arXiv survey argues agentic AI needs trajectory validation beyond component tests

An arXiv cs.AI survey titled Beyond Component Testing: Validating Agentic AI Systems synthesizes 257 papers on agent evaluation, software assurance, cyber-physical systems, runtime monitoring, and regulatory guidance to frame validation for multi-step agentic systems. The abstract states agentic AI systems act through multi-step trajectories that combine planning, tool use, memory, interaction, and adaptation, stretching practice beyond component testing and one-shot input-output checks. The review uses a five-dimension taxonomy covering behavioral, safety, temporal, regulatory, and multi-agent concerns to map approaches and coverage gaps. Behavioral evaluation is described as comparatively mature, while temporal validity, runtime evidence maintenance, regulatory legibility, and open-ended multi-agent assurance remain under-developed. Three cross-domain case studies in medical care, industrial operations, and smart-mobility systems illustrate how the taxonomy dimensions recur in safety-critical settings. The paper closes with a lifecycle-oriented research agenda centered on bounded-autonomy specifications, adversarial trajectory generation, runtime monitoring, and audit-ready evidence structures. Its central claim is that trustworthy deployment depends on validating trajectories in context rather than assessing isolated components alone.