Prompt with AI is like …. golfing
Let’s talk about using AI in this current age of 2025. Here’s a stock photo to help out:
As obvious as it sounds, writing prompts for an AI chat session is a lot like golf. Basically, honing down until you get a good answer OR run out of chat memory and have to move to a new session.
- Use the driver off the tee.
-Post your first prompt: “I’d like all the recent JIRA tickets that have work logged on them.” - On the fairway. Now you select an iron to get you close to the hole.
-“I mean the tickets that have worked logged in the last two weeks.” - Now you are on the green. You decide to putt.
-“For team members Sally and Frank.”
Sometimes though, AI completely misses it, for instance step two (real scenario with Rovo, Atlassians’s new AI assistant) – “Here you go, a JQL that lists all the tickets with time on them” — but since getting work time requires iteration through the worklog api data, AI didn’t do that correctly and just gave you tickets with recent activity that had work logged at any time, even three months ago. In this case 2A., need that pitching wedge to get onto the tee.
I’ve been working with Claude, Gemini, Copilot, ChatGPT, Ollama, and a lot of minor AI llms — they all do this as you can imagine, some just get you there faster. I am finding that smaller less known models need more time to get what you need. For instance, Junie in IntelliJ. Or some of the GitHub AI things, and definitely Rovo for Atlassian are less mature.
A big thing is learning how to summarize a session so as to continue on again at some point — have AI summarize, or save the chat context ported off for reloading in a new session.