Currently the spec only mentions
The times at which a browser agent performs an observation are implementation-defined
But to me it seems that in practice it would be the app itself that is best positioned to perform an observation, since only the app dev knows what part of the app state would most relevant to a browser agent, and therefore it can send observations when that state changes and avoid filling up the context of an agent with irrelevant stuff.
Currently the spec only mentions
But to me it seems that in practice it would be the app itself that is best positioned to perform an observation, since only the app dev knows what part of the app state would most relevant to a browser agent, and therefore it can send observations when that state changes and avoid filling up the context of an agent with irrelevant stuff.