#1 Prompt vs Context Engineering The Missing Pieces Explained

Added: 2026-05-25

[00:00:00]The problem is the three pieces of information that are missing. The user's location, the conference actual location, and the disambiguation logic.

[00:00:08]Let me give you an example that made this click for me. Imagine you're building a travel agent. Your users type, "Book me a hotel in Paris near the Devops conference." Your agent, working from the prompt alone, sees the word Paris and then sees the word Devops. It has no idea whether you mean Paris in France, or Paris in Kentucky, or Paris in Texas. It has to guess. And maybe 1 in 20 times, your user ends up booked into a Best Western hotel in Kentucky, when in fact they wanted to be in Paris.

[00:00:36]Now, you could fix this by using a better prompt. You could try something like writing, "Always book ho- hotels in major international cities." But what happens now if you have a user who actually wants to have a hotel in Paris, Texas? You could then try to do something like confirm the city before the booking, but that defeats the whole purpose of the agent in that you want the agent to be as autonomous as possible. Finally, you could try something like, "Use the user's home location to disambiguate." But the problem here is that your model doesn't have this information that is not part of the prompt, so this would need to be part of the context. And this is even a very simple example. Now, imagine what happens when we're trying to do something just a tiny bit more complicated than booking a hotel.

[00:01:22]The amount of pieces you need to put together to have a good, accurate context that lets the agent be autonomous when it needs to and ask questions when it can't do the thing autonomously is a really tough problem.

[00:01:35]This is the moment that the lesson lands. The problem isn't the prompt. The problem is the three pieces of information that are missing. The user's location, the conference actual location, and the disambiguation logic.

[00:01:47]No amount of phrasing fixes that. You have to engineer the context to include these things in a way that the agent can work autonomously when needed and ask questions when needed. This is the move from prompt engineering to context engineering in one example. Once you see that, you can't unsee it. And hopefully, this will land the message for you.

[00:02:07]You'll understand the difference between the two, and make sure that the idea stick.