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

# Cowork overview

> Meet Cowork, CogniAgent's virtual employees. Reusable AI workers that run tasks, write code, and use your tools.

Cowork is CogniAgent's autonomous "virtual employees" experience. Instead of building a workflow step by step, you hand work to a worker — a reusable AI colleague that plans its own steps, writes and runs code, edits files, and calls your connected tools to get the job done.

Think of Cowork as hiring staff instead of drawing a flowchart. You describe the outcome you want in a chat, the worker asks clarifying questions if it needs to, and then it works — in the background, in its own sandbox — until it has something to show you.

<Info>
  Cowork is a different product surface from [Conversation Flows](/conversation-flows/overview) (multi-agent, flow-authored conversations) and [Applications](/applications/create-application) (visual workflow automation). Cowork is chat-first and worker-first: you talk to a colleague, not a canvas.
</Info>

## The problem Cowork solves

Most automation tools ask you to specify every step in advance. That works for repeatable processes, but falls apart for open-ended work — reconciling a spreadsheet, researching a market, drafting an outreach sequence — where the right steps depend on what the worker discovers along the way.

Cowork flips that. A worker is given a goal, not a script. It reasons about what to do next, uses the tools it has, checks in with you when it needs input or approval, and keeps going. You stay in the loop as a reviewer and collaborator, not as the person clicking through every step.

## The hierarchy at a glance

Everything in Cowork nests under your workspace in three levels:

```mermaid theme={null}
flowchart TB
    WS[Workspace<br/>shared by everyone on the team]
    W1[Worker: Researcher]
    W2[Worker: SDR]
    T1[Task<br/>private to its owner]
    T2[Task<br/>private to its owner]
    T3[Task<br/>private to its owner]
    WT[Worker-less task<br/>no worker · private to its owner]

    WS --> W1
    WS --> W2
    WS --> WT
    W1 --> T1
    W1 --> T2
    W2 --> T3
```

* **Workspace** — the top-level container. Every worker in it is visible to, and editable by, everyone on the team — there's no admin-only tier in v1.
* **Worker** — a reusable AI colleague: a persistent definition of instructions, skills, assets, and connections that anyone in the workspace can assign work to.
* **Task** — a single, private run or conversation. A task is owned by the one person who started it — nobody else in the workspace can see it, even though the worker itself is shared.

<Note>
  Projects (a shared grouping layer above tasks) and sharing a task with teammates are on the roadmap, not in this hierarchy today. A task is private to its owner in v1.
</Note>

Read the full vocabulary and mental model on [Core concepts](/cowork/concepts).

## What a worker is, at a glance

A worker is defined by four things, all shared workspace-wide the moment they're created or edited: instructions, skills, assets, and connections. See [Workers](/cowork/workers/overview) for the full anatomy of each.

A worker also carries a status of `active` or `disabled`. Disabling a worker soft-retires it — a safe alternative to deleting one that still has running tasks — rather than deleting it outright. See [Workers](/cowork/workers/overview) for how to create and manage one.

## Worker-bound vs. worker-less tasks

A task is where the actual work happens, and it can start in one of two ways:

<Tabs>
  <Tab title="Worker-bound task">
    Started from a specific worker's page, inheriting that worker's instructions, skills, connections, and files — the common case, delegating to a colleague who already knows the job.
  </Tab>

  <Tab title="Worker-less task">
    Started from the top-level "New task" button with no worker attached — a fresh chat with standard tools like file editing and web search, but no inherited instructions or skills.
  </Tab>
</Tabs>

Both kinds of task are private to the person who started them and follow the same [agentic loop](/cowork/loop/agentic-loop). Full lifecycle details live on [Tasks](/cowork/tasks/overview).

## Where the work actually runs

When a task does more than chat — running code, editing files, calling a tool — it automatically gets its own isolated cloud sandbox, not your machine or a shared environment. See [Execution, sandbox & files](/cowork/tasks/execution) for the full picture, including how files sync and how sandboxes pause and resume.

## Next steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/cowork/quickstart">
    Create your first worker and run a task, end to end.
  </Card>

  <Card title="Core concepts" icon="book" href="/cowork/concepts">
    The full vocabulary and mental model behind Cowork.
  </Card>

  <Card title="Workers" icon="user-gear" href="/cowork/workers/overview">
    The anatomy of a worker and how to create and manage one.
  </Card>

  <Card title="Tasks" icon="list-check" href="/cowork/tasks/overview">
    How tasks run, from first message to completion.
  </Card>
</CardGroup>
