Jira has an AI feature called Rovo. It's terrible.
Giving an API key to Claude however gets you exactly what you need. Albeit in quite a risky way though.
Definitely, that's the issue with legacy tools and just plugging AI agents on top of existing API capabilities that were designed for human in the first place.
Retrofitting agent-safety onto APIs that predate agents is harder than it looks (guardrails, permissions scopes etc.)
Disclosure: I'm CTO of an ITSM product in this space (Siit), so I think about this a lot.
Russain roulette.