How your employee learns and grows
An EmAGI employee learns from every task, conversation and correction. This page covers what it does about that by itself, how to teach it something in one sentence, and where the dashboard shows what it has learned.
What does it do by itself?
- A correction said in plain words becomes a rule, and it keeps to it.
- It notices when it does the same work a second time and turns that work into a skill.
- You changed your mind: your latest word applies, the earlier one stays in the history but no longer guides the work.
That is why after the first week it stops asking the same things again.
How does it pick up new knowledge?
Every night it goes over the past day in all chats and writes itself new rules. After the main night task it studies a question about your field and saves what it learned.
Curiosity works in the daytime too: when its own question about your business comes up during a task, it looks into it by itself, in memory, past conversations, files and on the internet. What only you know, it asks briefly when you bring up that subject yourself.
How do you teach it something?
Say it once, the way you would tell a person: what is wrong and what you expected. "From now on always answer in English", "Ask me before writing to a client". What must be kept for good, mark with the word "remember".
A rule can be for all chats or for one. A ban on taking something out of a chat ("what we discuss here stays here, do not mention it in other chats") should be written as a rule for all chats: a rule written for one chat only is not seen by its other chats. Style and tone that differ in one place are written for that one chat.
To keep a piece of work that went well, say "turn this into a skill".
Where does the dashboard show what it has learned?
On the agent's page:
- "Memory & personality": the personality, main memory, work standards and other core files, an editor and a "Save" button. An edit applies from its next answer.
- "Skills": what it has been trained in and how often it uses each. A click opens the skill text and "Edit in editor"; one you no longer need goes to "Archive".
- "Reset memory" in the danger zone: tick what to erase, from conversations to rules. The name, character, settings and access stay.
One agent's memory opens to another of your agents only with your button in "Memory & personality". Usually it is simpler to tell the employee in words what to change: it will fix it itself.
How does it suggest better ways to work with it?
Once a week it looks at how you work together, and if it sees a way to save your time, it suggests one thing: split projects into topics, call instead of long voice messages, set a goal with a check, connect your Drive or mail. There is nothing to switch on.
The suggestion arrives where you write to it, or in the "How to work with me" topic if you have a group with topics. After your "yes" it does everything itself. Hear "no need", and it will not suggest that again; if it sees you are busy, it stays quiet.
Read next: Memory: what the agent keeps · One agent, several chats · Dashboard