Use cases for Autonomous Web Agents

Here is a list of use cases where we want autonomous agents that can flexibly operate on the Web:

List of domain areas: https://annuel2.framapad.org/p/webagents-uc-tf-areas-a7ed?lang=en

Use case 1: Assistant in discussion boards.

A Web agent that can monitor arbitrary discussion threads (Twitter feed, forums, Github issues, subreddit, etc.) so that whenever someone (or some other software agent) mentions @discussionBot the agent tries to interpret the text after this mention and act accordingly, especially producing a reply to the thread. For instance:

@chatGPTAgent, give me a code snippet in C++ for a sort algorithm.

or

@GitAgent, create an issue with tag "critical" titled "Security problem in GitAgent API" that links to this issue, and assign it to @Antoine-Zimmermann.

The former example should reply to the thread with the output of ChatGPT for prompt "give me a code snippet in C++ for a sort algorithm", which avoids going to a third party system, write the prompt, then copy paste the result back to the thread and put it in the right form. The later example must start a new discussion thread and probably respond to the initial with a link to the new issue.

Discussion boards of any kind have a limited set of possible actions that could be identified by URIs from a dedicated ontology: create thread, read thread, list (latest) threads, respond to thread/message, read message, create account, etc. These actions typically have the same type of input (e.g., a title, a body, where to submit and authentication details) so they could be described systematically in a standardised way in a knowledge graph. The agent should not have to be reimplemented differently when participating in a new discussion board. This means that some discussion sites would require a wrapper that takes standardised input as defined by the ontology (e.g. title, body, URI of thread and identity) and translate it to the appropriate operations on the website.

Use case 2: ...