BOLAA: Benchmarking and Orchestrating LLM-augmented Autonomous Agents