WEBVTT
00:00:01.360 --> 00:00:04.639
Why does no one talk about prompt engineering anymore?
00:00:04.879 --> 00:00:13.439
I mean, if you were whined back like two years ago, you would have sworn that prompt engineering would be the world's most popular future job title.
00:00:13.759 --> 00:00:15.839
But that's obviously not the case.
00:00:16.079 --> 00:00:20.800
And the essential disappearance of that term is twofold.
00:00:21.039 --> 00:00:28.320
One, models are smarter, and it doesn't always matter the exact way we talk to them, as long as we get the message cross.
00:00:28.559 --> 00:00:37.280
And two, an output that moves the needle is much more dependent on business context versus just wording something a certain way.
00:00:37.520 --> 00:00:41.759
Hence the resurgence of the term context engineering.
00:00:42.240 --> 00:00:44.240
But what does that even mean?
00:00:44.479 --> 00:00:53.600
And how can you understand the required inputs of context engineering to get better outputs out of a large language model?
00:00:53.840 --> 00:01:00.000
Well, if that's one of the things that you or your business is grappling with, then you're in luck.
00:01:00.240 --> 00:01:10.560
Because on today's episode of our start here series, we're tackling context engineering and how to get expert level outputs from AI chatbots.
00:01:10.719 --> 00:01:13.439
All right, I am excited for today's show.
00:01:13.599 --> 00:01:14.480
I hope you are too.
00:01:14.640 --> 00:01:16.159
If you're new here, welcome.
00:01:16.400 --> 00:01:20.239
This is the everyday AI Start Here series.
00:01:20.400 --> 00:01:25.920
So after 700 plus episodes, one of the most common questions I get is where do I start?
00:01:26.079 --> 00:01:29.120
So that's why we started the Start Here series.
00:01:29.280 --> 00:01:33.040
And this is actually volume seven of this exact series.
00:01:33.200 --> 00:01:41.359
So the Start Here series is the essential podcast series to both learn the AI basics and to double down on your knowledge.
00:01:41.519 --> 00:01:43.840
So if that's what you're trying to do, you're in luck.
00:01:44.000 --> 00:01:47.680
Make sure you go to starthearseries.com.
00:01:47.840 --> 00:01:53.040
That's going to redirect you and give you free access to our inner circle community.
00:01:53.120 --> 00:02:11.840
So there you can not only go take our context engineering course called Prime Prompt Polish for free, but also network with a bunch of other people and you'll be redirected right to our Start Here Series area where you can go and listen to every single episode in this series, all right there at your fingertips.
00:02:12.240 --> 00:02:12.639
All right.
00:02:12.800 --> 00:02:23.360
And if you missed our last episode of this series, we talked about how to train your team on AI and the seven steps to educate your organization on large language models.
00:02:23.520 --> 00:02:30.240
And the last step in there was well, making that step to go from operator to orchestrator.
00:02:30.400 --> 00:02:35.439
So that's where we kind of left you with the last step in our series.
00:02:35.520 --> 00:02:37.039
And that's where we're going to pick up.
00:02:37.280 --> 00:02:46.240
Because actually, one of the biggest things that you can do from going from an operator or essentially someone pushing all the buttons to an orchestrator, right?
00:02:46.319 --> 00:02:53.919
Which is when AI starts to do the work for you, is having the right data and providing that data to the model in the right way.
00:02:54.000 --> 00:02:57.919
And that is the backbone of what context engineering is.
00:02:58.319 --> 00:03:12.080
It is the process a human goes through to make sure a large language model has the right context about not just you, your role, what you're trying to accomplish, but maybe most importantly, your business and the competitive market.
00:03:13.039 --> 00:03:21.360
So this is the big differentiator because the same two people, they can use the same prompt, right?
00:03:21.439 --> 00:03:27.759
Going back to uh prompting and prompt engineering, uh, and you can get wildly different answers, right?
00:03:27.919 --> 00:03:46.319
Because if one person, whether it's in custom instructions, a GPT, a project, et cetera, or uh using Chat GPT apps or uh, you know, claude uh connectors, whatever you may be using, Google Gemini apps, uh right, you can put in the exact same prompt as someone sitting next to you.
00:03:46.479 --> 00:03:55.759
And if you have your context engineering 101 ducks in a row, your output will be light years better than that person who does not.
00:03:56.080 --> 00:04:14.240
And the difference isn't necessarily as difficult as it may sound, because the skill separating average AI users from expert level ones is just providing the model the needed context and understanding how it works in different scenarios.
00:04:14.400 --> 00:04:16.079
So that's what we're gonna cover today.
00:04:16.160 --> 00:04:26.800
We're gonna well first talk about why the AI industry as a whole and really just the business landscape has shifted away from prompt engineering and really just more focusing on context engineering.
00:04:27.040 --> 00:04:32.160
And I will say that that shift kind of happened and popularized in late 2025.
00:04:32.240 --> 00:04:39.279
Then I'm gonna lay out for you a six-part framework and also a four-layer system for structuring what your AI sees.
00:04:39.360 --> 00:04:50.399
And then I'm gonna walk you through how to build reusable context vaults or kind of skills one and the same and use platform features that are already at your fingertips.
00:04:51.040 --> 00:04:53.120
All right, let's get into it.
00:04:53.360 --> 00:04:57.120
So RIP prompt engineering, all right.
00:04:57.279 --> 00:05:01.279
Uh kind of, but here's here's the thing.
00:05:01.439 --> 00:05:16.480
Um, if you think back to the early days of Chat GPT, or even technically before Chat GPT, right, when the GPT technology was available uh to dozens and eventually hundreds of other platforms before ChatGPT even came out, right?
00:05:16.800 --> 00:05:23.839
So much of what you were able to get out of a model was how you talk to it.
00:05:24.079 --> 00:05:25.920
And that was for a couple of reasons.
00:05:26.079 --> 00:05:31.360
Number one, the AI models themselves were a little more old school, uh, right.
00:05:31.439 --> 00:05:33.439
And I'll talk about that here in a little bit.
00:05:33.600 --> 00:05:36.800
Uh, you couldn't always upload documents, right?
00:05:36.959 --> 00:05:43.519
And you couldn't always paste in a bunch of information either, because the model's context windows were smaller.
00:05:43.600 --> 00:05:46.959
So essentially, it would always forget things very quickly.
00:05:47.120 --> 00:05:51.759
Uh, you couldn't, for the most part, upload documents, at least not very easily.
00:05:51.920 --> 00:05:52.399
Uh, right.
00:05:52.480 --> 00:05:59.439
So this really changed that uh differentiator of if you were gonna get a good output versus a bad output.
00:06:00.079 --> 00:06:03.279
And it was really just how you talk to the model, right?
00:06:03.360 --> 00:06:10.240
If you use certain prompting techniques, uh you could kind of pull the best uh out of a model's training data, right?
00:06:10.319 --> 00:06:16.319
Because even if you think back to the way earlier days of large language models, they weren't connected to the internet, right?
00:06:16.480 --> 00:06:20.720
So they weren't connected to the internet, they didn't have tool calling, uh, right.
00:06:20.800 --> 00:06:28.639
And for the most part, you couldn't even upload files, which is why in the earlier days, prompt engineering was actually really important.
00:06:29.360 --> 00:06:32.399
Because even in that data set, right?
00:06:32.480 --> 00:06:39.519
Um, and if you go back and listen to the earlier uh episodes of our start here series, we talk about training data and everything that goes into it.
00:06:39.600 --> 00:06:54.800
Uh, but for the most part, in the earlier days of Chat GPT and you know, when Gemini was barred and you know, early copilot days, etc., it was really how you talk to the model because the model had so fewer capabilities.
00:06:55.040 --> 00:06:59.199
Yet there was still a lot of information there, right?
00:06:59.439 --> 00:07:07.839
Even the early models like GPT-3 or GPT-3.5, uh, or you know, the first you know version of Google Gemini.
00:07:08.160 --> 00:07:13.199
Um, even though we look back at those models now and we think, oh, they weren't very good, they were.
00:07:13.519 --> 00:07:16.399
You just really had to learn how to talk to it, right?
00:07:16.560 --> 00:07:24.959
So now you can go talk to any of today's smartest models, and you don't even have to really put a sentence that makes sense, right?
00:07:25.040 --> 00:07:31.759
Sometimes I find myself, you know, if I'm not using voice dictation with models and I'm just and if I'm just typing, right?
00:07:31.920 --> 00:07:35.759
I've become and I've probably programmed myself to know the models are so smart.
00:07:35.920 --> 00:07:46.639
You know, it's misspellings and I say the wrong thing and all these things, but I know in the end it doesn't matter because the today's models are so incredibly smart at understanding what I'm trying to say, right?
00:07:46.720 --> 00:07:50.800
Especially when I have personalization enabled, memory enabled, all of these other things.
00:07:51.040 --> 00:07:56.319
The actual words that I'm telling a model don't mean a ton.
00:07:58.319 --> 00:08:00.480
Now, right?
00:08:01.040 --> 00:08:03.680
That's how it is, but it isn't how it used to be.
00:08:04.000 --> 00:08:08.959
Because the prompt engineering used to focus on you had to say things the exact right way, right?
00:08:09.120 --> 00:08:16.720
But if you did, what you could get was just a step change different than what you could get if you didn't word something the direct way.
00:08:16.800 --> 00:08:20.959
It was like you almost had a password that no one else had and at the time, right?
00:08:21.120 --> 00:08:25.920
Proper prompt engineering, it was an amazing skill to have.
00:08:26.639 --> 00:08:31.040
But right now, it doesn't matter as much, right?
00:08:31.279 --> 00:08:37.039
You can have the best prompt engineering skills in the world, but if you don't have the context, it doesn't matter.
00:08:37.279 --> 00:08:43.360
And I think the industry really realized that the bottleneck was never about how we talked to the model, it was the information behind it.
00:08:43.600 --> 00:08:50.399
And I think that this shift really started to happen uh in probably mid-June of 2025.
00:08:50.559 --> 00:09:04.559
So, you know, two of the people that are kind of credited uh with popularizing this concept of context engineering were Shopify CEO Toby Ludke, um, who just kind of called for the move from prompts to context.
00:09:04.639 --> 00:09:10.320
And then in the same month, former OpenAI co-founder Andre Kapathy kind of endorsed that term publicly.
00:09:10.399 --> 00:09:19.840
And then in September, uh Anthropoc even published published a dedicated blog talking about moving from moving away from this concept of prompt engineering.
00:09:20.159 --> 00:09:22.720
And I will just go ahead and say this, right?
00:09:22.960 --> 00:09:30.720
Not one of those I told you so, uh, but we've been teaching this concept of context engineering, even though I didn't call it that.
00:09:30.879 --> 00:09:34.159
Um, I believe we started teaching it in late 2023.
00:09:34.399 --> 00:09:45.039
Um, so yeah, we've I've done more than uh, you know, 200, uh probably like 210 now uh live trainings on, you know, essentially prompting models, right?
00:09:45.200 --> 00:09:54.240
So even early on, right, I was teaching prompt engineering, but that really shifted uh in late 2023 and early 2024.
00:09:54.399 --> 00:10:04.159
You know, we kind of, you know, we have our prime prompt polish course, but if you have taken it in the last two and a half, three years, you know that we started to teach this concept called refine cue.
00:10:04.320 --> 00:10:07.039
And essentially that is context engineering at its core.
00:10:07.200 --> 00:10:10.720
So, you know, it's it's it's not a a new concept, right?
00:10:10.799 --> 00:10:12.879
Because I've been teaching people this for a long time.
00:10:13.120 --> 00:10:22.960
But the term, uh the terminology around context engineering, and it has really uh snowballed into more of a movement, uh, has really picked up in the last year or so.
00:10:23.200 --> 00:10:27.519
And that's because it's about designing the environment and not just the question.
00:10:27.759 --> 00:10:31.039
So an easy way to think about this is to think of the AI, right?
00:10:31.120 --> 00:10:38.399
Think of that whatever large language model that you that you're using as a processor and the context window is its working memory.
00:10:38.480 --> 00:10:43.440
Okay, so context window without getting uh too technical, that's like a hard drive, right?
00:10:43.600 --> 00:10:48.080
So in the same way, maybe you have a one terabyte hard drive, let's just say, right?
00:10:48.159 --> 00:10:55.600
If your hard drive's full, then if you try to put more information, maybe fortunately a computer will stop you from doing that.
00:10:55.759 --> 00:10:57.440
A large language model will not.
00:10:57.519 --> 00:11:04.320
So once its quote unquote hard drive gets full, it's just gonna delete the first file that you ever uploaded, right?
00:11:04.399 --> 00:11:12.320
So that's how a context window works with large language models, except you never really know when you are hitting that context window, right?
00:11:12.480 --> 00:11:24.799
So what this means is well, your context becomes very important and understanding that working memory and how the large language models work specifically with your data, right?
00:11:24.960 --> 00:11:28.000
That's grasping the basics of context engineering, right?
00:11:28.080 --> 00:11:29.519
And it's it's a little different, right?
00:11:29.600 --> 00:11:35.919
So it's different depending, uh, it does get a little convoluted depending on what large language model you're using, right?
00:11:36.000 --> 00:11:44.480
As an example, if you're using GPT-5.2 versus GPT-5.2 thinking versus if you're using GPT-5.2 Pro via the API.
00:11:44.639 --> 00:11:46.799
So it is, it is a little bit different.
00:11:47.120 --> 00:11:51.919
But essentially, the concept of prompt engineering it ends once you hit enter.
00:11:52.080 --> 00:12:02.879
And context engineering is an ongoing battle to make sure that your uh model or your session with the model has access to and can understand your business data.
00:12:03.039 --> 00:12:09.200
Um, and there was a study from Intuition Labs last year that said that 40% of AI projects fail.
00:12:09.360 --> 00:12:15.519
Uh and one of the reasons or the main reasons why is well, it comes from poor context and it's not actually the model.
00:12:15.679 --> 00:12:25.279
It's the model either number one, not understanding what you want out of it, or number two, it just doesn't have that data that can be the differentiator.
00:12:26.399 --> 00:12:30.240
So here's why it matters more than ever.
00:12:30.480 --> 00:12:41.600
Well, uh, if you talk about the big three or the big four, uh, right, so ChatGPT, Anthropics Claude, Google Gemini, uh, and Microsoft Copilot, they it used to be fairly hard.
00:12:41.759 --> 00:12:51.519
Aside from Copilot, it's always been uh more straightforward if you understand the uh the tech and the permissions uh landscape uh inside Microsoft Windows.
00:12:51.679 --> 00:12:52.799
That's a whole nother story.
00:12:52.879 --> 00:12:56.799
Uh but let's look at the other three with just ChatGPT, Claude, and Gemini.
00:12:57.200 --> 00:13:04.799
Um, I would say in early 2025, it was actually kind of difficult to use your company's data, right?
00:13:04.879 --> 00:13:14.720
You could even say, oh, well, Jordan, you know, there's been these things like GPTs where you know you could upload, you know, documents and have a specialized version of ChatGPT that had access to those documents.
00:13:14.879 --> 00:13:16.240
Yeah, but have you ever tested it?
00:13:16.399 --> 00:13:21.200
Have you ever run the needle in the haystack test on that GPT that has right?
00:13:21.279 --> 00:13:28.639
People would just assume, oh, I'm gonna, you know, upload a you know, 500-page PDF into a GPT and then it knows everything about me and my company.
00:13:28.720 --> 00:13:30.000
No, absolutely not.
00:13:30.240 --> 00:13:30.480
Right?
00:13:30.639 --> 00:13:37.360
That means you didn't really understand uh, you know, how uh these models tokenize that information or access that information.
00:13:37.519 --> 00:13:45.200
But now it's much easier, uh, because these models by default can create searchable indexes of your files.
00:13:45.360 --> 00:13:49.840
So think of it like this way: in the same way, let's say you have a Mac, right?
00:13:50.080 --> 00:13:57.840
And you can go to your Mac Finder and your search bar there, and you can search for maybe a word that is within a PDF.
00:13:57.919 --> 00:14:02.240
And it's gonna know because it's indexed that file and it understands it, right?
00:14:02.320 --> 00:14:07.279
That obviously requires a certain level of compute, right?
00:14:07.519 --> 00:14:10.639
That two years ago these models just didn't have, but now they do.
00:14:10.799 --> 00:14:18.240
So within ChatGPT, as an example, they have things like projects, they have things uh that were previously called connectors, now they're called apps.
00:14:18.320 --> 00:14:19.919
Uh Claude, same thing.
00:14:20.000 --> 00:14:22.720
Uh, you can have, you know, project-specific memory.
00:14:22.799 --> 00:14:27.679
Uh, you can have these reusable skills that can take advantage of your business context.
00:14:27.759 --> 00:14:29.440
Uh, same thing with Google Gemini.
00:14:29.679 --> 00:14:34.480
You can have these gems that connect live to your Google Drive, to your Gmail, to your calendar, right?
00:14:34.799 --> 00:14:47.759
So at various levels and in different ways, the three main players within a couple of clicks only can connect to your, in many cases, your dynamic business data.
00:14:47.840 --> 00:14:49.679
It's not always dynamic, uh, right.
00:14:49.840 --> 00:14:56.559
In some cases, it is, and it will create a searchable and live indexed of everything that you connect to it.
00:14:56.639 --> 00:14:56.960
All right.
00:14:57.120 --> 00:15:00.159
Again, I have to throw out that same asterisk as I always do.
00:15:00.240 --> 00:15:04.960
You know, always make sure you have permission to connect your company's data to a large language model, blah, blah, blah.
00:15:05.039 --> 00:15:05.679
Uh, right.
00:15:05.919 --> 00:15:19.519
Um, but once you do, that is the first step in context engineering is making sure, number one, the model has access uh to the context or the data that it needs.
00:15:19.759 --> 00:15:25.759
But you also, more importantly, need to understand how it works in each scenario, right?
00:15:25.919 --> 00:15:33.200
Like as an example, I would say most people, even if you listen to the show often, you might not know that ChatGPT had these things called connectors.
00:15:33.279 --> 00:15:39.360
But oh, wait, actually, in December, they changed them right before the holiday season, they changed them all to apps.
00:15:39.440 --> 00:15:40.799
And a lot of people miss that.
00:15:40.960 --> 00:15:49.600
And with that comes, uh, depending on what app you're talking about now, well, it maybe handles um your data a little bit differently than it did before.
00:15:49.759 --> 00:16:02.240
So you do have to, depending on the platform that you use, if you really want to understand context engineering, well, you have to understand how each of these different platforms connects to different data sources because it's not it's it's not uniform, right?
00:16:02.320 --> 00:16:03.840
It's really not, right?
00:16:04.000 --> 00:16:13.600
Even if you look at Chat GPT, uh their apps, there's four different ways uh that it can look at your data in four different ways that it can understand your data, right?
00:16:13.759 --> 00:16:25.039
In the same way, uh, right, if you think of cloud storage and there's these different permissions and different ways to access data, you know, that does uh trickle down to large language models as well.
00:16:26.000 --> 00:16:32.559
So uh let's talk about kind of what I'm calling these six building blocks of effective AI context.
00:16:32.879 --> 00:16:40.720
Because the step one is well, what I just covered, you have to make sure, depending on what large language model you're using, it has access to your data.
00:16:40.879 --> 00:16:42.960
Uh, but it's not just about data.
00:16:43.039 --> 00:16:46.960
And it's not just about telling a large language model, here's something about me, right?
00:16:47.039 --> 00:16:48.399
That's providing context, right?
00:16:48.480 --> 00:16:49.279
Context clues.
00:16:49.440 --> 00:16:52.320
Tell me more about yourself, what you're trying to accomplish, right?
00:16:52.399 --> 00:16:58.000
I'd like to break it down into these six different building blocks for building context within a context window.
00:16:58.080 --> 00:16:58.399
Okay.
00:16:58.559 --> 00:17:07.200
Once you're out of the context window, again, depending on how you're connected, uh, connecting your data, um, you might start to get poor results.
00:17:07.359 --> 00:17:13.119
So keep this in mind and keep these six building blocks in the context window of any given conversation.
00:17:13.359 --> 00:17:17.680
So goal, that's what you need the AI to produce and for whom.
00:17:17.920 --> 00:17:22.799
Constraints, understanding the boundaries, rules, things to avoid, and format requirements.
00:17:22.960 --> 00:17:27.680
Uh reference material, that's the approved facts, data, and source documents to draw from.
00:17:27.839 --> 00:17:33.920
Examples, those are representative samples of the output you want, plus context and examples.
00:17:34.079 --> 00:17:43.599
Um, then procedures, those are step-by-step instructions for how the AI model should approach the tax, uh, task, and then the evaluation rubric.
00:17:43.759 --> 00:17:48.559
So that's grading criteria so the AI can assess its own output quality.
00:17:48.880 --> 00:17:52.960
So this isn't a perfect formula and it is ever changing, right?
00:17:53.119 --> 00:18:01.440
Uh, but I think this is for the most part, number one, you should just go take our free prime prompt polish course and go through the whole thing.
00:18:01.519 --> 00:18:12.799
Uh, I think Refine Q is another version uh of building this essential uh building blocks of context, but this is another um kind of framework that I think works really well.
00:18:13.519 --> 00:18:20.559
Okay, now you don't just think of those six building blocks of okay, uh I'm good, right?
00:18:20.640 --> 00:18:24.960
Let me get those things because you have to actually apply them in different layers.
00:18:25.200 --> 00:18:27.599
So the first layer is well, personal.
00:18:28.000 --> 00:18:30.720
That's your own personal context.
00:18:31.039 --> 00:18:34.319
The second layer is your team, right?
00:18:35.440 --> 00:18:41.039
If you're on a small team, might be a little less difference between that first layer and the second layer.
00:18:41.200 --> 00:18:45.599
If you're on a large team, could be a huge difference.
00:18:45.759 --> 00:18:50.000
Then the third layer is obviously your company or your business, right?
00:18:50.079 --> 00:18:53.920
So those are things like your brand voice, your policies, your product details, et cetera.
00:18:54.079 --> 00:18:54.559
Uh, right.
00:18:54.640 --> 00:18:59.680
When we talk about layer one, that's your own personal role, your expertise, uh, right.
00:18:59.839 --> 00:19:04.240
Layer two, uh, that's kind of shared definitions, uh, your project goals, conversations.
00:19:04.319 --> 00:19:08.240
Like I said, layer three, that's at the company level, brand voice, policies, et cetera.
00:19:08.319 --> 00:19:10.160
And then number four, that's your market.
00:19:10.400 --> 00:19:16.799
That's the your position in the competitive market, uh, industry, invite uh insights, trends, et cetera.
00:19:17.039 --> 00:19:31.039
So it's not just about bringing the right folder in via a Chat GPT app, it's not just about those building blocks, it's also making sure that you apply those at the different layers that a large language model needs.
00:19:31.839 --> 00:19:37.200
And then when you do that, you can probably imagine by now, oh my gosh, this sounds extremely time consuming.
00:19:37.279 --> 00:19:38.319
Yeah, it is.
00:19:38.559 --> 00:19:38.880
Right.
00:19:39.039 --> 00:19:45.920
I always tell people if you're thinking about AI as if it's an easy button, you're looking at it all the wrong way, right?
00:19:46.160 --> 00:20:04.079
The best thing, the best analogy that I've probably ever taught, and it's from the very first, right, in early 2023, when I did my very first uh, you know, Chat GPT prompting course back when prompt engineering actually was a thing, and it still holds true to today.
00:20:04.559 --> 00:20:10.000
When you are working with a large language model, you have to think of it as you are training a new employee, right?
00:20:10.160 --> 00:20:16.160
A new college grad or someone that just transferred in and they're a capable person, right?
00:20:16.640 --> 00:20:22.559
But their output is going to largely be dependent on how much context you share with them, right?
00:20:22.960 --> 00:20:26.000
Because if you just throw down a giant prompt, right?
00:20:26.079 --> 00:20:34.640
If you throw down a giant training manual and then say, all right, first assignments due in an hour, they're gonna fail, right?
00:20:34.799 --> 00:20:40.400
You have to go through and give them the context and the conversation and the iteration that they need, right?
00:20:40.480 --> 00:20:50.079
And one of the ways that you can think about this, because what I just laid out in terms of uh, you know, applying the context, okay, those six different building blocks across four different layers, that's a lot.
00:20:50.240 --> 00:20:57.839
Well, you have to think in the same way that you would invest in an employee, you do it so in the long run, it is scalable, repeatable, reusable.
00:20:57.920 --> 00:20:58.319
Uh, right.
00:20:58.400 --> 00:21:03.200
So you can think of this as creating a context vault or maybe a skill, right?
00:21:03.279 --> 00:21:14.880
That's another terminology, uh, another term, uh, you know, kind of uh created by anthropic, but skills uh have been uh picked up by, you know, all the main players in the AI space.
00:21:14.960 --> 00:21:19.759
So think of a vault or a skill as kind of this folder of reusable context, right?
00:21:19.839 --> 00:21:28.799
So these different procedures, rubrics, key facts that you might be able to reuse or uh to modularly use uh with uh within each other.
00:21:28.880 --> 00:21:37.920
Uh so you can build a skill or a vault per role to start, then expand that as your team standardizes whatever process you're working on.
00:21:38.079 --> 00:21:43.359
And the most important content to the vault first is how you do things, right?
00:21:43.519 --> 00:21:46.559
And that's before the knowledge ever leaves.
00:21:48.400 --> 00:21:50.960
So here's another thing, right?
00:21:51.200 --> 00:21:54.319
People are always wondering okay, do I just paste this in?
00:21:55.440 --> 00:21:59.359
Do I upload this um, you know, in a in a Google Doc?
00:22:00.079 --> 00:22:03.680
Should I save this as a you know as an example if you're using clause?
00:22:03.839 --> 00:22:07.119
Should I, you know, save this as a skills markdown file?
00:22:07.200 --> 00:22:08.960
Uh, right, and then use it throughout.
00:22:09.039 --> 00:22:14.160
And I'd hate to use uh an SEO answer, but it depends.
00:22:14.480 --> 00:22:15.279
Uh, right.
00:22:15.920 --> 00:22:17.119
It depends on a lot of things.
00:22:17.200 --> 00:22:19.440
It depends on what model you're using.
00:22:19.599 --> 00:22:22.000
Uh, it depends on the context window.
00:22:22.160 --> 00:22:27.519
It depends on are you using something ultimately inside of a project, inside of a GPT, inside of a gem?
00:22:27.920 --> 00:22:33.759
Are you using it in a kind of quote unquote naked chat where it's not connected to one of those three things?
00:22:33.920 --> 00:22:44.400
Um, and the answer, it does depend, but I think it's um it's important to know well, you can kind of make it work anyway, right?
00:22:44.559 --> 00:22:49.440
As long as you have the right understanding of how all these models work.
00:22:49.680 --> 00:22:55.359
Um, so an easy example of that is to go back to GPTs, right?
00:22:55.440 --> 00:23:05.279
Because a lot of people were under the assumption uh that you could just throw in a large PDF in a GPT, and then at any point, well, okay, it has all my knowledge, right?
00:23:05.359 --> 00:23:24.720
But people didn't understand that, hey, you would probably have to number one, in the custom instructions, you would have to kind of provide a especially for larger and longer documents, you would have to provide kind of like uh an index, right, for how the model should treat that really long document.
00:23:24.880 --> 00:23:32.960
Or if you uploaded a handful of documents, you would have to have some simple rules like, hey, if this happens, then do that, right?
00:23:33.039 --> 00:23:45.119
If the user asks about marketing guidelines, you should check in, you know, page five of document A and page nine of document B.
00:23:45.359 --> 00:23:45.839
Uh right.
00:23:45.920 --> 00:23:48.960
So sometimes you can just upload static context.
00:23:49.279 --> 00:23:51.839
Sometimes you can just do the copy and paste method.
00:23:51.920 --> 00:24:00.480
Uh, sometimes if a certain, you know, app or connector uh syncs dynamically, that can solve a lot of your issues, right?
00:24:00.559 --> 00:24:04.160
You always still have to do the good old human in the loop stuff, right?
00:24:04.240 --> 00:24:13.839
And make sure that you're always testing these things and scoping them and measuring them and you know, making sure the models are properly connecting and pulling out the context that's needed.
00:24:14.000 --> 00:24:16.960
Because the other thing is, well, models are always changing.
00:24:17.200 --> 00:24:20.799
Behavior can be extremely finicky in a large language model, right?
00:24:20.960 --> 00:24:27.839
Taken from someone that's done dozens, if not more than a hundred live demos on this podcast.
00:24:28.000 --> 00:24:30.720
Things can go terribly awry, right?
00:24:30.799 --> 00:24:39.440
You can run the exact same quote unquote prompt, even if you have your context near uh context engineering all the exact same, and it can go in a different direction.
00:24:39.519 --> 00:24:43.200
That is the uh the source of generative AI.
00:24:43.279 --> 00:24:45.279
It is generative, it is non-deterministic.
00:24:45.440 --> 00:24:54.880
So uh you do have to still understand the different ways that you can bring that context in, uh, whether it's paste uh, you know, pasted context, as long as you understand the context window.
00:24:54.960 --> 00:25:02.400
Uh, again, you should always be looking at the um the chain of thought uh as well.
00:25:02.559 --> 00:25:04.880
Um, and and and that's another big thing, too.
00:25:05.039 --> 00:25:10.559
When we talk about why prompt engineering is all but dead, right?
00:25:11.039 --> 00:25:16.319
Well, chain of thought was a very popular prompting technique, right?
00:25:16.480 --> 00:25:28.079
And this was essentially a way that you could kind of uh go through this uh this this process of walking the model through how a human would think.
00:25:28.160 --> 00:25:35.839
This is the chain of thought, or how a smart human would go about uh you know getting this proper answer.
00:25:36.000 --> 00:25:41.359
And you would kind of think of okay, here's how a smart expert would go about uh getting an answer.
00:25:41.440 --> 00:25:49.440
And then you would have to deconstruct that and kind of reverse engineer that uh to a large language model, and that would be called a chain of thought prompting technique.
00:25:49.519 --> 00:25:53.599
But that was kind of when we just had these quote unquote old school transformer models, right?
00:25:53.680 --> 00:26:02.160
That scenario that I painted for you earlier when the models weren't connected to the internet, when you couldn't upload files, when they weren't dynamically connected to your data, right?
00:26:02.240 --> 00:26:05.759
When they didn't have tool calling and all of these different functions, right?
00:26:05.839 --> 00:26:10.319
That's when this kind of chain of thought uh really prompt engineering mattered, right?
00:26:10.400 --> 00:26:15.599
But now this is the default of how today's thinking or hybrid models work, right?
00:26:15.759 --> 00:26:26.559
If you ever read the summarized chain of thought, you'll see, oh, this is essentially they're doing this prompt engineering stuff that was popular in 2023 by default.
00:26:26.720 --> 00:26:28.160
They're doing this under the hood, right?
00:26:28.319 --> 00:26:36.559
Which is why how you talk to a model is way less important now than having the correct context.
00:26:37.680 --> 00:26:47.359
So here's three techniques that I think can uh help you turn good context into expert level outputs.
00:26:47.599 --> 00:26:48.240
All right.
00:26:48.480 --> 00:26:52.240
And these are technically still going back to prompt engineering basics.
00:26:52.319 --> 00:26:58.000
So, yes, prompt engineering stacked with proper context is obviously going to give you the best results.
00:26:58.160 --> 00:27:06.400
So it's not that prompt engineering is dead, you know, still some of its uh, you know, core uh you know, core foundations move uh live on.
00:27:06.640 --> 00:27:08.079
So few shot examples.
00:27:08.160 --> 00:27:14.640
Number one, you should still, even with all that context, give it the examples of what's good, what's bad, right?
00:27:14.720 --> 00:27:16.720
That's what we teach in our Prime Prompt Polish.
00:27:16.799 --> 00:27:27.039
Uh, that is the polish portion, is giving it kind of that uh multi-shot uh kind of um work, right?
00:27:27.200 --> 00:27:32.240
In the same way, think you're trading, you're training that employee, go back to that analogy that still works so well.
00:27:32.480 --> 00:27:41.839
The very first time they hand in their first project, their first assignment, their first deliverable, you're probably gonna go through and sit with them and say, hey, this is great because blank.
00:27:42.000 --> 00:27:45.119
Hey, this is incorrect because blank, right?
00:27:45.279 --> 00:27:48.240
So same thing that's the you know, few shot examples.
00:27:48.400 --> 00:27:52.240
Uh the second one, uh the second technique is rubric first.
00:27:52.640 --> 00:28:00.240
So give, you know, whatever large language model that you're working with a grading criteria in the context window before you even start working.
00:28:00.480 --> 00:28:00.960
Right.
00:28:01.119 --> 00:28:05.680
Uh, this is something, again, that we used to teach in our older PPP Pro course.
00:28:05.759 --> 00:28:12.720
Uh, we don't teach it as much anymore because I don't think it's as uh as useful as it once was, but I think it's still, right?
00:28:12.799 --> 00:28:17.359
I call that uh, you know, temperature, uh, I think I called it no gauges, right?
00:28:17.519 --> 00:28:21.359
Uh you need to give the model something it can gauge or give it a temperature, right?
00:28:21.440 --> 00:28:22.640
And give it examples too.
00:28:22.799 --> 00:28:26.319
So say, hey, uh, let's just talk about creative writing, right?
00:28:26.960 --> 00:28:34.240
Write a sentence as, you know, plainly as possible, as boring as possible, and you say this is a one, right?
00:28:34.319 --> 00:28:35.359
On the creative scale.
00:28:35.519 --> 00:28:38.480
You write a sentence that's kind of creative, this is a five.
00:28:38.720 --> 00:28:41.519
Write the world's most creative sentence ever, that's a 10.
00:28:41.599 --> 00:28:43.839
But tell the model that, tell it why, right?
00:28:44.000 --> 00:28:45.119
And then there you go.
00:28:45.279 --> 00:28:47.759
You've just kind of created a rubric, right?
00:28:47.839 --> 00:28:51.839
So then at any point, you can say, hey, let's do this as a three.
00:28:52.000 --> 00:28:54.079
Hey, let's do this as an eight, right?
00:28:54.240 --> 00:28:56.559
Uh, and that can help you think of not just creative writing.
00:28:56.640 --> 00:29:03.039
That's just an easy one uh to think about, but uh, there's so many different gauges or temperatures that you can put into a model.
00:29:03.279 --> 00:29:05.839
And then last but not least, show, don't tell.
00:29:06.000 --> 00:29:11.599
So sometimes you just need to paste the exact format you want instead of describing it.
00:29:11.759 --> 00:29:27.440
I think the combination of describing what you want and then also uh giving it examples, you know, that's just another uh uh, you know, kind of uh piling on here or doubling up here on our first uh technique, which is few shot examples.
00:29:27.519 --> 00:29:45.440
But if you combine that with the kind of the show don't tell, I think that's a great way, especially if you want outputs formatted uh in a certain way, or if you want outputs to always include, you know, XYZ, uh just giving it examples of that is helpful, but then also giving it the exact formula, uh right.
00:29:45.519 --> 00:29:51.920
So I would always do multiple uh versions uh of this, but the show don't tell is extremely important.
00:29:53.039 --> 00:30:08.079
All right, so we've covered a lot in this start here series, but let me wrap by saying this the context is everything, right?
00:30:08.480 --> 00:30:14.480
I've always said your data is the differentiator, using AI doesn't matter, right?
00:30:14.880 --> 00:30:20.880
Uh it doesn't people always think like, oh, we're using AI, so we're ahead of the curve.
00:30:20.960 --> 00:30:21.680
No, you're not.
00:30:21.839 --> 00:30:22.480
Um, right.
00:30:22.640 --> 00:30:26.960
And prompt engineering doesn't matter as much anymore, right?
00:30:27.119 --> 00:30:36.000
Because now these models by default are doing a lot of that heavy lifting that really paid off in 2023 and early 2024.
00:30:36.160 --> 00:30:42.480
So you have to stop thinking about talking to the model a certain way.
00:30:42.720 --> 00:30:51.359
And actually, especially if you are non-technical, especially if you're not a you know heavy AI user, that's actually a good thing.
00:30:51.680 --> 00:31:07.119
Because I think earlier on, right, in 2023 and 2024, a lot of non-technical people were kind of scared off of using large language models because you know, earlier on, there, you know, this prompt engineering, it's everything, it's everything, it's everything.
00:31:07.200 --> 00:31:09.519
And people are like, oh, I'm not an engineer, right?
00:31:09.680 --> 00:31:11.039
It doesn't matter anymore.
00:31:11.200 --> 00:31:17.680
You don't have to, you know, understand, you know, chain of thought or, you know, any of these other prompting techniques anymore.
00:31:17.920 --> 00:31:26.960
You can talk to it just even like a lazy human, because if you have the context side, how you talk to the model is not as important.
00:31:27.279 --> 00:31:33.039
But let me just give you three last pieces of advice to wrap up everything we talked about.
00:31:33.200 --> 00:31:36.319
So stop starting with a question.
00:31:36.559 --> 00:31:41.519
Instead, start with a tiny context pack, extend on it from there.
00:31:41.759 --> 00:31:44.319
Next, reuse what works, right?
00:31:44.480 --> 00:31:58.319
Whether you want to create skills, create GPTs, create projects, it doesn't matter, but you need to reuse what works because you do need to put much more in on the front end on the context side than you think you might need.
00:31:58.559 --> 00:32:00.480
More and more in more context.
00:32:00.640 --> 00:32:06.559
As long as you understand and don't exceed the context window, context will not bite you in the butt.
00:32:06.799 --> 00:32:07.119
Right?
00:32:07.279 --> 00:32:10.319
It will bite you in the butt if you don't provide enough context.
00:32:10.480 --> 00:32:14.240
But you need to be smart and think in a scalable fashion.
00:32:14.400 --> 00:32:18.000
Uh and you need to reuse what works.
00:32:18.640 --> 00:32:28.240
And then last but not least, remember that expert level results come from a system that you can repeat every time, right?
00:32:28.319 --> 00:32:40.000
It's all about if you want the expert level outputs, there's a good chance if you're doing this the right way that you're wasting time by not reusing it.
00:32:40.160 --> 00:32:40.480
Right.
00:32:40.559 --> 00:32:43.200
Here's a here's an extra pro tip, right?
00:32:43.279 --> 00:32:52.720
Especially if you're using as an example, uh, well, any of the models, if you have personalization and memory on, um, go through.
00:32:52.960 --> 00:32:59.440
Ask the model, what are the things that I'm most that I'm using you for that is most repetitive?
00:32:59.680 --> 00:33:04.640
You know, what are some systems that I can build uh that I can reuse, right?
00:33:04.799 --> 00:33:22.240
You'll probably be surprised, especially, you know, there's been some recent updates in the last, I'd say, three to four months uh with the big three players there and improving how the memory, how um, you know, remembering your past chat conversations, it's actually really good now, right?
00:33:22.319 --> 00:33:26.240
When it first started to debut like 15, 18 months ago, it wasn't that good, right?
00:33:26.319 --> 00:33:34.720
But I'll say the last three months, uh, these AI chat bots, essentially their memory and what they know about you is really good and being able to recall past conversations.
00:33:34.880 --> 00:33:39.039
So just ask, what am I wasting the most time on, right?
00:33:39.359 --> 00:33:40.480
Doing every single day?
00:33:40.640 --> 00:33:42.880
What are some skills that I should be building?
00:33:43.039 --> 00:33:46.960
What are some projects that I should be uh, you know, building?
00:33:47.119 --> 00:33:49.440
What are GPTs that I should be using?
00:33:49.599 --> 00:33:49.920
Right.
00:33:50.079 --> 00:33:59.599
Don't never, never feel like you're uh, you know, stupid or anything by just saying to a large language model, I'm not sure.
00:33:59.759 --> 00:34:00.480
Because guess what?
00:34:00.640 --> 00:34:02.559
Large language models are smarter than us.
00:34:02.799 --> 00:34:03.440
They are.
00:34:03.599 --> 00:34:11.840
They can pick up those uh those little pieces that humans tend to miss or even things that you may miss about yourself.
00:34:12.079 --> 00:34:19.119
So uh actually a great way to fill in that context is just to ask a large language model, right?
00:34:19.199 --> 00:34:28.559
There's kind of this more uh you know popular uh you know trend of creating like a skills markdown file or a role markdown file.
00:34:28.800 --> 00:34:29.920
Ask any of the models.
00:34:30.079 --> 00:34:39.280
Hey, based on everything you know about me, help me build out some of these blocks that you can then use modularly and take them with you as well, right?
00:34:39.440 --> 00:34:42.400
Because I can guarantee, and I've gone through this process.
00:34:42.559 --> 00:35:00.000
I actually did it like one or two months ago, where I just went through that process in all of the models because chances are I use them for different things because they each have their different strengths, and then I combine them into one big uh, you know, skills uh file, role file, all these different things.
00:35:00.079 --> 00:35:04.719
And now I have these building blocks that I can use whenever I need them.
00:35:04.880 --> 00:35:09.440
All right, so that is how you get those expert level results.
00:35:09.599 --> 00:35:14.880
So I hope that this version of the Start Here series was helpful.
00:35:15.039 --> 00:35:20.079
If it was, make sure if you haven't already, go to startheareseries.com.
00:35:20.239 --> 00:35:35.360
That is going to give you free access to not just our context engineering course that's been taken by more than 15,000 business leaders called Prime Prompt Polish, but it is going to give you free access to our inner circle community.
00:35:35.519 --> 00:35:47.280
And then you can also go to the Start Here Series uh space inside of our community and go listen to and catch up on all of our Start Here series, all in one easy place.
00:35:47.440 --> 00:35:47.840
All right.
00:35:48.000 --> 00:35:49.119
I hope this was helpful.
00:35:49.280 --> 00:35:50.239
Thank you for tuning in.
00:35:50.400 --> 00:35:53.760
I hope to see you back tomorrow and every day for more everyday AI.
00:35:53.840 --> 00:35:54.400
Thanks, y'all.
00:00:01.360 --> 00:00:04.639
Why does no one talk about prompt engineering anymore?
00:00:04.879 --> 00:00:13.439
I mean, if you were whined back like two years ago, you would have sworn that prompt engineering would be the world's most popular future job title.
00:00:13.759 --> 00:00:15.839
But that's obviously not the case.
00:00:16.079 --> 00:00:20.800
And the essential disappearance of that term is twofold.
00:00:21.039 --> 00:00:28.320
One, models are smarter, and it doesn't always matter the exact way we talk to them, as long as we get the message cross.
00:00:28.559 --> 00:00:37.280
And two, an output that moves the needle is much more dependent on business context versus just wording something a certain way.
00:00:37.520 --> 00:00:41.759
Hence the resurgence of the term context engineering.
00:00:42.240 --> 00:00:44.240
But what does that even mean?
00:00:44.479 --> 00:00:53.600
And how can you understand the required inputs of context engineering to get better outputs out of a large language model?
00:00:53.840 --> 00:01:00.000
Well, if that's one of the things that you or your business is grappling with, then you're in luck.
00:01:00.240 --> 00:01:10.560
Because on today's episode of our start here series, we're tackling context engineering and how to get expert level outputs from AI chatbots.
00:01:10.719 --> 00:01:13.439
All right, I am excited for today's show.
00:01:13.599 --> 00:01:14.480
I hope you are too.
00:01:14.640 --> 00:01:16.159
If you're new here, welcome.
00:01:16.400 --> 00:01:20.239
This is the everyday AI Start Here series.
00:01:20.400 --> 00:01:25.920
So after 700 plus episodes, one of the most common questions I get is where do I start?
00:01:26.079 --> 00:01:29.120
So that's why we started the Start Here series.
00:01:29.280 --> 00:01:33.040
And this is actually volume seven of this exact series.
00:01:33.200 --> 00:01:41.359
So the Start Here series is the essential podcast series to both learn the AI basics and to double down on your knowledge.
00:01:41.519 --> 00:01:43.840
So if that's what you're trying to do, you're in luck.
00:01:44.000 --> 00:01:47.680
Make sure you go to starthearseries.com.
00:01:47.840 --> 00:01:53.040
That's going to redirect you and give you free access to our inner circle community.
00:01:53.120 --> 00:02:11.840
So there you can not only go take our context engineering course called Prime Prompt Polish for free, but also network with a bunch of other people and you'll be redirected right to our Start Here Series area where you can go and listen to every single episode in this series, all right there at your fingertips.
00:02:12.240 --> 00:02:12.639
All right.
00:02:12.800 --> 00:02:23.360
And if you missed our last episode of this series, we talked about how to train your team on AI and the seven steps to educate your organization on large language models.
00:02:23.520 --> 00:02:30.240
And the last step in there was well, making that step to go from operator to orchestrator.
00:02:30.400 --> 00:02:35.439
So that's where we kind of left you with the last step in our series.
00:02:35.520 --> 00:02:37.039
And that's where we're going to pick up.
00:02:37.280 --> 00:02:46.240
Because actually, one of the biggest things that you can do from going from an operator or essentially someone pushing all the buttons to an orchestrator, right?
00:02:46.319 --> 00:02:53.919
Which is when AI starts to do the work for you, is having the right data and providing that data to the model in the right way.
00:02:54.000 --> 00:02:57.919
And that is the backbone of what context engineering is.
00:02:58.319 --> 00:03:12.080
It is the process a human goes through to make sure a large language model has the right context about not just you, your role, what you're trying to accomplish, but maybe most importantly, your business and the competitive market.
00:03:13.039 --> 00:03:21.360
So this is the big differentiator because the same two people, they can use the same prompt, right?
00:03:21.439 --> 00:03:27.759
Going back to uh prompting and prompt engineering, uh, and you can get wildly different answers, right?
00:03:27.919 --> 00:03:46.319
Because if one person, whether it's in custom instructions, a GPT, a project, et cetera, or uh using Chat GPT apps or uh, you know, claude uh connectors, whatever you may be using, Google Gemini apps, uh right, you can put in the exact same prompt as someone sitting next to you.
00:03:46.479 --> 00:03:55.759
And if you have your context engineering 101 ducks in a row, your output will be light years better than that person who does not.
00:03:56.080 --> 00:04:14.240
And the difference isn't necessarily as difficult as it may sound, because the skill separating average AI users from expert level ones is just providing the model the needed context and understanding how it works in different scenarios.
00:04:14.400 --> 00:04:16.079
So that's what we're gonna cover today.
00:04:16.160 --> 00:04:26.800
We're gonna well first talk about why the AI industry as a whole and really just the business landscape has shifted away from prompt engineering and really just more focusing on context engineering.
00:04:27.040 --> 00:04:32.160
And I will say that that shift kind of happened and popularized in late 2025.
00:04:32.240 --> 00:04:39.279
Then I'm gonna lay out for you a six-part framework and also a four-layer system for structuring what your AI sees.
00:04:39.360 --> 00:04:50.399
And then I'm gonna walk you through how to build reusable context vaults or kind of skills one and the same and use platform features that are already at your fingertips.
00:04:51.040 --> 00:04:53.120
All right, let's get into it.
00:04:53.360 --> 00:04:57.120
So RIP prompt engineering, all right.
00:04:57.279 --> 00:05:01.279
Uh kind of, but here's here's the thing.
00:05:01.439 --> 00:05:16.480
Um, if you think back to the early days of Chat GPT, or even technically before Chat GPT, right, when the GPT technology was available uh to dozens and eventually hundreds of other platforms before ChatGPT even came out, right?
00:05:16.800 --> 00:05:23.839
So much of what you were able to get out of a model was how you talk to it.
00:05:24.079 --> 00:05:25.920
And that was for a couple of reasons.
00:05:26.079 --> 00:05:31.360
Number one, the AI models themselves were a little more old school, uh, right.
00:05:31.439 --> 00:05:33.439
And I'll talk about that here in a little bit.
00:05:33.600 --> 00:05:36.800
Uh, you couldn't always upload documents, right?
00:05:36.959 --> 00:05:43.519
And you couldn't always paste in a bunch of information either, because the model's context windows were smaller.
00:05:43.600 --> 00:05:46.959
So essentially, it would always forget things very quickly.
00:05:47.120 --> 00:05:51.759
Uh, you couldn't, for the most part, upload documents, at least not very easily.
00:05:51.920 --> 00:05:52.399
Uh, right.
00:05:52.480 --> 00:05:59.439
So this really changed that uh differentiator of if you were gonna get a good output versus a bad output.
00:06:00.079 --> 00:06:03.279
And it was really just how you talk to the model, right?
00:06:03.360 --> 00:06:10.240
If you use certain prompting techniques, uh you could kind of pull the best uh out of a model's training data, right?
00:06:10.319 --> 00:06:16.319
Because even if you think back to the way earlier days of large language models, they weren't connected to the internet, right?
00:06:16.480 --> 00:06:20.720
So they weren't connected to the internet, they didn't have tool calling, uh, right.
00:06:20.800 --> 00:06:28.639
And for the most part, you couldn't even upload files, which is why in the earlier days, prompt engineering was actually really important.
00:06:29.360 --> 00:06:32.399
Because even in that data set, right?
00:06:32.480 --> 00:06:39.519
Um, and if you go back and listen to the earlier uh episodes of our start here series, we talk about training data and everything that goes into it.
00:06:39.600 --> 00:06:54.800
Uh, but for the most part, in the earlier days of Chat GPT and you know, when Gemini was barred and you know, early copilot days, etc., it was really how you talk to the model because the model had so fewer capabilities.
00:06:55.040 --> 00:06:59.199
Yet there was still a lot of information there, right?
00:06:59.439 --> 00:07:07.839
Even the early models like GPT-3 or GPT-3.5, uh, or you know, the first you know version of Google Gemini.
00:07:08.160 --> 00:07:13.199
Um, even though we look back at those models now and we think, oh, they weren't very good, they were.
00:07:13.519 --> 00:07:16.399
You just really had to learn how to talk to it, right?
00:07:16.560 --> 00:07:24.959
So now you can go talk to any of today's smartest models, and you don't even have to really put a sentence that makes sense, right?
00:07:25.040 --> 00:07:31.759
Sometimes I find myself, you know, if I'm not using voice dictation with models and I'm just and if I'm just typing, right?
00:07:31.920 --> 00:07:35.759
I've become and I've probably programmed myself to know the models are so smart.
00:07:35.920 --> 00:07:46.639
You know, it's misspellings and I say the wrong thing and all these things, but I know in the end it doesn't matter because the today's models are so incredibly smart at understanding what I'm trying to say, right?
00:07:46.720 --> 00:07:50.800
Especially when I have personalization enabled, memory enabled, all of these other things.
00:07:51.040 --> 00:07:56.319
The actual words that I'm telling a model don't mean a ton.
00:07:58.319 --> 00:08:00.480
Now, right?
00:08:01.040 --> 00:08:03.680
That's how it is, but it isn't how it used to be.
00:08:04.000 --> 00:08:08.959
Because the prompt engineering used to focus on you had to say things the exact right way, right?
00:08:09.120 --> 00:08:16.720
But if you did, what you could get was just a step change different than what you could get if you didn't word something the direct way.
00:08:16.800 --> 00:08:20.959
It was like you almost had a password that no one else had and at the time, right?
00:08:21.120 --> 00:08:25.920
Proper prompt engineering, it was an amazing skill to have.
00:08:26.639 --> 00:08:31.040
But right now, it doesn't matter as much, right?
00:08:31.279 --> 00:08:37.039
You can have the best prompt engineering skills in the world, but if you don't have the context, it doesn't matter.
00:08:37.279 --> 00:08:43.360
And I think the industry really realized that the bottleneck was never about how we talked to the model, it was the information behind it.
00:08:43.600 --> 00:08:50.399
And I think that this shift really started to happen uh in probably mid-June of 2025.
00:08:50.559 --> 00:09:04.559
So, you know, two of the people that are kind of credited uh with popularizing this concept of context engineering were Shopify CEO Toby Ludke, um, who just kind of called for the move from prompts to context.
00:09:04.639 --> 00:09:10.320
And then in the same month, former OpenAI co-founder Andre Kapathy kind of endorsed that term publicly.
00:09:10.399 --> 00:09:19.840
And then in September, uh Anthropoc even published published a dedicated blog talking about moving from moving away from this concept of prompt engineering.
00:09:20.159 --> 00:09:22.720
And I will just go ahead and say this, right?
00:09:22.960 --> 00:09:30.720
Not one of those I told you so, uh, but we've been teaching this concept of context engineering, even though I didn't call it that.
00:09:30.879 --> 00:09:34.159
Um, I believe we started teaching it in late 2023.
00:09:34.399 --> 00:09:45.039
Um, so yeah, we've I've done more than uh, you know, 200, uh probably like 210 now uh live trainings on, you know, essentially prompting models, right?
00:09:45.200 --> 00:09:54.240
So even early on, right, I was teaching prompt engineering, but that really shifted uh in late 2023 and early 2024.
00:09:54.399 --> 00:10:04.159
You know, we kind of, you know, we have our prime prompt polish course, but if you have taken it in the last two and a half, three years, you know that we started to teach this concept called refine cue.
00:10:04.320 --> 00:10:07.039
And essentially that is context engineering at its core.
00:10:07.200 --> 00:10:10.720
So, you know, it's it's it's not a a new concept, right?
00:10:10.799 --> 00:10:12.879
Because I've been teaching people this for a long time.
00:10:13.120 --> 00:10:22.960
But the term, uh the terminology around context engineering, and it has really uh snowballed into more of a movement, uh, has really picked up in the last year or so.
00:10:23.200 --> 00:10:27.519
And that's because it's about designing the environment and not just the question.
00:10:27.759 --> 00:10:31.039
So an easy way to think about this is to think of the AI, right?
00:10:31.120 --> 00:10:38.399
Think of that whatever large language model that you that you're using as a processor and the context window is its working memory.
00:10:38.480 --> 00:10:43.440
Okay, so context window without getting uh too technical, that's like a hard drive, right?
00:10:43.600 --> 00:10:48.080
So in the same way, maybe you have a one terabyte hard drive, let's just say, right?
00:10:48.159 --> 00:10:55.600
If your hard drive's full, then if you try to put more information, maybe fortunately a computer will stop you from doing that.
00:10:55.759 --> 00:10:57.440
A large language model will not.
00:10:57.519 --> 00:11:04.320
So once its quote unquote hard drive gets full, it's just gonna delete the first file that you ever uploaded, right?
00:11:04.399 --> 00:11:12.320
So that's how a context window works with large language models, except you never really know when you are hitting that context window, right?
00:11:12.480 --> 00:11:24.799
So what this means is well, your context becomes very important and understanding that working memory and how the large language models work specifically with your data, right?
00:11:24.960 --> 00:11:28.000
That's grasping the basics of context engineering, right?
00:11:28.080 --> 00:11:29.519
And it's it's a little different, right?
00:11:29.600 --> 00:11:35.919
So it's different depending, uh, it does get a little convoluted depending on what large language model you're using, right?
00:11:36.000 --> 00:11:44.480
As an example, if you're using GPT-5.2 versus GPT-5.2 thinking versus if you're using GPT-5.2 Pro via the API.
00:11:44.639 --> 00:11:46.799
So it is, it is a little bit different.
00:11:47.120 --> 00:11:51.919
But essentially, the concept of prompt engineering it ends once you hit enter.
00:11:52.080 --> 00:12:02.879
And context engineering is an ongoing battle to make sure that your uh model or your session with the model has access to and can understand your business data.
00:12:03.039 --> 00:12:09.200
Um, and there was a study from Intuition Labs last year that said that 40% of AI projects fail.
00:12:09.360 --> 00:12:15.519
Uh and one of the reasons or the main reasons why is well, it comes from poor context and it's not actually the model.
00:12:15.679 --> 00:12:25.279
It's the model either number one, not understanding what you want out of it, or number two, it just doesn't have that data that can be the differentiator.
00:12:26.399 --> 00:12:30.240
So here's why it matters more than ever.
00:12:30.480 --> 00:12:41.600
Well, uh, if you talk about the big three or the big four, uh, right, so ChatGPT, Anthropics Claude, Google Gemini, uh, and Microsoft Copilot, they it used to be fairly hard.
00:12:41.759 --> 00:12:51.519
Aside from Copilot, it's always been uh more straightforward if you understand the uh the tech and the permissions uh landscape uh inside Microsoft Windows.
00:12:51.679 --> 00:12:52.799
That's a whole nother story.
00:12:52.879 --> 00:12:56.799
Uh but let's look at the other three with just ChatGPT, Claude, and Gemini.
00:12:57.200 --> 00:13:04.799
Um, I would say in early 2025, it was actually kind of difficult to use your company's data, right?
00:13:04.879 --> 00:13:14.720
You could even say, oh, well, Jordan, you know, there's been these things like GPTs where you know you could upload, you know, documents and have a specialized version of ChatGPT that had access to those documents.
00:13:14.879 --> 00:13:16.240
Yeah, but have you ever tested it?
00:13:16.399 --> 00:13:21.200
Have you ever run the needle in the haystack test on that GPT that has right?
00:13:21.279 --> 00:13:28.639
People would just assume, oh, I'm gonna, you know, upload a you know, 500-page PDF into a GPT and then it knows everything about me and my company.
00:13:28.720 --> 00:13:30.000
No, absolutely not.
00:13:30.240 --> 00:13:30.480
Right?
00:13:30.639 --> 00:13:37.360
That means you didn't really understand uh, you know, how uh these models tokenize that information or access that information.
00:13:37.519 --> 00:13:45.200
But now it's much easier, uh, because these models by default can create searchable indexes of your files.
00:13:45.360 --> 00:13:49.840
So think of it like this way: in the same way, let's say you have a Mac, right?
00:13:50.080 --> 00:13:57.840
And you can go to your Mac Finder and your search bar there, and you can search for maybe a word that is within a PDF.
00:13:57.919 --> 00:14:02.240
And it's gonna know because it's indexed that file and it understands it, right?
00:14:02.320 --> 00:14:07.279
That obviously requires a certain level of compute, right?
00:14:07.519 --> 00:14:10.639
That two years ago these models just didn't have, but now they do.
00:14:10.799 --> 00:14:18.240
So within ChatGPT, as an example, they have things like projects, they have things uh that were previously called connectors, now they're called apps.
00:14:18.320 --> 00:14:19.919
Uh Claude, same thing.
00:14:20.000 --> 00:14:22.720
Uh, you can have, you know, project-specific memory.
00:14:22.799 --> 00:14:27.679
Uh, you can have these reusable skills that can take advantage of your business context.
00:14:27.759 --> 00:14:29.440
Uh, same thing with Google Gemini.
00:14:29.679 --> 00:14:34.480
You can have these gems that connect live to your Google Drive, to your Gmail, to your calendar, right?
00:14:34.799 --> 00:14:47.759
So at various levels and in different ways, the three main players within a couple of clicks only can connect to your, in many cases, your dynamic business data.
00:14:47.840 --> 00:14:49.679
It's not always dynamic, uh, right.
00:14:49.840 --> 00:14:56.559
In some cases, it is, and it will create a searchable and live indexed of everything that you connect to it.
00:14:56.639 --> 00:14:56.960
All right.
00:14:57.120 --> 00:15:00.159
Again, I have to throw out that same asterisk as I always do.
00:15:00.240 --> 00:15:04.960
You know, always make sure you have permission to connect your company's data to a large language model, blah, blah, blah.
00:15:05.039 --> 00:15:05.679
Uh, right.
00:15:05.919 --> 00:15:19.519
Um, but once you do, that is the first step in context engineering is making sure, number one, the model has access uh to the context or the data that it needs.
00:15:19.759 --> 00:15:25.759
But you also, more importantly, need to understand how it works in each scenario, right?
00:15:25.919 --> 00:15:33.200
Like as an example, I would say most people, even if you listen to the show often, you might not know that ChatGPT had these things called connectors.
00:15:33.279 --> 00:15:39.360
But oh, wait, actually, in December, they changed them right before the holiday season, they changed them all to apps.
00:15:39.440 --> 00:15:40.799
And a lot of people miss that.
00:15:40.960 --> 00:15:49.600
And with that comes, uh, depending on what app you're talking about now, well, it maybe handles um your data a little bit differently than it did before.
00:15:49.759 --> 00:16:02.240
So you do have to, depending on the platform that you use, if you really want to understand context engineering, well, you have to understand how each of these different platforms connects to different data sources because it's not it's it's not uniform, right?
00:16:02.320 --> 00:16:03.840
It's really not, right?
00:16:04.000 --> 00:16:13.600
Even if you look at Chat GPT, uh their apps, there's four different ways uh that it can look at your data in four different ways that it can understand your data, right?
00:16:13.759 --> 00:16:25.039
In the same way, uh, right, if you think of cloud storage and there's these different permissions and different ways to access data, you know, that does uh trickle down to large language models as well.
00:16:26.000 --> 00:16:32.559
So uh let's talk about kind of what I'm calling these six building blocks of effective AI context.
00:16:32.879 --> 00:16:40.720
Because the step one is well, what I just covered, you have to make sure, depending on what large language model you're using, it has access to your data.
00:16:40.879 --> 00:16:42.960
Uh, but it's not just about data.
00:16:43.039 --> 00:16:46.960
And it's not just about telling a large language model, here's something about me, right?
00:16:47.039 --> 00:16:48.399
That's providing context, right?
00:16:48.480 --> 00:16:49.279
Context clues.
00:16:49.440 --> 00:16:52.320
Tell me more about yourself, what you're trying to accomplish, right?
00:16:52.399 --> 00:16:58.000
I'd like to break it down into these six different building blocks for building context within a context window.
00:16:58.080 --> 00:16:58.399
Okay.
00:16:58.559 --> 00:17:07.200
Once you're out of the context window, again, depending on how you're connected, uh, connecting your data, um, you might start to get poor results.
00:17:07.359 --> 00:17:13.119
So keep this in mind and keep these six building blocks in the context window of any given conversation.
00:17:13.359 --> 00:17:17.680
So goal, that's what you need the AI to produce and for whom.
00:17:17.920 --> 00:17:22.799
Constraints, understanding the boundaries, rules, things to avoid, and format requirements.
00:17:22.960 --> 00:17:27.680
Uh reference material, that's the approved facts, data, and source documents to draw from.
00:17:27.839 --> 00:17:33.920
Examples, those are representative samples of the output you want, plus context and examples.
00:17:34.079 --> 00:17:43.599
Um, then procedures, those are step-by-step instructions for how the AI model should approach the tax, uh, task, and then the evaluation rubric.
00:17:43.759 --> 00:17:48.559
So that's grading criteria so the AI can assess its own output quality.
00:17:48.880 --> 00:17:52.960
So this isn't a perfect formula and it is ever changing, right?
00:17:53.119 --> 00:18:01.440
Uh, but I think this is for the most part, number one, you should just go take our free prime prompt polish course and go through the whole thing.
00:18:01.519 --> 00:18:12.799
Uh, I think Refine Q is another version uh of building this essential uh building blocks of context, but this is another um kind of framework that I think works really well.
00:18:13.519 --> 00:18:20.559
Okay, now you don't just think of those six building blocks of okay, uh I'm good, right?
00:18:20.640 --> 00:18:24.960
Let me get those things because you have to actually apply them in different layers.
00:18:25.200 --> 00:18:27.599
So the first layer is well, personal.
00:18:28.000 --> 00:18:30.720
That's your own personal context.
00:18:31.039 --> 00:18:34.319
The second layer is your team, right?
00:18:35.440 --> 00:18:41.039
If you're on a small team, might be a little less difference between that first layer and the second layer.
00:18:41.200 --> 00:18:45.599
If you're on a large team, could be a huge difference.
00:18:45.759 --> 00:18:50.000
Then the third layer is obviously your company or your business, right?
00:18:50.079 --> 00:18:53.920
So those are things like your brand voice, your policies, your product details, et cetera.
00:18:54.079 --> 00:18:54.559
Uh, right.
00:18:54.640 --> 00:18:59.680
When we talk about layer one, that's your own personal role, your expertise, uh, right.
00:18:59.839 --> 00:19:04.240
Layer two, uh, that's kind of shared definitions, uh, your project goals, conversations.
00:19:04.319 --> 00:19:08.240
Like I said, layer three, that's at the company level, brand voice, policies, et cetera.
00:19:08.319 --> 00:19:10.160
And then number four, that's your market.
00:19:10.400 --> 00:19:16.799
That's the your position in the competitive market, uh, industry, invite uh insights, trends, et cetera.
00:19:17.039 --> 00:19:31.039
So it's not just about bringing the right folder in via a Chat GPT app, it's not just about those building blocks, it's also making sure that you apply those at the different layers that a large language model needs.
00:19:31.839 --> 00:19:37.200
And then when you do that, you can probably imagine by now, oh my gosh, this sounds extremely time consuming.
00:19:37.279 --> 00:19:38.319
Yeah, it is.
00:19:38.559 --> 00:19:38.880
Right.
00:19:39.039 --> 00:19:45.920
I always tell people if you're thinking about AI as if it's an easy button, you're looking at it all the wrong way, right?
00:19:46.160 --> 00:20:04.079
The best thing, the best analogy that I've probably ever taught, and it's from the very first, right, in early 2023, when I did my very first uh, you know, Chat GPT prompting course back when prompt engineering actually was a thing, and it still holds true to today.
00:20:04.559 --> 00:20:10.000
When you are working with a large language model, you have to think of it as you are training a new employee, right?
00:20:10.160 --> 00:20:16.160
A new college grad or someone that just transferred in and they're a capable person, right?
00:20:16.640 --> 00:20:22.559
But their output is going to largely be dependent on how much context you share with them, right?
00:20:22.960 --> 00:20:26.000
Because if you just throw down a giant prompt, right?
00:20:26.079 --> 00:20:34.640
If you throw down a giant training manual and then say, all right, first assignments due in an hour, they're gonna fail, right?
00:20:34.799 --> 00:20:40.400
You have to go through and give them the context and the conversation and the iteration that they need, right?
00:20:40.480 --> 00:20:50.079
And one of the ways that you can think about this, because what I just laid out in terms of uh, you know, applying the context, okay, those six different building blocks across four different layers, that's a lot.
00:20:50.240 --> 00:20:57.839
Well, you have to think in the same way that you would invest in an employee, you do it so in the long run, it is scalable, repeatable, reusable.
00:20:57.920 --> 00:20:58.319
Uh, right.
00:20:58.400 --> 00:21:03.200
So you can think of this as creating a context vault or maybe a skill, right?
00:21:03.279 --> 00:21:14.880
That's another terminology, uh, another term, uh, you know, kind of uh created by anthropic, but skills uh have been uh picked up by, you know, all the main players in the AI space.
00:21:14.960 --> 00:21:19.759
So think of a vault or a skill as kind of this folder of reusable context, right?
00:21:19.839 --> 00:21:28.799
So these different procedures, rubrics, key facts that you might be able to reuse or uh to modularly use uh with uh within each other.
00:21:28.880 --> 00:21:37.920
Uh so you can build a skill or a vault per role to start, then expand that as your team standardizes whatever process you're working on.
00:21:38.079 --> 00:21:43.359
And the most important content to the vault first is how you do things, right?
00:21:43.519 --> 00:21:46.559
And that's before the knowledge ever leaves.
00:21:48.400 --> 00:21:50.960
So here's another thing, right?
00:21:51.200 --> 00:21:54.319
People are always wondering okay, do I just paste this in?
00:21:55.440 --> 00:21:59.359
Do I upload this um, you know, in a in a Google Doc?
00:22:00.079 --> 00:22:03.680
Should I save this as a you know as an example if you're using clause?
00:22:03.839 --> 00:22:07.119
Should I, you know, save this as a skills markdown file?
00:22:07.200 --> 00:22:08.960
Uh, right, and then use it throughout.
00:22:09.039 --> 00:22:14.160
And I'd hate to use uh an SEO answer, but it depends.
00:22:14.480 --> 00:22:15.279
Uh, right.
00:22:15.920 --> 00:22:17.119
It depends on a lot of things.
00:22:17.200 --> 00:22:19.440
It depends on what model you're using.
00:22:19.599 --> 00:22:22.000
Uh, it depends on the context window.
00:22:22.160 --> 00:22:27.519
It depends on are you using something ultimately inside of a project, inside of a GPT, inside of a gem?
00:22:27.920 --> 00:22:33.759
Are you using it in a kind of quote unquote naked chat where it's not connected to one of those three things?
00:22:33.920 --> 00:22:44.400
Um, and the answer, it does depend, but I think it's um it's important to know well, you can kind of make it work anyway, right?
00:22:44.559 --> 00:22:49.440
As long as you have the right understanding of how all these models work.
00:22:49.680 --> 00:22:55.359
Um, so an easy example of that is to go back to GPTs, right?
00:22:55.440 --> 00:23:05.279
Because a lot of people were under the assumption uh that you could just throw in a large PDF in a GPT, and then at any point, well, okay, it has all my knowledge, right?
00:23:05.359 --> 00:23:24.720
But people didn't understand that, hey, you would probably have to number one, in the custom instructions, you would have to kind of provide a especially for larger and longer documents, you would have to provide kind of like uh an index, right, for how the model should treat that really long document.
00:23:24.880 --> 00:23:32.960
Or if you uploaded a handful of documents, you would have to have some simple rules like, hey, if this happens, then do that, right?
00:23:33.039 --> 00:23:45.119
If the user asks about marketing guidelines, you should check in, you know, page five of document A and page nine of document B.
00:23:45.359 --> 00:23:45.839
Uh right.
00:23:45.920 --> 00:23:48.960
So sometimes you can just upload static context.
00:23:49.279 --> 00:23:51.839
Sometimes you can just do the copy and paste method.
00:23:51.920 --> 00:24:00.480
Uh, sometimes if a certain, you know, app or connector uh syncs dynamically, that can solve a lot of your issues, right?
00:24:00.559 --> 00:24:04.160
You always still have to do the good old human in the loop stuff, right?
00:24:04.240 --> 00:24:13.839
And make sure that you're always testing these things and scoping them and measuring them and you know, making sure the models are properly connecting and pulling out the context that's needed.
00:24:14.000 --> 00:24:16.960
Because the other thing is, well, models are always changing.
00:24:17.200 --> 00:24:20.799
Behavior can be extremely finicky in a large language model, right?
00:24:20.960 --> 00:24:27.839
Taken from someone that's done dozens, if not more than a hundred live demos on this podcast.
00:24:28.000 --> 00:24:30.720
Things can go terribly awry, right?
00:24:30.799 --> 00:24:39.440
You can run the exact same quote unquote prompt, even if you have your context near uh context engineering all the exact same, and it can go in a different direction.
00:24:39.519 --> 00:24:43.200
That is the uh the source of generative AI.
00:24:43.279 --> 00:24:45.279
It is generative, it is non-deterministic.
00:24:45.440 --> 00:24:54.880
So uh you do have to still understand the different ways that you can bring that context in, uh, whether it's paste uh, you know, pasted context, as long as you understand the context window.
00:24:54.960 --> 00:25:02.400
Uh, again, you should always be looking at the um the chain of thought uh as well.
00:25:02.559 --> 00:25:04.880
Um, and and and that's another big thing, too.
00:25:05.039 --> 00:25:10.559
When we talk about why prompt engineering is all but dead, right?
00:25:11.039 --> 00:25:16.319
Well, chain of thought was a very popular prompting technique, right?
00:25:16.480 --> 00:25:28.079
And this was essentially a way that you could kind of uh go through this uh this this process of walking the model through how a human would think.
00:25:28.160 --> 00:25:35.839
This is the chain of thought, or how a smart human would go about uh you know getting this proper answer.
00:25:36.000 --> 00:25:41.359
And you would kind of think of okay, here's how a smart expert would go about uh getting an answer.
00:25:41.440 --> 00:25:49.440
And then you would have to deconstruct that and kind of reverse engineer that uh to a large language model, and that would be called a chain of thought prompting technique.
00:25:49.519 --> 00:25:53.599
But that was kind of when we just had these quote unquote old school transformer models, right?
00:25:53.680 --> 00:26:02.160
That scenario that I painted for you earlier when the models weren't connected to the internet, when you couldn't upload files, when they weren't dynamically connected to your data, right?
00:26:02.240 --> 00:26:05.759
When they didn't have tool calling and all of these different functions, right?
00:26:05.839 --> 00:26:10.319
That's when this kind of chain of thought uh really prompt engineering mattered, right?
00:26:10.400 --> 00:26:15.599
But now this is the default of how today's thinking or hybrid models work, right?
00:26:15.759 --> 00:26:26.559
If you ever read the summarized chain of thought, you'll see, oh, this is essentially they're doing this prompt engineering stuff that was popular in 2023 by default.
00:26:26.720 --> 00:26:28.160
They're doing this under the hood, right?
00:26:28.319 --> 00:26:36.559
Which is why how you talk to a model is way less important now than having the correct context.
00:26:37.680 --> 00:26:47.359
So here's three techniques that I think can uh help you turn good context into expert level outputs.
00:26:47.599 --> 00:26:48.240
All right.
00:26:48.480 --> 00:26:52.240
And these are technically still going back to prompt engineering basics.
00:26:52.319 --> 00:26:58.000
So, yes, prompt engineering stacked with proper context is obviously going to give you the best results.
00:26:58.160 --> 00:27:06.400
So it's not that prompt engineering is dead, you know, still some of its uh, you know, core uh you know, core foundations move uh live on.
00:27:06.640 --> 00:27:08.079
So few shot examples.
00:27:08.160 --> 00:27:14.640
Number one, you should still, even with all that context, give it the examples of what's good, what's bad, right?
00:27:14.720 --> 00:27:16.720
That's what we teach in our Prime Prompt Polish.
00:27:16.799 --> 00:27:27.039
Uh, that is the polish portion, is giving it kind of that uh multi-shot uh kind of um work, right?
00:27:27.200 --> 00:27:32.240
In the same way, think you're trading, you're training that employee, go back to that analogy that still works so well.
00:27:32.480 --> 00:27:41.839
The very first time they hand in their first project, their first assignment, their first deliverable, you're probably gonna go through and sit with them and say, hey, this is great because blank.
00:27:42.000 --> 00:27:45.119
Hey, this is incorrect because blank, right?
00:27:45.279 --> 00:27:48.240
So same thing that's the you know, few shot examples.
00:27:48.400 --> 00:27:52.240
Uh the second one, uh the second technique is rubric first.
00:27:52.640 --> 00:28:00.240
So give, you know, whatever large language model that you're working with a grading criteria in the context window before you even start working.
00:28:00.480 --> 00:28:00.960
Right.
00:28:01.119 --> 00:28:05.680
Uh, this is something, again, that we used to teach in our older PPP Pro course.
00:28:05.759 --> 00:28:12.720
Uh, we don't teach it as much anymore because I don't think it's as uh as useful as it once was, but I think it's still, right?
00:28:12.799 --> 00:28:17.359
I call that uh, you know, temperature, uh, I think I called it no gauges, right?
00:28:17.519 --> 00:28:21.359
Uh you need to give the model something it can gauge or give it a temperature, right?
00:28:21.440 --> 00:28:22.640
And give it examples too.
00:28:22.799 --> 00:28:26.319
So say, hey, uh, let's just talk about creative writing, right?
00:28:26.960 --> 00:28:34.240
Write a sentence as, you know, plainly as possible, as boring as possible, and you say this is a one, right?
00:28:34.319 --> 00:28:35.359
On the creative scale.
00:28:35.519 --> 00:28:38.480
You write a sentence that's kind of creative, this is a five.
00:28:38.720 --> 00:28:41.519
Write the world's most creative sentence ever, that's a 10.
00:28:41.599 --> 00:28:43.839
But tell the model that, tell it why, right?
00:28:44.000 --> 00:28:45.119
And then there you go.
00:28:45.279 --> 00:28:47.759
You've just kind of created a rubric, right?
00:28:47.839 --> 00:28:51.839
So then at any point, you can say, hey, let's do this as a three.
00:28:52.000 --> 00:28:54.079
Hey, let's do this as an eight, right?
00:28:54.240 --> 00:28:56.559
Uh, and that can help you think of not just creative writing.
00:28:56.640 --> 00:29:03.039
That's just an easy one uh to think about, but uh, there's so many different gauges or temperatures that you can put into a model.
00:29:03.279 --> 00:29:05.839
And then last but not least, show, don't tell.
00:29:06.000 --> 00:29:11.599
So sometimes you just need to paste the exact format you want instead of describing it.
00:29:11.759 --> 00:29:27.440
I think the combination of describing what you want and then also uh giving it examples, you know, that's just another uh uh, you know, kind of uh piling on here or doubling up here on our first uh technique, which is few shot examples.
00:29:27.519 --> 00:29:45.440
But if you combine that with the kind of the show don't tell, I think that's a great way, especially if you want outputs formatted uh in a certain way, or if you want outputs to always include, you know, XYZ, uh just giving it examples of that is helpful, but then also giving it the exact formula, uh right.
00:29:45.519 --> 00:29:51.920
So I would always do multiple uh versions uh of this, but the show don't tell is extremely important.
00:29:53.039 --> 00:30:08.079
All right, so we've covered a lot in this start here series, but let me wrap by saying this the context is everything, right?
00:30:08.480 --> 00:30:14.480
I've always said your data is the differentiator, using AI doesn't matter, right?
00:30:14.880 --> 00:30:20.880
Uh it doesn't people always think like, oh, we're using AI, so we're ahead of the curve.
00:30:20.960 --> 00:30:21.680
No, you're not.
00:30:21.839 --> 00:30:22.480
Um, right.
00:30:22.640 --> 00:30:26.960
And prompt engineering doesn't matter as much anymore, right?
00:30:27.119 --> 00:30:36.000
Because now these models by default are doing a lot of that heavy lifting that really paid off in 2023 and early 2024.
00:30:36.160 --> 00:30:42.480
So you have to stop thinking about talking to the model a certain way.
00:30:42.720 --> 00:30:51.359
And actually, especially if you are non-technical, especially if you're not a you know heavy AI user, that's actually a good thing.
00:30:51.680 --> 00:31:07.119
Because I think earlier on, right, in 2023 and 2024, a lot of non-technical people were kind of scared off of using large language models because you know, earlier on, there, you know, this prompt engineering, it's everything, it's everything, it's everything.
00:31:07.200 --> 00:31:09.519
And people are like, oh, I'm not an engineer, right?
00:31:09.680 --> 00:31:11.039
It doesn't matter anymore.
00:31:11.200 --> 00:31:17.680
You don't have to, you know, understand, you know, chain of thought or, you know, any of these other prompting techniques anymore.
00:31:17.920 --> 00:31:26.960
You can talk to it just even like a lazy human, because if you have the context side, how you talk to the model is not as important.
00:31:27.279 --> 00:31:33.039
But let me just give you three last pieces of advice to wrap up everything we talked about.
00:31:33.200 --> 00:31:36.319
So stop starting with a question.
00:31:36.559 --> 00:31:41.519
Instead, start with a tiny context pack, extend on it from there.
00:31:41.759 --> 00:31:44.319
Next, reuse what works, right?
00:31:44.480 --> 00:31:58.319
Whether you want to create skills, create GPTs, create projects, it doesn't matter, but you need to reuse what works because you do need to put much more in on the front end on the context side than you think you might need.
00:31:58.559 --> 00:32:00.480
More and more in more context.
00:32:00.640 --> 00:32:06.559
As long as you understand and don't exceed the context window, context will not bite you in the butt.
00:32:06.799 --> 00:32:07.119
Right?
00:32:07.279 --> 00:32:10.319
It will bite you in the butt if you don't provide enough context.
00:32:10.480 --> 00:32:14.240
But you need to be smart and think in a scalable fashion.
00:32:14.400 --> 00:32:18.000
Uh and you need to reuse what works.
00:32:18.640 --> 00:32:28.240
And then last but not least, remember that expert level results come from a system that you can repeat every time, right?
00:32:28.319 --> 00:32:40.000
It's all about if you want the expert level outputs, there's a good chance if you're doing this the right way that you're wasting time by not reusing it.
00:32:40.160 --> 00:32:40.480
Right.
00:32:40.559 --> 00:32:43.200
Here's a here's an extra pro tip, right?
00:32:43.279 --> 00:32:52.720
Especially if you're using as an example, uh, well, any of the models, if you have personalization and memory on, um, go through.
00:32:52.960 --> 00:32:59.440
Ask the model, what are the things that I'm most that I'm using you for that is most repetitive?
00:32:59.680 --> 00:33:04.640
You know, what are some systems that I can build uh that I can reuse, right?
00:33:04.799 --> 00:33:22.240
You'll probably be surprised, especially, you know, there's been some recent updates in the last, I'd say, three to four months uh with the big three players there and improving how the memory, how um, you know, remembering your past chat conversations, it's actually really good now, right?
00:33:22.319 --> 00:33:26.240
When it first started to debut like 15, 18 months ago, it wasn't that good, right?
00:33:26.319 --> 00:33:34.720
But I'll say the last three months, uh, these AI chat bots, essentially their memory and what they know about you is really good and being able to recall past conversations.
00:33:34.880 --> 00:33:39.039
So just ask, what am I wasting the most time on, right?
00:33:39.359 --> 00:33:40.480
Doing every single day?
00:33:40.640 --> 00:33:42.880
What are some skills that I should be building?
00:33:43.039 --> 00:33:46.960
What are some projects that I should be uh, you know, building?
00:33:47.119 --> 00:33:49.440
What are GPTs that I should be using?
00:33:49.599 --> 00:33:49.920
Right.
00:33:50.079 --> 00:33:59.599
Don't never, never feel like you're uh, you know, stupid or anything by just saying to a large language model, I'm not sure.
00:33:59.759 --> 00:34:00.480
Because guess what?
00:34:00.640 --> 00:34:02.559
Large language models are smarter than us.
00:34:02.799 --> 00:34:03.440
They are.
00:34:03.599 --> 00:34:11.840
They can pick up those uh those little pieces that humans tend to miss or even things that you may miss about yourself.
00:34:12.079 --> 00:34:19.119
So uh actually a great way to fill in that context is just to ask a large language model, right?
00:34:19.199 --> 00:34:28.559
There's kind of this more uh you know popular uh you know trend of creating like a skills markdown file or a role markdown file.
00:34:28.800 --> 00:34:29.920
Ask any of the models.
00:34:30.079 --> 00:34:39.280
Hey, based on everything you know about me, help me build out some of these blocks that you can then use modularly and take them with you as well, right?
00:34:39.440 --> 00:34:42.400
Because I can guarantee, and I've gone through this process.
00:34:42.559 --> 00:35:00.000
I actually did it like one or two months ago, where I just went through that process in all of the models because chances are I use them for different things because they each have their different strengths, and then I combine them into one big uh, you know, skills uh file, role file, all these different things.
00:35:00.079 --> 00:35:04.719
And now I have these building blocks that I can use whenever I need them.
00:35:04.880 --> 00:35:09.440
All right, so that is how you get those expert level results.
00:35:09.599 --> 00:35:14.880
So I hope that this version of the Start Here series was helpful.
00:35:15.039 --> 00:35:20.079
If it was, make sure if you haven't already, go to startheareseries.com.
00:35:20.239 --> 00:35:35.360
That is going to give you free access to not just our context engineering course that's been taken by more than 15,000 business leaders called Prime Prompt Polish, but it is going to give you free access to our inner circle community.
00:35:35.519 --> 00:35:47.280
And then you can also go to the Start Here Series uh space inside of our community and go listen to and catch up on all of our Start Here series, all in one easy place.
00:35:47.440 --> 00:35:47.840
All right.
00:35:48.000 --> 00:35:49.119
I hope this was helpful.
00:35:49.280 --> 00:35:50.239
Thank you for tuning in.
00:35:50.400 --> 00:35:53.760
I hope to see you back tomorrow and every day for more everyday AI.
00:35:53.840 --> 00:35:54.400
Thanks, y'all.
