> ## 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.
# Quickstart: your first coworker
> Hire an AI coworker and finish a real task in about ten minutes. No configuration, no setup — just pick a role and give it something to do.
Ten minutes, one real task, nothing to configure first.
## Before you start
Nothing. You don't need to connect an app, write instructions, or set anything up. All five ready-made coworkers do useful work with what they arrive with.
Have one real job in mind — something you were going to do this week anyway. A test question tells you very little; a real task tells you everything.
## Hire a coworker
Go to **Cowork → Roster** and choose **Hire**.
Six cards: five ready-made roles and **Start from scratch**. Take a ready-made one — even an imperfect fit beats an empty coworker on your first go.
Not sure? The **Research Analyst** is the easiest to judge, because you can check its work.
It appears on your roster with a name and an avatar, ready to work. There's no setup step.
## Give it a real task
Open your new coworker and start a task. Write what you want the way you'd brief a colleague.
**A good first task for each role:**
| | Try |
| ------------------- | ---------------------------------------------------------------------------------- |
| Research Analyst | "Research the three biggest players in \[your market] and tell me how they price." |
| Executive Assistant | "Go through my inbox and tell me what actually needs me today." |
| Support Agent | "Here are ten open tickets — group them and draft a reply to each." |
| SDR | "Research these five companies and draft a first email to each." |
| Data Analyst | "Here's last quarter's export — what should I be worried about?" |
**Say what you want to end up with.** "A one-page summary I could send to my boss" produces something better than "look into this".
Attach a file if the job needs one — the paperclip, or drag it onto the conversation.
## Watch the first one
Stay for this one, even though you won't need to later.
You'll see a row per step: what it searched for, what it opened, what it worked out. Expand any of them. On a longer job it keeps a visible checklist and ticks items off.
You're not supervising. You're learning where this coworker is reliable and where it needs a tighter brief — which is what makes the next nine tasks quick.
## Correct it
When something isn't quite right, say so in plain words, right there in the chat:
> "Too long — half that next time."
>
> "Lead with the recommendation, then the reasoning."
It [remembers](/cowork/control/memory). This is the single highest-return thing you'll do, and it takes a sentence.
## Approve what it asks for
If your task involves sending or changing something, your coworker will stop and show you exactly what it's about to do. Approve it and it continues; decline and it carries on with everything else.
You can [decide these in advance](/cowork/control/permission-rules) once you know what you're being asked about.
## Then what
Five minutes here explains most of what your coworker can and can't do.
Your inbox or CRM makes it considerably more useful.
Once you've corrected the same thing three times.
When you trust it enough to run without you.
## Common questions
A few minutes for something straightforward, longer for real research. You can close the tab — it keeps working.
Say so and it adjusts. You don't start over, and it doesn't need re-briefing from scratch.
Anything leaving your workspace waits for your approval, and a new coworker has no apps connected — so its first task can only read, research and produce files.
Not yet. Get one working well first. The signal for a second is catching yourself wishing the first were better at something specific.