Autonomous Agent

Actor agents published

Also known as: Fully Autonomous Agent, Self-Directed Agent

Definition

An agent that operates with minimal or no human intervention during execution. Autonomous agents make decisions, take actions, and handle errors independently, only surfacing to humans for high-stakes decisions or unrecoverable failures. The degree of autonomy is a spectrum: fully autonomous agents require no human input after initialization; semi-autonomous agents may request approval at defined checkpoints.

What this is NOT

  • Not the same as 'agentic' (autonomy is about oversight level, not capability type)
  • Not defined by capability but by governance (a weak agent can be autonomous; a strong agent can require approval)
  • Not synonymous with 'unsupervised' (autonomous agents should still be monitored)

Alternative Interpretations

Different communities use this term differently:

robotics

A physical or simulated robot that navigates and acts in an environment without continuous human control, using sensors and onboard computation.

Sources: Autonomous robotics literature (1990s-present)

llm-practitioners

An LLM-based agent that can execute multi-step tasks end-to-end without human approval at each step. Examples include AutoGPT, Devin, and similar "set it and forget it" systems.

Sources: AutoGPT project documentation, Devin (Cognition Labs) announcements

enterprise-ai

An AI system authorized to complete business processes independently, typically within defined guardrails and audit requirements.

Sources: Enterprise AI governance frameworks

Examples

  • AutoGPT running a multi-hour research task without human input
  • A trading bot executing strategies based on market conditions
  • A CI/CD pipeline that deploys on green tests without human approval
  • Devin completing a coding task from spec to PR

Counterexamples

Things that might seem like Autonomous Agent but are not:

  • A chatbot that requires a human to send each message
  • An agent that pauses for approval before every external API call
  • A copilot that suggests but never executes

Relations

  • specializes agent (Autonomous agents are agents with high autonomy)
  • inTensionWith human-in-the-loop (Human-in-the-loop reduces autonomy by definition)
  • overlapsWith supervisor-agent (Supervisors may manage autonomous sub-agents)

See Also

Implementations

Tools and frameworks that implement this concept: