rng.md

I’ve been thinking about my recent post, Maybe It’s Time To Start Looking at Local Coding Models.

Last year, I spent some time building SlateDB’s deterministic simulation tests (DSTs).

I like Obsidian’s open ecosystem and flexible feature set, but I really love Bear’s user interface.

This weekend, I experimented with AI image generation for this blog.

SlateDB occasionally has a flaky test failure–one that occurs randomly.

I used to think that AI would end up writing code in a language that had formal methods properties.

My friend is vibe coding a “personal OS” for their new job.

I have mostly been writing code like this lately: Spend a lot of time in plan mode to generate a truly good plan.

I’ve had a series of notable conversations with friends over the past couple of months.

I use lizard.py along with a PROMPT.md to keep AI-generated code from getting too complex.

This is how I’m currently writing most of my code: Copy code # Run codex in a simple loop against PROMPT.md. # # Behavior: # - Reads input from PROMPT.md on each iteration

Several people have asked about my experience with UniFFI after landing SlateDB binding work in PR #1462.