Really? You don't think it's a matter of updating the prompt and expanding the context to include more of the code?
I don't know the exact method of how they generated these comments, but I think just focusing a prompt on a single method with the entire codebase in context would yield much better results.
* Past incidents
* Regulations
* What the code used to do, and why that was bad
The whole point of a really good comment is that you can't infer it just by looking at the code. That's why it's there at all!
Yes they become paragraph long. But that's the point. That's where I search for "why". And not just "why", "when, why and with what reasoning?".
Imagine if there were some interference that could derive a comment based on history and current code!
I'm certainly a fan of including lots of detail in commit messages, but often I do so and then realize "hey, this particular information would be better situated close to the code", and I move it.
Other times, yes, it's really just best to leave in the commit message because no one looking at the after-version of the code will care. For example, removal of a deprecated feature.
It doesn't even need fucking AI, just regexp and just a configuration, e.g. if your Jira project is called NAI, any word that has the format NAI-[0-9]* should be such a link to https://jira.lookatmenotusingfancyai.com/task/NAI-xx
Edit: then again, Jira tasks can end up being a place for long discussions. If the AI can summarize the end-result/the decisions taken, it would be useful.. but also as a Jira plugin.
AI isn't magic. If the "why" isn't in the code, there's no way for it to magically synthesize it.
Comments explaining "there's a check for this edge case because we saw a problem in production when client X did something, see JIRA-1234 for more details" are what you should be aspiring to. If a human doesn't write that down, the AI can't invent it.