-
The Frontier Model Is Overqualified
Most of your token bill goes to summaries, tags, and routing, all billed at genius rates. Open-weight models do the cheap tier for cents.
-
Pay the Setup Tax Once
The model forgets your repo between every session, so someone re-teaches it the same facts forever. A committed, layered CLAUDE.md is how the repo briefs the model for the whole team and every session that follows.
-
Stratify: The Size of What One Person Ships Now
A polyglot static-analysis engine. Five languages and six analyses through one model, reaching your terminal, CI, editor, AI agent, and dashboards. I built it solo with the spec-driven loop. The headline is not the tool. It is how little it took.
-
Don't Make the Model Do a Build Step
An agent spends the first minutes of every session rebuilding a map of your codebase. A script already has that map. The model should read the facts, not re-derive them.
-
Conduit: The Plumbing Every RAG Team Rebuilds
A RAG application is 20 percent retrieval logic and 80 percent data plumbing, and every team rebuilds the same 80 percent from scratch. I built an open-source engine that owns it. Today it is public.
-
What Goes in the Prompt, What Goes in the Tool
An agent keeps its knowledge in two places, and most broken agents have something on the wrong side. One question decides the split: does this change between calls? No: prompt. Yes: tool.
-
The Context Window Is a Desk, Not a Memory
The model holds nothing between your messages. Every turn, the entire conversation is packed onto one working surface and read from scratch. Once you see the desk, long-chat costs, sudden forgetting, and product memory all stop being mysterious.
