Hey all, I hope you had a good New Year. Mine was a bit of a blur, but I played a lot of Deadlock, so that was nice.
Weekly Notes
The new year feels like a good time to reflect on my journals. We're about 12 weeks into these, and my feelings are mixed.
There are things that I appreciate about these journals:
They help me ground myself each week
They help me remind myself that I exist (strange that I find myself needing this)
Interactions on my journals brings me immense joy
At the same time, I'm not convinced they're helping me in the ways I hoped they would. These journals take a long time for me to compose—which is funny, because I don't think I'm covering very much.
Part of that feeling is because I don't have great daily journaling practices, so I don't remember what happens day-to-day—while I'm recapping my week, the information is... kinda just pulled from vibes.
The topics I want to talk about from week to week also varies. Usually, I get to Monday, and I have a thought that's been nagging at me, so I type it up. It's not really a recap, but more just some yapping about a thread that's been nagging at me for a few days.
The duration piece is due to not having a good revision style. Over time, my ability to write has slowed further and further, because I take my time revising sentence-by-sentence as I write them. It's unsustainable, frankly, although—adding this message on my proofreading pass—these notes were probably my quickest to draft, at about an hour and a half. So there's improvement!
All of this to say: I'm going to continue writing these journals, because it makes me happy when I write something that resounds with others. However, I'm going to spend some time reflecting on their structure, and find ways to take notes during the week that might improve these journals as publicized work logs.
Coding with an Agent (AKA progress on Leadlight)
Let's get to the real stuff that happened last week. I made some progress on Leadlight (previously Artwall), my Jetstream-based, media-centric Bluesky client. I'm finally at the point where I can render posts to the webpage, so hopefully I'll have a screenshot to share this week.
Why is this section about coding agents? Well, I spent some time working with an agent on the specifics of state machine composition. State machines (with XState) are usually the easiest way for me to conceptualize how my programs work, so I reach for them first. They make fixing/catching bugs pretty easy, but they can be a pain in the ass to refactor—hence, I wanted to have an agent with me.
I'm still struggling to find a way to work with a coding agent that maintains my interest and attention, and doesn't feel like I'm wasting my time. In this attempt, I instructed a Letta agent to write only one function at a time, and to warn me if I asked it to do anything too substantial.
This was to help me maintain my understanding of the code. When getting to an MVP, I want to know the why of every file and line—so, I figured when working with an agent in the early stages, it would help to make sure it dangled every line in front of me before they were written to disk.
This approach worked for a couple hours, until we ended up in the weeds on a machine design that didn't matter at the time (it was like two steps down the line), and I didn't catch on to that fact until my interest was waning.
Ideally, I would have the realization "Communicating this to the agent is taking too long, I have a design in mind and I should just implement it myself." Then, I'd do it! Unfortunately, I got frustrated by the time lost on the theoretical discussion, executive dysfunction kicked in, and I had to close out of the terminal for the day. Sigh.
I have a long way to go to determine how much or how little I want to have a coding agent involved in my workflow. They're a double-sided blade: With one edge, they're a way to guide me to releasing a project, even when I'm losing interest. With the other, they can completely torpedo my interest in a project with a bad interaction, and result in shelving that project indefinitely.
I was going to share some other thoughts on standardized atproto lexicons, but I think I've yapped enough for the week. Long story short—standardized lexicons are cool, but the ecosystem has a lexicon discovery problem.
That's for another time, though. Thanks for reading this far—I hope you have a great week!