Type: url
Original: https://arxiv.org/pdf/2508.11126
Status: generated
Summary
This paper is a systematic literature review of 152 academic papers (2022-2025) surveying AI agentic programming, an emerging paradigm in which LLM-based coding agents autonomously plan, execute, and interact with tools such as compilers, debuggers, and version control systems. It introduces a taxonomy of agent behavioral dimensions and system architecture categories, reviews enabling techniques such as prompting strategies, tool integration, and context/memory management, and surveys coding benchmarks used to evaluate agents. It also compares AI agentic programming to five related automation paradigms plus robotics/reinforcement-learning agents, and discusses open challenges (evaluation and benchmarking, multi-agent communication protocols, domain-specific models, safety and privacy, human-centric toolchain design, and scalable memory) and future research directions.