Guide

Working with AI mates

A short guide for everyone who meets an AI mate at work: what it is, what it does, where it can be wrong, and how people stay in control. Share it with your team before a mate joins your first meeting.

Last updated: October 7, 2026

What an AI mate is

An AI mate is software, not a person. It joins a meeting as "Name (AI)" and says that it is an AI when it joins, by voice and in the chat.
It listens, answers questions by voice from the knowledge your organisation connected (for example Confluence pages and Jira tickets), and suggests decisions, action items, tickets and pages.
Its answers are generated by a language model. It does not understand things the way a colleague does: it predicts a likely answer from what it can read.

Where it can be wrong

It can sound sure and still be wrong. Check anything important against the page or ticket it comes from before you rely on it.
It only knows what it can read. If a page is missing, outdated or outside what the mate may read, its answer will be incomplete.
It can mishear. Speech recognition makes mistakes, especially with names, numbers and crosstalk. Review what it recorded.
It does not see the screen or the room. It hears the meeting audio and reads the chat, nothing else.

You stay in control

Nothing is written without a person. Before a mate creates a ticket, writes a page, records a decision or hands work over, an approver your organisation chose must approve it.
How approval works. In Zoom, approvers approve in the console (Approve or Reject on the meeting card); a spoken "yes" does not count. In Microsoft Teams, an approver of your organisation can also say "yes" in the meeting. Guests never approve.
Code stays a draft. A developer mate opens draft pull requests only. People review and merge.
Everything is logged. Every action, with who asked and who approved, is in the console's audit log.
You can ask it to leave. The host can remove it from the meeting at any time, like any participant.

How to recognise its work

Tickets, pages and comments a mate writes end with a note such as "(created by Alex, AI, after approval)".
Jira issues and Confluence pages it writes carry the label ai-generated, so you can search for them (labels = ai-generated in Jira, label = ai-generated in Confluence).
Draft pull requests come from the platform's GitHub app (shown as a bot) and say which mate wrote them.
Keep these marks when you share the content outside your organisation.

What it is not for

Do not use AI mates, or what they record, to evaluate, monitor or rank colleagues, to decide on hiring, pay, promotion or who gets which work based on personal traits, or to read emotions. These uses are outside the service's intended purpose (see the Terms of Service) and are restricted or prohibited by the EU AI Act.

Report a problem

If a mate says something harmful, acts without approval, or you find a security issue, write to support@hgb.fi. Include the meeting and the time; the audit log helps us find what happened.