How I use AI
I use AI as a practical collaborator for shaping ideas, editing content, changing the site, and checking the result before I publish it.
Start with a question
I start with a question about my practice or about the site itself: how should an experience be represented, how should two ideas connect, or how can a reader find the right context? I give the agents the repository, the relevant source material, and the constraint I am trying to respect. We work through possible structures and wording together.
The useful output is something I can respond to: a page, a small prototype, a clearer explanation, or a test that exposes a broken link. I can then change the direction and try again.
The definitive resources card is the reference shelf I use while working: Jekyll, Ruby, Liquid, kramdown, Git, Cloudflare, web standards and the Jekyll plugins that support this site.
For this website specifically, my working set is Codex, Ollama with Hermes 3, WebStorm, and Aider. Codex helps me work through repository changes and tests; Ollama and Hermes give me a local model path to explore; WebStorm is my web IDE; and Aider gives me another terminal-oriented way to work with code.
Make, inspect, revise
On this site, AI agents help me with page structure, draft wording, code, source research, and checks. The git history shows that pattern in practice: I used the agents while moving the site from its earlier renderer to Jekyll, adding connected cards and automatic backlinks, adding the Food page, and preparing the Cloudflare Workers deployment. Each change left a reviewable file or build result that I could inspect.
I stay involved in the choices. An agent’s confident wording does not establish that a project happened, that a qualification exists, or that a technical claim is correct. Those claims need my source material or a relevant reference. I am responsible for what I publish.
Keep the thinking connected
My shared-brain article explores using small connected notes with a generative AI collaborator. This site applies a similar idea: each card has a place of its own, and meaningful links explain why another note is worth following.
The distinction matters. An explained relationship carries context. An ordinary mention simply says that one note refers to another. Neither should quietly become a claim of expertise or a completed project.
Be clear about what is made
This website is built by me and my AI agents. I want that assistance to be visible without confusing generated material with evidence of my practice. AI can help draft a description or propose a connection, but I decide whether it accurately represents my experience and whether it belongs in the published site.
This is also part of my exploration of MySotongkoo: making local AI useful in executive learning, with Python, Textual interfaces and AI-assisted development. The questions are about how people use, question and learn from these tools as much as about what the tools can produce.