Technical writing on AI coding agents, context engineering, and developer productivity.
Benchmark results from 21 runs on a real FastAPI project: 65% fewer input tokens, 57% lower cost, 14pp better task completion. Full methodology and setup guide.
Stale context causes AI coding bugs that look like hallucinations but aren't. Here's why it happens, why it's getting worse, and how to detect it.
Compare manual Claude Code token-saving tactics with automated context engines like vexp, with real numbers on savings, tradeoffs, and setup.
Prompt engineering vs context engineering: which actually saves more tokens? Real numbers show context engineering delivers 10x the cost reduction — here's why.
Why Claude Code sessions degrade over time, what’s really filling your context window, and how vexp-style context engineering fixes it.
Most tokens in a typical Claude Code session are exploration overhead and history accumulation — not useful work. Here's the data and what to do about it.
Claude Code starts every session with zero memory of previous work. Here's how to build a three-layer memory bank - CLAUDE.md files, dynamic mem dir, automated session memory with vexp.
Every AI coding session starts blank. You re-explain the architecture, the constraints, the decisions. Here's what persistent memory changes and how it works.
Without a code graph, AI agents navigate blind and read 15 files to answer a 3-file question. Here's why it happens and how graph-based context selection fixes it.