# AI agent

An AI system that can select steps towards a goal and carry out actions through available tools.

Source: https://aecon.ai/glossary/ai-agent
Updated: 2026-09-07

## In practice

A repair agent finds a booking incomplete, checks earlier visits, and chooses whether to request an equipment photo or ask the coordinator.

## The distinction

A model produces outputs; an agent combines that capability with information, tools and control over actions. It remains limited by its permissions.

## What makes it a working system?

Consider a constructed repair-booking example. “Prepare tomorrow’s visit” supplies a goal, but the model cannot complete that job from the instruction alone.

| Part | Its role in this example |
| --- | --- |
| Model | Interprets the booking and selects a useful next step |
| Information | The equipment description and authorised visit history |
| Tools | Functions for reading records or requesting a label photo |
| Permissions | The boundary between collecting information and committing to a repair |
| Feedback | Whether a record was found or the customer supplied the requested detail |

The system might find two conflicting model numbers. It should make that uncertainty visible and refer the case to the coordinator, not manufacture a confident answer. The physical inspection and repair remain with the technician.

## An agent is not an employee or an account

“Agent” is a broad technical term with a longer history than language models. This reference focuses on current AI systems used for work. An agent does not become a legal person, take responsibility for a business or acquire permission simply by having a connection to its systems.

## Judge the whole job

A demonstration that retrieves a record is useful evidence about that action. It does not establish that the visit was well prepared. Evaluate missing records, conflicting information and requests outside the system’s permission, as well as straightforward cases.

## Sources & context

- [Machine Learning Glossary](https://developers.google.com/machine-learning/glossary) — Google for Developers. Technical terminology, paraphrased for a business reader. Checked 2026-09-07.
- [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) — Erik S. and Barry Zhang, Anthropic, December 2024. The workflow–agent distinction used here; terminology varies across the field. Checked 2026-09-07.

## Related reading

- [AI model](https://aecon.ai/glossary#ai-model)
- [Tool](https://aecon.ai/glossary#tool)
- [Agentic](https://aecon.ai/glossary/agentic)
- [Authority](https://aecon.ai/glossary#authority)