This paper aims to examine where large language models (LLMs) and AI agents create the most value across the broader HR cycle – from workforce planning and attraction to offboarding – and where human oversight remains essential in higher-stakes decisions.
The paper is a selective narrative review of recent academic literature on AI-enabled HRM, with emphasis on studies published between 2021 and 2025 that describe concrete applications of LLMs or agentic systems across employee-lifecycle stages and related HR functions. The review synthesises evidence for a practitioner audience and translates it into a decision framework for HR leaders.
Three patterns stand out. Firstly, LLMs are most useful where HR work is knowledge intensive and language heavy, such as onboarding, learning support, employee self-service and document-rich compensation queries. Secondly, AI agents add value when HR work requires orchestration across multiple steps, tools or rules, particularly in recruitment workflows, guided onboarding and case handling. Thirdly, a full-cycle view shows uneven evidence: recruitment, onboarding and learning are the most developed domains, whereas workforce planning, engagement and retention and offboarding remain promising but less mature. Across all stages, the main barriers to responsible adoption are fairness, privacy, explainability, data quality and employee trust. High-stakes decisions such as selection, pay and performance evaluation therefore require human review, documented controls and clear appeal paths.
The paper moves beyond broad “AI in HR” claims by distinguishing between LLMs and AI agents, comparing their value across the HR cycle and offering practical guidance on where augmentation is more appropriate than automation. It also extends the discussion beyond the usual focus on recruitment and onboarding by mapping emerging use cases across the full employee lifecycle.
