My new app has a name. Classis
Posts in "Coding"
Slice Done
The author prefers GLM-5.2 over GPT-5.6 Sol due to GLM’s transparency and ability to provide contextual information, which fosters trust and efficiency in evaluating the model’s outputs.
Don't Prompt, Loop
The text critiques the concept of “don’t prompt, loop” in AI-assisted development, emphasizing the importance of thorough planning and clear specifications over vague goals to ensure quality and intentionality in software creation.
I spent the last week nailing down the specs of a new project. Nearly there. Maybe a day or two more to nail down the last few bits. Getting antsy to build.
Mittens and Orbs
Dashboards and GUIs hinder understanding and effective interaction with AI agents, while a direct terminal interface provides clarity and control.
The Architect and the Builder
Automating the coding process through AI loops overlooks the critical thinking and design needed in software development, ultimately leading to inefficient and unsatisfactory results.
It’s almost funny how quickly Fable 5 blasts through my token limit when I have it analyze my codebase with ultracode. Hint: it is measured in seconds.
The Dashboard Graveyard
Abandoned dashboards illustrate a struggle with maintaining interest in complex projects, revealing a preference for simple, functional tools that address immediate needs.
I am back to enjoying LLM coding again. I had switched to GSD-pi and it suffered from a ton of AI slop PRs lately. I just did not enjoy it anymore, but was a bit stuck using it in the middle of a project. That’s done, I switched to vanilla Pi Coding Agent and am so much happier.
The Glitch Diary
A new blog series titled “The Glitch Diary” will candidly explore the challenges and failures of working with AI systems in daily life, highlighting the learning that emerges from mistakes.