This week, I applied in practice what we learned in class last week. In my Week 2 reflection, I described learning how an AI agent could be asked to do tasks like build a website and write the necessary code itself rather than being prompted through every individual step. This week, I actually did that. I used Claude Code to rebuild this reflection journal, and this very site is the result.

The process was different from much of what I had done with ChatGPT. I described what I wanted and the agent went and did it. It looked at my existing Google Site, pulled my first two posts off of it, created the new pages, started a local preview so I could see the site before anything was public, and then created a public repository on GitHub and published the site. Along the way it explained what each step did, which mattered to me because I would not have been able to follow along otherwise.

What stood out to me was that the agent did not just execute. Before it built anything, it asked me a few specific questions about where the site should live, what sections I wanted, and what it should look like. It also stopped and asked for my permission before creating the public repository. I also noticed how much the agent's suggestions went beyond what I had asked for. It initially proposed adding a progress grid, a running log of every AI mistake I caught, and a page cataloging every AI tool I have tried. These were decent ideas, but they were not what I wanted. I asked it to keep the same simple structure as my original site, a home page with a page for each week, and it did. A lesson from this that I want to remember as I work with AI moving forward is that it will often offer, as well as describe/detail way more than what you need but it is important for me as the user to decide what actually belongs given the context of the work.

I learned that part of using AI means accepting that I won’t know how to do every step of a workflow myself but I need to continue to work on how I describe what I want clearly, ask the right questions along the way, and make sure to verify the result.