Introduction
In 2025, I founded ANDOOR alone. Zero employees. But the scale of projects exceeded the limits of “one person.” So I chose to build AI agents as “team members.”
Why an “AI Team”?
Outsourcing and freelancers were on the table. But I prioritized AI team building for these reasons:
- Responsiveness: Available instantly — late at night, on weekends, whenever needed
- Context Retention: Once a project context is learned, there’s zero handoff loss
- Scalability: Parallel task capacity can scale up or down on demand
- Cost Structure: Variable cost, not fixed overhead
The Build Process
Phase 1: Single Agent Era
Started with simple back-and-forth with Claude. Prompt engineering yielded decent quality, but context fragmentation became the bottleneck.
Phase 2: Specialization & Role Assignment
Defined “characters” and “domains” for each agent, introducing role division. Research, implementation, review — each with a dedicated agent.
Phase 3: Autonomous Team
By integrating with task management systems, agents began autonomously picking up tasks, reporting completion, and moving to the next item. I (the human) could focus on direction-setting and review.
Current Setup
- Strategy & Direction: Human (me)
- Research & Analysis: AI agents
- Implementation & Coding: AI agents
- Documentation & Communication: Human + AI collaboration
Conclusion
An AI team is not a silver bullet. But designed correctly, it can reliably elevate one person’s output to team-level capacity. The key is clearly defining what AI handles and concentrating human effort on judgment that only humans can make.