> ## Documentation Index
> Fetch the complete documentation index at: https://open-index.io/llms.txt
> Use this file to discover all available pages before exploring further.

# What are autonomous agents?

> A working definition of autonomous agents, the human-in-the-loop spectrum, and what it actually takes to remove the human.

An **autonomous agent** is a system that pursues a goal by repeatedly deciding its
own next action — choosing tools, gathering information, and adapting — rather than
executing a fixed script. The word doing the work is *deciding*: a workflow follows
steps you wrote; an agent chooses steps you didn't.

## The autonomy spectrum

"Autonomous" is not binary. Most real systems sit somewhere on a spectrum defined by
**how often a human has to intervene**:

| Level      | The human…                             | Example                              |
| ---------- | -------------------------------------- | ------------------------------------ |
| Assisted   | drives; the agent suggests             | Autocomplete, copilots               |
| Supervised | approves each meaningful action        | "Agent proposes, human clicks run"   |
| Steered    | is pulled in when the agent gets stuck | Most "agents" in production today    |
| Autonomous | reviews outcomes, not steps            | Handles the whole task, reports back |

The jump that creates business value is **Steered → Autonomous** — and the thing
that blocks it is almost never raw model capability. It's the context layer and the
harness around the model. See [Human steering](/guide/human-steering).

## Removing the human in the loop

"Remove the human" is a misleading goal if it means *no human ever*. The right goal
is to **move the human from the inner loop to the outer loop** — from approving every
step to reviewing outcomes and correcting the system.

Three things have to be true before that's safe:

<Steps>
  <Step title="The agent can find the right context on its own">
    No pre-injection, no human deciding what to paste in. The agent searches and
    discovers. See [Context search accuracy](/guide/context-search-accuracy).
  </Step>

  <Step title="Its context is trustworthy and traceable">
    Facts are current, non-conflicting, and carry provenance, so an outcome can be
    audited without a human reconstructing what the agent "must have" used. See
    [Provenance](/guide/provenance) and [Conflicts & staleness](/guide/conflicts-and-staleness).
  </Step>

  <Step title="The harness catches its own failures">
    Soft gates stop early exits; critic sub-agents catch bad reasoning in-session,
    before it reaches the user. See [The harness](/guide/the-harness).
  </Step>
</Steps>

Get those three right and the human's job shifts from *steering* to *supervising the
system* — which is where autonomy actually pays off.

## Domain agents raise the bar

General agents can lean on the model's world knowledge. **Domain agents** — legal,
healthcare, sales, infrastructure — can't: the knowledge they need is private,
structured, and changing. That's exactly why the context layer matters more the more
specialized the agent is, and why a [structured context engine](/why-open-index)
tends to be the highest-leverage investment for a domain agent.

<Card title="Next: The harness" icon="gears" href="/guide/the-harness">
  What sits around the model to make the loop reliable.
</Card>