00:00:00 You understand the machine now, the loop, what the agent actually sees, how it searches, and how its window fills up.
00:00:10 And you've also got the vocabulary from the last module, tools, skills, MCP, subagents.
00:00:17 Basically, that's how the Agent works.
00:00:20 Now, we go all the way back to Module 1. the very first problem I named, and we'll fix it directly.
00:00:28 Remember the two root causes of every AI coding failure we talked about.
00:00:33 No real plan and no real context.
00:00:36 I've been circling around that second one the entire time, but in this lesson, we will go straight to it.
00:00:43 The trap that I have seen all developers fall into is that when the output is bad, people reach for the same tool every time.
00:00:52 Prompt engineering.
00:00:53 Word it better, add more detail, find the phrasing.
00:00:57 There are whole guides built on the idea that the prompt is the lever, that if you could just say it right, the agent would get it.
00:01:07 And I want to reframe that for you because it's the wrong lever.
00:01:11 Think about what actually happens when you hit enter.
00:01:14 Your prompt doesn't land in a vacuum.
00:01:17 it lands on top of everything the agent already knows in that moment.
00:01:22 The files it's read, the conversation so far, whatever's loaded into its always-on setup.
00:01:27 That whole environment is the context.
00:01:30 And your prompt is only one small thing arriving into it.
00:01:34 and the quality of what comes back depends far more on that environment than on the exact wording of the one line you typed.
00:01:43 I think I talked about this in module two as well.
00:01:46 the agent only knows what's been loaded.
00:01:49 So, watch what that means.
00:01:52 write the most beautiful, detailed prompt in the world and send it into an empty context, it still has to guess stuff because you gave it perfect words
00:02:01 and nothing to work with.
00:02:03 Now, if you flip it, send a rough, plain prompt into a context where the right file, the convention, and the constraint is already loaded and gets it right
00:02:13 the first time.
00:02:14 Same agent, even the same model, but the difference wasn't the prompt.
00:02:18 It was what was sitting around it.
00:02:21 So the real skill was never writing better prompts.
00:02:25 It's designing what the agent knows before it ever reads a prompt at all.
00:02:30 That's context engineering.
00:02:32 And that word, engineering, is doing the work here.
00:02:36 This isn't wording, it's design.
00:02:39 You're deciding, deliberately, ahead of time, what gets loaded and what doesn't.
00:02:46 What the agent should always know and should only reach for when it needs it.
00:02:51 Or what it should never be carrying around.
00:02:53 It's about systems architecture.
00:02:55 Because an architect doesn' t just hope the crew figures out from a vague brief.
00:03:00 They set up everything that's known before the first brick ever goes down.
00:03:05 So context engineering is how you actually do that for an agent.
00:03:10 And to be clear, prompting still matters, but it's the last and smallest step in the chain.
00:03:16 The leverage lives upstream in context, and that's what I'll teach you in this module.
00:03:22 Over the next couple of lessons, the different kinds of context and which one to use when, why the always-on kind has to stay ruthlessly small,
00:03:32 because every line of it costs you every single session.
00:03:35 How these files actually load And the one mistake that silently loads nothing at all.
00:03:41 Oh, and one lesson on structuring your code so the agent works better because your own code is context too.
00:03:48 Every failure that I named back in module one, when I talked about why AI coding falls apart and how you should direct the AI and not hope,
00:03:58 trace back to a context decision that somebody didn't make on purpose.
00:04:03 So this is where you learn how to make these decisions on-purpose.
00:04:08 Now let's start with the three types of context.
