AI Agent

What is an AI Agent?

An AI Agent is a software system built on a large language model (LLM such as GPT-5, Claude, Gemini) that, unlike a simple chatbot, can autonomously plan steps, call external tools and iterate repeatedly until the goal set by the user is achieved. An agent is not just a “text generator” — it has a set of tools (API calls, database queries, document reading, sending emails) and decides for itself which to use and in what order.

A typical agent cycle:

1. Planning — the LLM breaks the task into a sequence of steps
2. Tool selection — decides which function to call
3. Execution and evaluation — processes the output and continues or revises the plan
4. Iteration — continues until the goal is achieved (or a limit is reached)

A practical scenario in a B2B company: an agent receives an email with a PDF invoice, OCR extracts the data from it, validates it against the purchase order, checks for duplicates, proposes a ledger entry and prepares payment — all without manual intervention.

When it is used

AI Agents are deployed for tasks that are:

• Repetitive — but require context and decision-making
• Structured — with a clear goal and output
• Cross-system — spanning multiple systems (email, ERP, CRM)

See the AI page, the Document Extraction module, and the Workflows module.

Related terms

• RAG — a technique that allows an agent to work with company documents. See /en/glossary/rag.
• MCP — protocol for standardised tool calling by an agent. See /en/glossary/mcp.
• Workflow automation — classic automation, from which AI agents differ by their autonomy. See /en/glossary/workflow-automation.

In Modulario

Modulario deploys AI agents in modules such as Document Extraction (autonomous invoice processing), for drafting replies in CRM, and for automating tasks in Workflows. Details on AI architecture are available at /ai.

AI agents in Modulario are always under human control — before irreversible actions (payment, sending an email to a client, deleting a document) the system requires confirmation from a responsible person. The entire agent workflow is audited and visible in a timeline — ideal for meeting the requirements of the AI Act.