Had some “agentic AI” training at work today. Work are providing us with Cursor and Copilot for coding tasks, as well as Gemini, Atlassian Rovo, and an LLM Gateway for non-coding tasks. Today was mostly getting to grips with terminology. One person walked through installing Opencode, adding an MCP for Atlassian. They then pasted a Jira link into an Opencode agent that planned some work and attempted to implement it. I think everyone was suitably impressed, but it was clear after a few questions and comments that a lot of stuff we do is either novel or very delicate. Novel means we don’t fully know what we would put on a ticket or what to tell a coding agent. Delicate because we don’t really want an agent snooping around our Azure tenant full of PII data and tens of production customers working in a highly regulated industry. I’ve also recently been told I need to start logging my time again after six years. So this afternoon was spent putting together requirements for an agent to code a small command line utility to complete that for me - can be run as a cronjob on my server. Some other people put together an AI skill to do it - but that was going too far for me.