Task: Add editing of information entities via AI chat

As practice has shown, AI is very unstable when it comes to uncontrolled work with information. It is almost impossible to reliably control it so that the AI agent reliably performs the same actions according to the same logic in similar situations. But not involving the agent in information management is also not cool, otherwise what are they needed for.

Here I am thinking of testing a new interesting mechanic: when something needs to be updated, instead of letting the AI perform the final database update itself, it outputs a card with proposed changes into the chat. It should turn out to be quite convenient. That is, I tell it something like "Find the coordinates and other useful data for this object," the AI finds it, but instead of sending an update mutation, it simply outputs all this in the form of a separate card with proposed changes, and through a personal interface I can see what kind of entity this is and what fields are proposed for change, and simply apply all suggestions or just specific ones. It should be convenient.