[ AUG 10, 2026 ]
The Eye of Sauron (Metaphor for AI)

Claude code does amazing things. It also does blindly stupid things. Things that I wouldn't expect a first year CS student to do. Understanding how and why it does these things requires an understanding of attention, the transformer architecture and machine learning in general. I'm not going to cover those things here, but if you have a cursory understanding of those things you'll know what I mean when I say that Context is King.
LLMs have a lot of general knowledge baked into them that they rely on for much of what they do, but they augment this general knowledge (encoded in the 100's of billion or whatever parameters of the model) with context about what they are concerned with right at this moment - the conversation they're having, the code they're looking at - and gaps in that context often lead to unexpected outcomes.
Much like the Eye of Sauron from the Lord of the Rings was incredibly powerful, but failed to notice a couple of out-of-place hobbits walking an all-powerful ring straight into it, LLMs will fail to notice very obvious things if you don't point them at the right hobbit.
They'll apply some general knowledge, but get the specific details wrong. They might use a completely different form approach in a React component than the rest of the app, or reimplement a variation of something that already exists in shared. Solving this problem is one part of Context Engineering.
Claude Code does a decent job of searching for existing patterns and reviewing existing code, but as codebases grow it's necessary to point the eye in the right direction with nested CLAUDE.md or AGENTS.md files that help focus attention. If there's an existing example or pattern you want it to copy, especially if it's somewhere else in the codebase, point it at it or ask it to find it before starting on work.
If there had been an Orc jumping up and down pointing at the two hobbits, the Eye of Sauron would have been able to muster its abilities a lot better - remember that when you're working with Claude and things will go a lot smoother.
AUG 10, 2026
Metaphors for AI Development
I've always had a soft spot for a good analogy or metaphor. I've found that there's a massive amount of understanding that can be communicated succinctly with...
AUG 5, 2026
The 3x3 Rule of AI Assisted Software Development
The goal of AI Assisted Software Development, as I see it, is to figure out how to make a repeatable, consistent process out of building software with LLMs. To...