2026-01-29 Thu

• Person/Steve Yegge

  • Updated with Beads, Gas Town, and Medium (Jan 2026).
  • Person/Steve Yegge/Blog/Welcome to Gas Town
    • I really enjoyed this essay. I listened to it once last weekend and it's been on my mind ever sense. I need to sit down and read it in more detail because it's dense and I keep hearing references to it. Welcome to Gas Town. For example, I heard it discussed in Inventing the Ralph Wiggum Loop | Creator Geoffrey Huntley - YouTube and Steve Yegge's Vibe Coding Manifesto: Why Claude Code Isn't It & What Comes After the IDE - YouTube, as well as Angie Jones on Ralphing 25k repos at Block, GPT-5.2 Codex, and CES weirdness - Dev Interrupted | Podcast on Spotify.

• AI/Coding

  • Claude/Code/Plugin/Marketplace/Report/26/01/Automated Testing
  • AI/Coding/Ralph Wiggum
    • You're Using Ralph Wiggum Loops WRONG - YouTube helped break it down at a nuts and bolts level.

      Basically:
      

    • create a spec with Person/Brian Madison/BMAD or GitHub/SpecKit or with a conversation with AI
    • create an implementation plan with AI that results in a bunch of checkboxes in markdown docs, github issues, or something like steveyegge/beads: Beads - A memory upgrade for your coding agent
    • then, in a loop

      
        a have the LLM pick the next best task
        b have the LLM do exactly that one task then mark it as complete
        c discard the context window and then move on to a fresh prompt
      

• GitHub/Desktop

  • Alternatives to GitHub Desktop

    • I'm finding that this takes me out of the CLI flow. Also, I can't use it in devcontainers.
      • Right now, I'm finding that it takes me too many clicks to just get the path to a repo, which is one of the most common things that I need to do with AI coding. It turns out there's no GitHub/Desktop/Keyshort for this.
    • Formerly, I've referenced these as possible github desktop alternatives:
      • gitopolis - GitHub Desktop–like but CLI and “more systematic” for a specific collection of repos (clone, run commands, organize with tags), useful for onboarding
        • See Person/Tim Abell/GitHub/cloner
      • Lazygit - GitHub abstraction layer (more of a replacement for JetBrains/Git) workflow in a TUI
        • not really a "repo manager" - gitopolis is better for that

• The AI/Engineer Mindset

  • Tip
    • I think there's a good chance that a strong component of why AI-assisted software engineering is the most effective application of LLMs is because software engineers have already deeply internalized the "inputs -> outputs" mentality, and instinctively don't think there's a "someone else" that's responsible for changing the behavior of the system. It's always the inputs that determine the outputs.
    • If you find yourself complaining about interactions with LLMs starting with the words "The AI XYZ ...," consider instead starting with the words "The way I'm prompting this, I'm noticing XYZ..." This helps mindset shift helps us see better see how to achieve our aims with LLMs.
    • There's a comparison here to relationships with other humans. If you say to another human, "you make me feel angry," that divests you of agency. If you say to another human, "when you do/say XYZ, then I often find that afterward I am feeling ABC," it opens up the discussion from a point of curiosity.