FRENES

Glossary · Agents and autonomy

Autonomous agent

An autonomous agent is an AI agent that can pursue a goal and choose several actions without involving you at every step. Its independence remains bounded by instructions, permissions and stopping conditions, and does not guarantee that the task will succeed.

Autonomous agent: the explanation

You provide an objective, resources and limits. The agent then chooses steps, uses permitted tools and examines results to adjust its approach. It might repeat a search or try another method without immediately asking for help. The surrounding software still needs stopping conditions, such as repeated failures, a spending limit or an action requiring your approval. Those controls define how far the agent may proceed before it must pause or return the task to you.

Autonomy reduces the amount of intervention needed during a long or repetitive task. An agent might monitor a folder, prepare a report or investigate several possible solutions. You do not have to specify every click or search. This capability reflects design and delegation choices, rather than a universal property of the model. A system can operate independently while gathering information yet still require approval before sending messages, changing records or taking other consequential actions.

Autonomy is not a single switch that is either fully on or fully off. A system may choose its own research steps while asking you to confirm every edit. Another may act automatically within a very narrow area. Examine which decisions are delegated, how long execution can continue and what access is available. Greater freedom can amplify mistakes when the system misreads the objective or fails to recognize that it should stop and ask for help.

A concrete example

You ask an agent to monitor an invoice folder and prepare a weekly summary. It finds new files, extracts amounts and flags missing information without waiting for instructions at each step. You allow it to read documents and create a draft, but not to make payments. When an invoice is ambiguous, it sets that item aside and explains what you need to review.

Why it matters

The level of autonomy determines what you are actually delegating and how far an error might spread. Before using an agent, check time limits, spending controls, permissions and the ability to interrupt it. Also examine how it reports a blockage. An agent that keeps running without progress can consume resources or repeat an action you only intended it to perform once.

Common misconceptions

  • Autonomous does not mean authorized to do everything: an agent can choose its own steps while remaining limited to reading selected files and requiring your approval before making any consequential change to those files.
  • More autonomy does not automatically improve reliability: when an objective is unclear or checks are weak, an agent may pursue an unhelpful approach for longer instead of recognizing uncertainty and asking you for clarification.

Related terms: agents and autonomy

AI agent
An AI agent is a system that uses a model to decide which steps to take toward a task.
multi-agent system
A multi-agent system is a collection of agents that divide tasks or exchange information to pursue an objective.
planning (AI agents)
Planning in AI agents is the preparation of steps an agent intends to follow toward a goal.
agent memory
Agent memory is the set of mechanisms that lets an AI agent retain and retrieve information useful to its work.
human in the loop
Human in the loop is an arrangement in which a person participates at selected stages of an AI system’s work.
tool calling
Tool calling is a mechanism through which a model asks software to run a specific function, such as searching documents or checking a calendar.
computer use
Computer use is a capability that lets an AI agent interact with a graphical interface by clicking, typing or scrolling.
coding agent
A coding agent is software that combines an AI model with tools to perform programming tasks.

Where this site covers it

The term appears on 17 pages of this site. A few of them:

Same family

Go further

Frequently asked questions

Can it run unattended?

For some bounded tasks, yes, if the system supports it. Unattended operation needs suitable permissions, stopping conditions and error monitoring. Actions with significant consequences can still require human approval even when the surrounding research runs independently.

Can I restrict its independence?

Depending on the product, you can limit available tools, require confirmation for particular actions or set an execution budget. These restrictions should be enforced by the software, rather than relying only on instructions given to the model.

What happens when it gets stuck?

A well-designed system can stop, explain what is missing and request your help. The exact behavior varies. Check whether it avoids repeated attempts and preserves an understandable record of the steps it has already taken.

Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score