Single AI agents are constrained by the context window length and single-inference capability of large language models, and often struggle when faced with large codebases.