AI agent

#How AI agents work

An agent runs a loop. The model receives a goal and a list of tools. It chooses a tool and its arguments, the application runs the tool, and the result goes back into the model's context. The loop repeats until the model decides the goal is met or a limit stops it.

Anthropic's guide separates workflows from agents. In a workflow, code sets the path, and the model is called at predefined steps. In an agent, the model decides its own steps and its own tool use.

#Why it matters for AI agents

An agent is only as reliable as its tools. A model can plan well and still fail when a tool returns wrong data. Web actions are often the weakest link, because websites change and some block automated access. Each step also costs model calls, which adds time and money.

#Wield and AI agents

Wield is not an agent. It is the layer agents call. Your agent, for example one in Claude Desktop, decides what to do. Wield gives it tools such as find_recipe and run, which execute a site task in your own browser. No model runs inside a recipe, so a recipe run costs no LLM tokens. See how Wield works.

#Related terms

• Tool calling: the mechanism an agent uses to act.
• Browser agent: an agent that works through a web browser.
• MCP server: a common way to offer tools to an agent.

#FAQ

#What is the difference between an agent and a workflow?

A workflow follows steps that code defines. An agent lets the model choose its next step. Anthropic's guide advises using the simplest design that works and adding agent behavior only where it is needed.

#Does an AI agent need a browser?

No. An agent needs tools. It might use a browser, an API, a local file or an MCP server. A browser agent is one kind of agent, and it works through a browser. Wield lets an agent reach some sites through recipes instead of driving a browser step by step.