An AI agent is an AI system that does more than respond to a single prompt. Agents can plan sequences of actions, call external tools and APIs, read and write data, make decisions based on intermediate results, and work toward a goal across multiple steps, often without human input at each stage.
The distinction from a standard language model is one of agency. A standard LLM responds. An agent acts. It might search the web, run a database query, send an email, create a calendar event, or update a CRM record, choosing which tools to use based on the task and what it has learned from prior steps in the same session.
Agent architectures vary widely. Some agents use a single model with tool access. Others use multi-agent systems where specialized agents collaborate, with one agent orchestrating and others executing specific tasks. Frameworks like LangChain, LlamaIndex, and Salesforce Agentforce provide infrastructure for building and deploying agents in enterprise contexts.
The governance challenges of AI agents are more significant than those of standard AI assistants, because agents take real actions with real consequences. An agent with write access to a CRM, email system, or financial database can cause irreversible harm if it misinterprets an instruction or encounters an edge case it was not designed to handle. Defining clear action boundaries, logging all agent actions, and maintaining human-in-the-loop requirements for high-stakes decisions are essential components of a responsible agent deployment.