I had Google’s Gemini 3 write a little CLI tool for writing and publishing posts and notes from the command line. It’s nothing fancy. Basically some template files that get loaded up into NeoVim so I can write posts. Then saves and let’s me publish via options wrapped around git commands.

I keep trying to reduce the friction for writing in hopes that the easier it is to publish a post on this blog, the more I’ll be likely to publish more. It’s sort of working. Moving to include a note type helped a little I think. It puts less pressure on writing a well thought out article on a specific topic or thought and reduces the barrier of entry to posting smaller, Twitter/Bluesky/Mastodon type of posts.

Developing little tools like this has been the best use of “AI” agents/tools I’ve found so far. There are tons of little things that I have wanted to automate but haven’t had the time to work on. Using these tools I can write out a description of what I’ve planned out in my head. Next I’ll feed that in and let it do some research and ask me any clarifying questions about implementation and overall preferences. Then I can just send it off to work on it while I’m doing something else. The ability to run the tasks in the background has been really convenient and allowed me to start getting some of my ideas put into action.

Not really sure where I’m going with this post but the other thing I’ve found the LLMs are good at is running a D&D game. With Google Gemini I was able to upload a few of my character sheets, describe a broad scenario and play a text adventure D&D with it. It’s great to be able to just open up that conversation and have a few back and forth messages to continue a story that will never really end. It even rolls dice for me when necessary and generates images for important scenes or if I ask for clarification on the setup and environment of an encounter.

I’ve also been able to do all of this on the free tier of Gemini. It’s very rare that I run out of “credits” or get rate limited. Overall I am still a skeptic that any of these AI companies can turn a profit unless there is a major leap in one or more of the technologies or hardware used to train and run these models. But I will keep using them for these two purposes until the companies either crash or force me to pay to use them.