Core Principles
Context Engineering Beats Prompt Engineering
Good context + sloppy prompt beats bad context + perfect prompt.
Henrik Kniberg's insight from building with agents: good context + sloppy prompt beats bad context + perfect prompt. The spec isn't instructions for an agent to blindly follow. It's the context layer that lets the agent make good decisions autonomously.
In LDD, specs are:
- Thin (one hypothesis per spec, not a feature catalog)
- Testable (clear acceptance criteria that prove the hypothesis)
- Disposable (will likely be wrong; that's the point)
- Contextualized (include why this matters, what we're learning)