Context Engineering vs Prompt Engineering: Differences and Debugging Methods

On November 11, 2026, a post titled “Analyzing LLM Inputs: Deciding Between Prompt, Context, RAG, and Memory” appeared on Juejin. The article uses a single document comparison task as a shared testbed to sort out the application scenarios of four ways of supplying information to an LLM: a Prompt, the current input, external retrieval (RAG), and long-term Memory. Rather than describing the four in abstract terms, it asks a practical question for developers: when the model has to answer a question, which kind of information should you supply?

The post frames its guidance around an offline example built on Open ETF Agent, using the offline setting to show how the decision plays out when no live request is being served. A document comparison task is used as the running example because it requires distinguishing between what the task explicitly demands and what the model already knows from its parameters, which is precisely the split between Prompt, Context, RAG, and Memory. Developers are encouraged to trace, for a given question, whether the missing piece belongs to the task instructions, the current exchange, retrieved material, or recalled long-term history, and to add only that class of information. No other reports on this topic have been found so far, so no subsequent developments can be documented.