> ## 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.
# AI coworkers that do the work
> Hire an AI coworker instead of building a workflow. Give it a goal, not a script — it researches, runs the numbers, uses your apps and asks when it needs you.
Most automation asks you to describe every step in advance. Cowork asks you to describe the outcome.
You hire a coworker, tell it what you need, and it works out how — researching, opening your spreadsheets, using your apps, asking when it's genuinely unsure. You review what comes back.
## Why this is different
Step-by-step automation is excellent when the steps never change. It falls apart on the work that fills most days: reconciling two exports that don't agree, researching a market you don't know, drafting outreach that has to say something specific about each company.
That work can't be drawn as a flowchart, because the right next step depends on what you find. So Cowork doesn't ask you to draw one.
You give it a goal. It decides how to get there, and adjusts when reality doesn't match the plan.
Your inbox, your CRM, your spreadsheets — and a real browser for anything without an integration.
When something needs a decision or a password, it stops and asks. You approve; it continues.
Correct it once and it remembers. Over weeks it needs telling less.
## What people actually use it for
* **Research that ends in a decision** — market sizing, competitor pricing, whether a claim stands up, with sources attached
* **The spreadsheet nobody wants to open** — reconcile two exports, clean a contact list, chart a quarter
* **Inbox triage** — what actually needs you today, and drafts for the rest
* **Outreach worth replying to** — research the company first, then write something true about it
* **First-line support** — triage the queue, draft grounded replies, escalate what's risky
* **The portal with no export button** — a real browser, driven for you, handing over at the login
## How it works, briefly
Pick one of [five ready-made roles](/cowork/coworkers/ready-made) or build your own. It arrives with instructions, a personality, and the right capabilities switched on.
In plain words. Attach files if it needs them. It asks clarifying questions before starting if something's genuinely ambiguous.
You can follow every step as it happens, or close the tab and read the recap later.
Anything that leaves your workspace — an email, a change to your systems — waits for your yes.
## Is Cowork the right tool?
CogniAgent gives you three ways to put AI to work, and picking the right one matters more than anything else on this page.
| You want… | Use |
| ------------------------------------------------- | ------------------------------------------------------- |
| A judgement call made, or a messy job figured out | **A coworker** |
| The same steps run the same way, every time | **[A workflow](/applications/define-workflow)** |
| A conversation held with a customer | **[A conversation flow](/conversation-flows/overview)** |
The full comparison is on [Cowork vs workflows vs conversation flows](/cowork/vs-workflows).
## Where to go next
Hire a coworker and finish a real task in ten minutes.
Research analyst, assistant, support, SDR, data analyst.
Every capability, one page each.
The mental model, in five minutes.
## Common questions
No. You describe what you want in plain English. Your coworker writes and runs whatever it needs to behind the scenes — that's the point of it.
Only where you've allowed it to. Anything that leaves your workspace waits for approval by default, and you can tighten or loosen that per action. See [Permission rules](/cowork/control/permission-rules).
Each task works in its own private workspace, isolated from every other task, and it's discarded when the task ends. Nothing runs on your computer.
A chatbot answers. A coworker does — it opens the file, runs the numbers, sends the email, and comes back with the finished thing rather than instructions for making it.
You see every step it took and can check its work, including exactly what it changed in a file. Correct it in the chat and it remembers for next time.