The science and practice of managing, optimizing, and engineering context for AI coding agents.
MCP (Model Context Protocol) servers extend AI coding agents with new tools and data sources. They're the plugin system that makes agents truly powerful.
Three approaches to code indexing for AI: embeddings, dependency graphs, and RAG. Each has trade-offs in accuracy, token efficiency, and maintenance cost.
RAG retrieves relevant code from your codebase before the AI generates a response. But vector-based RAG misses structural relationships that matter for coding.
Loading more files into the context window doesn't improve AI output — it degrades it. Quality context with 5 relevant files beats 50 random ones every time.
A reproducible framework for benchmarking AI coding context engines across codebases, tasks, and session lengths, with vexp vs manual context as a worked example.
65-80% of input tokens in typical Claude Code sessions are irrelevant. Here's where they come from, how to measure them, and how to eliminate them.
Using Cursor, Claude Code, and Codex? Each tool starts from zero every session. Here's how to build shared context across AI coding agents — and why it matters.
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.
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.