Give AI employees their roles
Treat it like a new hire: write the Agent a job description, grant system access, and set who approves.
Treat it like a new hire: write the Agent a job description, grant system access, and set who approves.
The company is set up. Now give your AI employee a real role.
The brain is the model, as in Lesson 1. These three are what make it good at your company's work.
Example: the sales lead says in the Sales Room, "Follow up on the Maple Kitchen quote."
It recalls the relevant preferences. No need to repeat yourself.
If a Skill matches the task, it reads it first, then follows it.
Only when it needs email or sheets, and it uses your connection.
It also reads earlier messages in the thread, so follow-ups don't start from scratch.
Set access by type. Teach the rest on the job.
The company from Lesson 2: sales, ops, finance, and the owner.
Anyone can ask it first.
Owner only, with the owner's email and calendar.
For the sales team. Runs quote follow-ups.
Finance and owner only. Runs reconciliation.
Example: the Sales Agent sits in the Sales Room, uses the Gmail the sales lead shared, and the sales team can use it.
No need to write everything up front. Tell it as you go and it remembers.
From now on, follow-up emails stay under five lines. That's the Sales Room rule.
Noted: follow-up emails in the Sales Room stay under five lines.
You can ask an Agent what it remembers. More: kylon.io/help/ask-agent-to-remember-preference kylon.io/help/correct-outdated-agent-memory
More: kylon.io/help/give-an-agent-a-clear-task
Most common tools have ready-made connections. For the rest, the tech lead connects with an API key. After that, what matters is what it may do and who approves.
Pick it on the Connections page and sign in yourself.
No ready-made option? The tech lead connects with a key.
Have a brought-in Agent export CSVs on a schedule.
The Connections page shows what's available. More: kylon.io/help/connect-a-tool
More: kylon.io/help/connect-a-tool kylon.io/help/share-a-connection
Reads and edits happen directly. Only outbound messages wait for a person.
Searches email and sheets to answer you.
Creates events, edits sheets. No approval step.
Email and X: drafts first.
An Agent can only do what you checked when authorizing. More: kylon.io/help/what-agents-do-with-connections
A chat reply doesn't count. Click the button on the card.
More: kylon.io/help/approve-agent-actions
Week one: read-only and drafting. Expand once it's steady.
It looks things up and writes summaries.
It writes outbound emails and messages. You send them.
Edits sheets, books events, no need to ask.
For read-only: check only read permissions when authorizing, or give it a read-only account.
Then fill in the job description on the next page.
What does it take on? Use the job you picked in Lesson 1.
Main, personal, or function?
Which Rooms? Whose connections? Who can use it?
Who signs off on outbound emails and messages?
| Spell out | Notes | Example: Quote follow-up Agent |
|---|---|---|
| Which type | Main, personal, or function | e.g. Function Agent |
| The job | What it owns day to day | e.g. Check quotes daily, find any silent for 7 days, draft follow-ups |
| Which Rooms | Where its threads live | e.g. Sales |
| Which connections | Accounts it needs | e.g. Sales lead's Gmail, quote tracker |
| Who can use it | Who's on the list | e.g. Sales team |
| Who clicks Send | Who signs off on outbound | e.g. Sales lead |
The example is a template. Swap in your own role.
If not, your message never reaches it.
Unclicked draft or approval cards keep it paused.
A brought-in Agent stops replying when its computer is off.
More: kylon.io/help/agent-not-responding
More: kylon.io/help/approve-agent-actions kylon.io/help/share-a-connection kylon.io/help/reconnect-a-connection
A job description, connected systems, and every outbound message waits for the lead's sign-off.
Spreadsheets become Apps. SOPs become Skills.