Context Engineering

Context engineering is the practice of deliberately designing everything an AI model can see, remember, and access during a working session. This goes beyond writing good prompts: it includes persistent instruction files, project documentation, tool configurations, and conversation history, all structured to keep the AI aligned with your goals across sessions. Where prompt engineering focuses on crafting individual messages, context engineering treats the entire information environment as a system to be designed.

The Simple Version

If prompt engineering is crafting the perfect question, context engineering is setting up the entire room before the conversation starts. It’s not just about what you say to the AI. It’s about what files the AI can see, what decisions it already knows about, what conventions it follows, and what tools it has access to.

Think of it like preparing for a strategy meeting. You could walk in and start talking. Or you could distribute a briefing document ahead of time, set an agenda, bring the right people into the room, and have relevant data on the screen. The meeting itself might take the same amount of time, but the quality of the output is completely different. Context engineering is that preparation, but for AI.

Why It Matters

Most AI frustrations come from context problems, not capability problems. The AI is smart enough to help. It just doesn’t know enough about your specific situation to help well. Context engineering fixes that gap.

It includes strategies like:

• Persistent files (PLAN.md, CLAUDE.md) that tell the AI about the project before the conversation starts
• Context pruning to keep only relevant information in the context window
• Offloading decisions to files the AI can reference, instead of trying to hold everything in one conversation
• Tool access so the AI can read your codebase, search for files, and understand the full picture

The term is broader than prompt engineering because it covers the entire information environment, not just the words you type.

How It’s Used on This Site

This site’s architecture is a context engineering system. PLAN.md provides the strategic context: what the project is, what’s been built, what’s next. CLAUDE.md provides the operational context: conventions, linking rules, tone, content workflows. Together, they ensure that Claude Code starts every session with the right information, even when the conversation history has been completely reset. The result: consistent output across dozens of sessions without manual re-briefing.