An AI agent is a software system that uses artificial intelligence — most commonly a large language model as its reasoning core — to autonomously perceive inputs from its environment, plan a course of action, execute that plan through available tools or APIs, and iterate based on feedback until a goal is achieved. Unlike a simple chatbot that responds to a single query, an AI agent maintains state across multiple steps, can invoke external tools (web search, code interpreters, databases, communication platforms), and may spawn sub-agents to parallelize complex tasks. The degree of autonomy ranges from "human-in-the-loop" designs requiring approval at key decision points to fully autonomous agents that operate without human intervention.
AI agents are increasingly deployed in enterprise settings for use cases including customer support automation, code generation and review, data analysis pipelines, research workflows, and IT operations. They offer significant productivity gains by automating multi-step knowledge work that previously required human judgment at every stage. However, this autonomy also introduces novel risks: agents can take unintended actions that are difficult to reverse, accumulate permissions beyond what is necessary for a given task, be manipulated through prompt injection attacks embedded in the data they process, and operate outside the visibility of IT and security teams when deployed as Shadow AI.
Enterprise governance of AI agents requires a purpose-built policy framework that addresses the unique characteristics of agentic systems. Key governance requirements include: minimal privilege access (agents should only have the permissions they need for each specific task), action audit logs (every tool call and state transition should be recorded), human escalation triggers (high-stakes or irreversible actions should require human approval), input monitoring for prompt injection (data the agent reads from external sources can contain malicious instructions), and approved agent registries (a formal process for evaluating and sanctioning AI agents before enterprise deployment). Regulatory frameworks including the EU AI Act are beginning to address autonomous agents as a distinct AI system category with heightened oversight requirements.