Transcription
We've all heard of the term vibe coding, the ability to instantly prototype your apps using natural language as the AI dynamically generates all the code for your project. This concept was first coined by Andre Kaparthy, and it perfectly captured the emerging shift in how we build software.
Now, Kaparthy has also introduced another powerful concept called context engineering. Recently, Shopify CEO Toby made a statement on Twitter that resonated with many in the AI community, where he stated that "I've really like the term context engineering over prompt engineering. It describes the core skills better, the art of providing all the context for the task to be plausibly solvable by the large language model."
Kaparthy had then expanded on this, emphasizing that "it's no longer about writing clever prompts. It's about context engineering, which is where you carefully curate the right precise information for the large model to perform at its best. It's a complex blend of science, intuition, and system design that powers real-world AI applications far beyond what people call ChatGPT wrappers." We're entering this new phase where context engineering may become a core skill of AI-assisted development. Just as vibe coding revolutionized prototyping, context engineering is about to transform how we think, build, and collaborate with AI systems.
But let's take a step back and explain why this conversation even came up in the first place. There was a study done by Codo, and it focused on the state of AI code quality. They surveyed huge numbers of developers, and one thing stood out the most. Most developers, 76.4%, don't trust AI-generated code without human review. The main reason why is because hallucinations and mistakes still happen way too often.
Now, the problem isn't that AI coding is bad. The problem is when AI gets shipped without human review. This is because, right now, the biggest gap in AI coding tools is context. AI often doesn't have enough of it, and it completely misses it most of the times. Which is why on this channel, I showcased a couple of things like Taskmaster. You have context 7. And without these tools, the AI doesn't get the right context it needs for the AI to generate the correct code. That's why context engineering is something that matters. We don't just need better AI. We need better structure and better ways to feed the right information to the AI so that it can actually succeed with its generation.
This is why today I'm going to be showcasing context engineering, a template that was developed by Cole Medan, and he is absolutely amazing, guys. He is the one who had created this template, which will drastically improve your code generation with context engineering. I highly recommend that you take a look at his channel. He does a lot of in-depth tutorials on various sorts of AI tooling, as well as plugins and different concepts. But today, we're going to be showcasing context engineering in action and showcasing why it truly matters and how it will help you.
Now, just to briefly reiterate what context engineering is. It's a skill of carefully selecting, organizing, and even managing the right information that an AI or an AI agent needs at each step to perform a task efficiently and effectively without overwhelming it or missing critical details. And this is a visual way to see how it basically functions. With context engineering, you have all these different contexts that will be used. We first need to understand this definition because it's not a single prompt or something that you just send into the large language model. Context is everything the model sees before it actually generates. This entails the state, the history, the user prompt, the available tools, RAG instructions, as well as long-term memory. And with this full context, it's able to refer to all these things in one singular area to get the best structured output out of it. In essence, it's about getting the right balance and feeding the AI the necessary, useful, and structured information at the right time, step by step.
So, now that we have gotten all that out of the way, let's now take a look at context engineering, cuz it's way better than prompt engineering, like 10 times better, and easily 100 times better than vibe coding. Now, what this is going to help us do is that it's going to help us work with any AI coding assistant, but this particular template works with Claude Code. It's a template built to easily connect with Claude's strengths, and it's going to be able to help it become more flexible, be more precise with its generation. It's going to use less tokens, and you're essentially just going to get better generation from it.
Now, there are a couple of prerequisites that you'll need. Make sure you have Git installed to help us clone the repository. Node.js for the functionality of Claude Code. Make sure you have Claude Code installed as well. It's super easy to install. Just run the code for within your operating system's command prompt, and you can easily then get started with context 7.
Now, once all these prerequisites are fulfilled, head over to the GitHub repository of the template, and then you want to first clone this repository. So copy this first link, open your command prompt, and then paste it in to clone the repo. Once the repo has been cloned, you can then head into the context engineering intro to get a good familiar understanding of this template.
Before we get started, allow me to introduce today's video sponsor, Zapier. Zapier took AI automation to the next level with Zapier Agents. Not chatbots, but autonomous agents that quietly get real work done while you focus on what matters. If you're tired of bouncing between apps, managing different tasks, or losing precious time, well, Zapier Agents are like 24/7 AI teammates that can handle lead scoring, customer routing, content creation, and much more, all on its own. Zapier's agent templates make it easier to get started with ready-to-use frameworks that can automate your work across 7,000 apps in just a few clicks. These agents run in the background, proactively automating your workflows across multiple apps. They're simple to build. Just type what you want or grab a ready-to-use template like a viral content creation agent or a sales call analysis agent or even a travel booking organizer. There's something for every workflow. So, start building your own AI-powered dream team today. And this is something that you can access completely for free using my link below. Thanks again to Zapier for sponsoring today's video. But with that thought, let's get right back into today's video.
After cloning the repository, I want you to go ahead and open it up within your IDE so that you can easily preview and configure these different files. Now, what I want you to do first is head over to the Claude MD. This is essentially the global rules for your Claude to actually follow through. It is something that will define the project-wide rules, like how you would with client rules or with cursor rules. And essentially, it's going to guide the AI assistant to follow in every conversation, like the code structure that you can set. This is the base template that is already set, but you can configure things like the testing, as well as the reliability, task completion, system, and conversation. So you can easily configure this based off your own preference.
Next, you want to then create your initial feature request, and that is within the initial MD. Now, there's already an example one which showcases what you should provide. But within this initial MD file, this is essentially where you describe the features you want to build. So within the features tab, you insert a clear description of what you want the AI to focus on. Within the example tab, you can provide as many files as you want, and it'll reference it so that it could possibly build based off of your own context. And you can provide the examples within this folder over here. For the documentation, you can link it to relevant docs, APIs, or even MCP server resources. And for additional considerations, this is like edge cases or specific requirements that the AI could focus on.
So, now that we have set our global rules as well as our feature requests, what we're going to do next is work on generating the PRP, which is the product requirement prompt, and it's something that you probably might have heard of within Claude or within Cursor. It's essentially a PRD that will help craft a more specific instruction to the AI coding assistant. And you just need to simply run the `generate PRP initial.md` command to execute it within Claude Code to generate this PRP. And essentially, within the context engineering intro template, there are already these commands set to help you generate the PRP. These are slash commands that are custom commands defined in this commands folder. And it's something that will help you create the PRPs, as well as executing the PRPs, and the arguments, the variables that you will see after you execute it will reference the initial MD file after you have sent it in to generate the PRP.
So this is where, within my terminal, I can open up Claude Cloud Code, and what I can do is set my mode as well as whatever login I want by setting the API key. And once I've done that, I can then generate the PRP. So now that Claude Code has been authenticated, I can run this within Claude Code to generate the PRP based off the initial MD requirements I have set. So I can just simply paste this in and run this command. And it's going to take a while because it is going to thoroughly draft this PRP for us based off of all the references that we have provided. And right now, you can see that it is developing this PRP based off of the PRP template that Coleman had created, basing the PRP based off of all these requirements that were set within the template. And you can see right now it is working on creating this new PRP based off the initial MD that we had created.
Now, guys, what's amazing with this whole process, while it's generating the PRPs as well as working with the context engineering template, is how AI doesn't just generate the code, but it actually plans it thoroughly. You can see that this is a pretty detailed to-do list. It focuses on researching the APIs. It checks out the codebase that we give it, the review examples that we had provided, as well as reading the documentation that we had also set within the initial MD. And this right here is huge because one of the biggest issues with AI coding assistants is hallucinating, like we mentioned at the start. It is also not able to properly get the right API calls, or it misses critical details. But with context engineering, the AI takes care of the heavy lifting, doing the real research and building a readable and reliable plan before writing anything. It's not just creating a single file. It's actually creating it with confidence. And there we go. Our PRP is now fully developed. And we can now reference it within the PRP's folder. And you can access it as the multi-agent research email system. So, if I actually open this up as a preview, you're going to be able to look at our overall PRP that it was able to generate. It has our documentation and reference all the files that it needs, the links that we had provided within the example file. It also has the current codebase tree as well as the desired codebase tree with files to be added. So overall, this is going to reduce the hallucination immensely, and it's going to save you so much on token output, as well as making it super efficient to generate anything. This is why context engineering is really important, as it's a way to streamline this whole process of generating code.
Now, guys, since we have generated our PRP, now all we got to do is execute the PRP. And this is where it's going to easily code out our application or whatever we're generating, in this case, an AI agent. And you just need to simply run the `execute PRP` command, the folder of which the PRP generated in, and the name of the MD file that it created. So simply paste this in, and it's going to work thoroughly on now generating this application for us. This might take some time and it will take up a lot of tokens, so keep that in mind. But let's see what it's capable of generating for us. Looks like it first created the to-do list, which is creating the project directory, the requirements, implementing the agent from Pydantic, and then running all the validation commands to fix issues. So you can see this in-depth process of creating this application for us. It should take some time, but I'll be back once it's finished.
Before we get started, I just want to mention that you should definitely go ahead and subscribe to the World of AI newsletter. I'm constantly posting different newsletters on a weekly basis. So this is where you can easily get up-to-date knowledge about what is happening in the AI space. So definitely go ahead and subscribe, as this is completely for free.
Now, right now it is knocking down each task one by one. And this is something that is absolutely amazing, guys. It was able to thoroughly reference the to-do list, the PRP that we had created, and it is able to create high-quality outputs based off of that PRP. Now, if you were to single-shot send in a prompt to a model to do this, you wouldn't be able to do it at the same sort of quality due to hallucination, not able to reference the correct context or the examples. Whereas in this case, like we saw at the start, it pulls all the contents in one single area and feeds it equally through the large language model to process. And that is why it is going to be able to output the best contents in comparison to something that you would just send in all on its own with a single file. And just like that, our PRP is now fully implemented. And you can see the step-by-step process as to how you can set it up has been listed as well. You need to first go into your environments and you want to copy it, configure the API keys, dependencies. You just want to install it and then run the `python main.py` file, and then you can test it out using the `pytest`. But overall, you can see our AI agent has been fully created and constructed based off our PRP. But let's test it out. Also, it looks like it spent $3.32 for all of that to be fully configured.
And guys, now that I have configured my API key, I have now opened up our agent. Now, this is essentially where I have now created this multi-agent research and email system with a single PRP. And you can see that I'm able to research with it, where it can search with the Brave API. It can email using our Gmail. It's able to create the drafts, and the multi-agent is able to seamlessly have delegation between different agents. So let's now try out our multi-agent that it had created. We can ask it, "Create me an in-depth research on the world of AI," and we can process this request with the multi-agent that we had developed. You can see right now it is using the Brave search tool, the research agent itself, and it's going to be able to compile all the results to develop this research for us, and within a couple seconds, it'll output it over here. And guys, this was all fully constructed with the help of AI without me even doing anything. It did a pretty quick job, a pretty cheap job as well, and it was able to follow through due to the context engineering process where it's able to reference everything thoroughly. And you can see right now it is linking all the sources of what it found, and it has created this research on what the world of AI is.
If you like this video and would love to support the channel, you can consider donating to my channel through the Super Thanks option below. Or you can consider joining our private Discord where you can access multiple subscriptions to different AI tools for free on a monthly basis, plus daily AI news and exclusive content, plus a lot more.
Now, I really hope that you can understand a bit more of what context engineering entails and that it is something that will definitely help you elevate your coding experience. It's something that I highly recommend that you take a look at with all these different resources that I used in today's video in the description below. But that's basically it, guys. I hope you enjoyed today's video. Spent a lot of time today working on it, so I would greatly appreciate it if you can comment, like, and subscribe. It would mean the whole world to me. But with that thought, guys, thank you guys so much for watching. Make sure you subscribe to the second channel. Join our newsletter as well as our Discord. Follow me on Twitter. And lastly, make sure you guys subscribe, turn on the notification bell, like this video, and please take a look at our previous videos because there is a lot of content that you will truly benefit from. But with that thought, guys, thank you guys so much for watching. Have an amazing day. Spread positivity, and I'll see you guys fairly shortly. Peace out, fellas.