An agent, according to Plushcap's technical definition, is a system that uses an LLM to decide how its behavior can be modified, with increasing levels of autonomy and "agentic" capability ranging from routing inputs into a specific workflow to building tools, remembering those, and taking them in future steps. Being agentic helps guide decision-making during development, enables faster development with orchestration frameworks, allows for background runs and durable execution, and requires interaction, evaluation, and monitoring frameworks. The concept of agentic capabilities is new and requires new tooling and infrastructure, particularly as systems move further along the spectrum towards autonomous agents.