- [RAG (Retrieval-Augmented Generation)](https://parallel.ai/learn/rag-retrieval-augmented-generation)
Reading time: 2 min
September 8, 2026
# Autonomous agent
An autonomous agent is software that pursues a goal on its own, choosing its steps and taking actions to reach it. It keeps working until it meets the goal instead of waiting for a new prompt each time. It builds on what an AI agent is by adding independence and persistence.
Reading time: 3 min
An autonomous agent is an AI system that plans and acts toward a goal set by a person, with little or no supervision at each step.
## What is an autonomous agent?
An autonomous agent is software that pursues a goal on its own, choosing its steps and taking actions to reach it. It keeps working until it meets the goal instead of waiting for a new prompt each time. It builds on what an AI agent is[what an AI agent is] by adding independence and persistence.
A spring 2025 MIT Sloan survey[MIT Sloan survey] measured the rise. MIT SMR and BCG found that 35% of respondents had adopted AI agents by 2023, with another 44% planning to deploy them.
## Key characteristics
An agent shares a few core traits, shaped by its AI agent architecture[AI agent architecture].
- - **Goal-driven:** The agent decides the steps to reach a goal you set.
- - **Tool use:** The agent calls outside tools and application programming interfaces (APIs), such as web search, code execution, and databases, to carry out plans.
- - **Memory and reasoning:** The agent remembers past steps so it can reason about what to do next.
- - **Limited supervision:** Gartner[Gartner] predicts agentic AI will make at least 15% of daily work decisions by 2028, up from 0% in 2024.
- - **Human oversight for high stakes:** The agent pauses for a human before it takes risky or costly actions.
## Example
An autonomous research agent[autonomous research agent] shows the idea in action. You give it a question, and it searches the web, then reads and combines the best sources into an answer. It decides which pages to open and when it has gathered enough, without supervision at each step.
_Parallel's agentic web research benchmark results, comparing accuracy and cost across leading providers. Source: Parallel benchmarks[Parallel benchmarks]._
## Related terms
- - **AI agent:** the broader category an agent belongs to.
- - **Agent harness[Agent harness]:** the framework that runs an agent's loop.
- - **Deep research[Deep research]:** the multi-step research an agent can produce.
- - **Web search API for agents[Web search API for agents]:** the live search layer agents call.
## FAQ
**Are autonomous agents fully independent?**
Not yet, since most still need human oversight for safety. The 2025 MIT AI Agent Index[2025 MIT AI Agent Index] found that across 30 leading deployed agents, only four provide agent-specific safety evaluations.
**Why do many agent projects fail?**
Gartner predicts[Gartner predicts] that companies will cancel over 40% of agentic AI projects by the end of 2027. The causes include escalating costs, unclear business value, or inadequate risk controls.
**How is an agent different from a chatbot?**
An agent takes actions and completes tasks toward a goal you set, going beyond a chatbot that only answers the questions you ask.
You can put these ideas to work with Parallel's web APIs built for agents. Start Building[Start Building] with our docs.
By Parallel
September 8, 2026