jonas.splinter
Search
Topic

AI

Artificial intelligence and language models: how I use them in my work, what I build with them, and what I build for them.

Most of what is called AI today is built on large language models: systems trained on very large amounts of text that can write, summarise, translate, answer questions and write code. They are useful and also unreliable in specific ways. They can state wrong things with confidence, so the real work is in giving them the right context and checking what comes back.

I am deeply into AI. I experiment with ChatGPT and Claude, build custom solutions on top of them, and since December 2025 I have been building software with Claude Code. Much of this website was made that way.

I also build for AI. Automate & Tweak is a directory of about 1,400 AI and automation tools in 12 languages, made to be read by assistants as well as people. It has an llms.txt file and a free public MCP server, so Claude, ChatGPT or Cursor can search and compare tools directly.

AI also runs through my automation work. A language model is often the step that turns messy input, such as free text, into something structured that the rest of a workflow can use.

2 blog posts

Longer articles about AI

1 update

About AI, from Blockspot only

Blockspot

New post: Let an AI sort the inbox, then snooze what can wait

New

How a small AI model labels each new mail as high or low value, why low-value mail is snoozed to two fixed times a day, and why I let it err towards important.