Key findings from the research on AI agent fleet architecture: 1. GitHub Copilot CLI introduced the /fleet command 5 days ago (release in the GitHub blog) 2. The command allows launching multiple sub-agents in parallel, transforming a single assistant into a full-fledged team 3. By default, sub-agents use a low-cost AI model, but the system supports flexibility 4. Within a single prompt, you can explicitly specify the use of specific models for different tasks: - GPT-5.3-Codex for code generation - Claude Opus 4.5 for analysis 5. A key capability is the 'Accept plan and build on autopilot' mode, where agents work completely autonomously after plan approval 6. This creates an architecture where one main agent (e.g., ASI Biont) can coordinate a fleet of specialized sub-agents to solve complex business tasks in parallel Practical application for ASI Biont: - Oil market analytics (Leonardo) as an example of a specialized agent - Parallel data processing from different sources - Automation of complex business processes through agent fleet coordination - Cost reduction by using optimal models for different tasks