AI agent
A system in which a language model decides its own next steps, calling tools and acting on the results in a loop until it reaches a goal or a stopping point.
An AI agent is a software system in which a language model decides what to do next, calls tools to act or to gather information, and repeats until it reaches its goal or a stopping condition.
Agents and workflows. Anthropic’s “Building effective agents” (December 2024) draws a useful line. In a workflow, code sets the path and the model fills in steps. In an agent, the model directs its own process and tool use. Anthropic calls both agentic systems. Microsoft Agent Framework’s documentation makes the same split, recommending an agent for open-ended tasks and a workflow for well-defined steps, and adds a practical rule: if a plain function can handle the task, write the function.
How one works. OpenAI’s Agents SDK describes an agent as a language model configured with instructions and tools, plus optional handoffs, guardrails and structured output. A runtime sends the conversation to the model, executes any tool calls it returns, feeds the results back, and repeats until the model gives a final answer or hands off to another agent. Anthropic stresses two habits of agents that work well: they check progress against real results from the environment, such as tool outputs, and they pause for a person at checkpoints or when they get stuck.
What an agent is made of. The model supplies judgment. Tools supply actions and fresh information. Memory keeps state beyond a single model call. Instructions and guardrails set the limits. Remove the tools and the loop, and what remains is a chatbot.
Agents across organizations. Inside one application, an agent reaches capabilities through tools, often over MCP. When the capability belongs to another organization’s agent, which keeps its own state, policies and goals, the two need an agent-to-agent protocol. The A2A specification describes such agents as independent and potentially opaque: they collaborate through declared capabilities and exchanged information without sharing their internal plans or tools.
Neighbouring terms. Agentic AI is the umbrella term for systems built this way. A multi-agent system splits work across several agents.