least let it be a demanding one.”

The manager now uses the AI to handle routine administrative tasks, freeing up time for strategic work and one-on-one coaching.

The Broader Implications

This experiment touches on several workplace trends:

Middle Management Automation

Research suggests 40-60% of middle management tasks could be automated:

• Scheduling and coordination
• Status reporting
• Routine approvals
• Policy enforcement

But the human elements—coaching, conflict resolution, strategic thinking—remain difficult to automate.

Worker Agency in AI Deployment

The Uber case is unusual because workers built the AI themselves, rather than having it imposed by leadership. This bottom-up approach may lead to better adoption:

• Workers understand the tool’s limitations
• Implementation addresses actual pain points
• Trust is higher when peers build the solution

The Authenticity Question

Can an AI truly represent a person’s judgment? Critics argue that management involves contextual understanding and emotional intelligence that AI cannot replicate.

Supporters counter that much of management is pattern recognition and policy application—tasks well-suited to AI.

Key Takeaways

• Uber engineers built an AI replica of their manager using 50K+ Slack messages and 200+ hours of meetings
• Capabilities: Meeting attendance, email responses, time-off approval, standup updates
• Accuracy: 85-95% on routine tasks, 75% on performance feedback
• Manager reaction: Amused, now uses AI to handle administrative tasks
• Automation potential: 40-60% of middle management tasks could be automated
• Bottom-up approach: Worker-built AI may have better adoption than top-down mandates
• Remaining challenges: Coaching, conflict resolution, strategic thinking still require humans

The Bottom Line

The Uber AI boss experiment is both a technical achievement and a workplace Rorschach test. Optimists see a future where AI handles administrative drudgery, freeing managers to focus on genuinely human work. Pessimists see the beginning of management automation that could eliminate middle-layer jobs.

The truth likely lies in between. AI will transform management roles, automating routine tasks while elevating the importance of distinctly human skills: empathy, creativity, strategic thinking, and relationship-building.

The engineers who built this tool weren’t trying to replace their boss. They were asking a question: what parts of management are actually necessary, and what parts are just organizational friction? The answer may reshape how we think about work itself.

FAQ

How did Uber engineers build an AI version of their boss?

They trained a fine-tuned language model on their manager’s communication patterns: 50,000+ Slack messages, 200+ hours of meeting transcripts, and 10,000+ emails. The AI can attend meetings, respond to emails, and approve time-off requests.

What tasks can the AI boss handle?

The AI handles routine management tasks including meeting attendance (85% accuracy), email responses (90%), time-off approval (95%), standup updates (88%), and basic performance feedback (75%).

Is this replacing human managers?

Not entirely. The actual manager uses the AI to handle administrative tasks, freeing time for strategic work and coaching. While 40-60% of middle management tasks could be automated, human skills like empathy, conflict resolution, and strategic thinking remain difficult to automate.

—

Sources: TechCrunch, Uber Engineering Blog

Tags: Uber, AI Agents, Workplace Automation, Management, AI Ethics, Engineering Culture