Glossary · AI
AI agent
Also called: LLM agent, autonomous agent, tool-using AI
An AI agent is a system in which a language model decides its own next step, calls tools (APIs, searches, other systems), reads the results and repeats until it judges the goal reached.
The defining feature is control flow. In an ordinary workflow, you decide the order of steps. In an agent, the model does: given a goal and a set of tools, it picks a tool, looks at what came back, and picks again. Anthropic's engineering team draws the line the same way in Building effective agents: workflows follow predefined code paths, agents direct their own process.
An example
Goal: "Find out why order 10482 hasn't shipped and draft a reply to the customer." The agent calls the order API, sees status on hold, calls the payments API, sees a failed charge, and drafts a reply asking the customer to update their card. Nobody wrote "if on hold, check payments": the model chose that path.
Where agents fail
- Cost and latency. Each loop is another model call. A task that takes one call in a workflow can take eight in an agent.
- Unpredictability. The same input can produce different paths, which makes testing and auditing harder.
- Tool misuse. An agent with a "send email" tool can send an email you didn't intend, especially under prompt injection.
When to use one
Use an agent when the path genuinely can't be known in advance. If you can write the steps down, an agentic workflow or a plain workflow with AI steps is cheaper, faster and easier to trust. Either way, put a person in front of irreversible actions.
In Fluxo Flow
The Agent node is an AI step that can reason over several steps and call an HTTP tool, running on your own key from any of 14 providers. Most production workflows pair it with deterministic steps: a Validator to check its output, a Guardrail for banned terms, and a Human Review node before it sends or changes anything. Each server run stops if its estimated AI cost passes a per-run cap, and the trace shows tokens and estimated cost per step.
Read next
Related terms
- Agentic workflowAn agentic workflow is a workflow with a fixed overall structure in which some steps use a language model to read, classify, decide between predefined branches or draft content.
- LLMAn LLM, or large language model, is a neural network trained on large amounts of text that generates text by repeatedly predicting the next token, which lets it summarize, classify, extract and draft.
- AI guardrailsAI guardrails are automated checks placed before or after a model call that block, correct or route away inputs and outputs that break your rules, such as banned content, leaked data or invalid formats.
- Human in the loop (HITL)Human in the loop (HITL) is a design in which an automated process pauses at chosen points for a person to review, correct, approve or reject its output before it continues.