Prompt and Context Engineering 101 for ChatGPT & Claude

ChatGPT and other generative AI tools can serve as your own personal learning assistant!

In this section of the course, we will explore various prompting techniques so that you can learn new topics with the help of your new AI learning assistant. We will accomplish this by employing three common learning methodologies with the follow focus:

• Subject-Focused

• Problem-Focused

• Learner-Focused

Let’s engineer prompts to learn something new, starting with subject-focused prompting, starting in the next lesson!

Subject-focused prompting allows us to steer ChatGPT toward the heart of the subject you want to study. This is ideal when you're learning a new topic from scratch. Subject-focused prompting involves explicitly mentioning the desired subject or providing context related to that subject within the conversation. This helps ensure that the model's responses stay relevant to the intended topic.

If you could sum up our prior lesson’s topic of subject-focused prompting, you would characterize those types of prompts as something along the lines of “Tell me more about____ fill in the blank. For this lesson we are now dealing with problem-focused learning prompts. The key here is that we are trying to come up with solutions to our problems, and to get to the heart of the matter, we need to get specific with ChatGPT regarding what our problems we are trying to solve are.

Learner-focused prompts, also known as the Learner-Centered Approach, prioritize your interests, needs, and individual learning styles. This approach aims to personalize your learning experience by adapting to your pace, interests, and preferred ways of learning. With learner-focused prompts, you are encouraged to actively engage with the content generated by ChatGPT and explore topics that align with your personal interests. This not only enhances your motivation but also greatly improves retention when learning something new.

Let’s walk through several examples of prompts you could use to elicit learner-focused responses from ChatGPT, encouraging open-ended exploration and reflective questioning. For demonstration purposes, I am going to be prompting ChatGPT learner-focused prompts in order to learn more about the topic of Chess. I encourage you to follow along in your own ChatGPT account and feel free to replace Chess with your own topic of interest. And don’t forget to access my specific prompts that I have shared in the resources! Let me share my screen and we’ll walk through several learner-focused prompts together!

In this section of the course, you’ll be introduced to several different types of prompts.

Here we will learn zero-shot prompts, one-shot prompts, few shot prompts and in addition to these various ‘shot’ approaches to prompting, we will explore Chain of Thought prompting.

Chain of Thought prompts are often represented as the acronym C.O.T or COT.

Becoming familiar with these different prompt types will help you elicit desired responses from an AI such as ChatGPT.

So let’s start with chain of thought prompts, starting in the next lesson!

You may have often performed chain of thought prompts without even realizing it. As you prompt and receive responses from ChatGPT, this can often result in new thoughts and questions coming to your mind. You can follow your fancy and dive deeper into different paths and branches inside of the same chat and end up far afield from where you originated.

This phenomenon has been called many things, such as a prompt wheel, a prompt chain, but what seems to be coming into wider acceptance is the term chain-of-thought prompts, often shortened to CoT.

Let’s build upon our work in the previous section where we explored the different focused prompts for learning new topics. As a reminder, those were:

• subject-focused

• problem- focused

• learner-focused prompts

While I could simply demo a single example of a subject-focused CoT prompt, instead, I want to show you how you can prompt ChatGPT to generate an unlimited number of examples for you. The beauty of this approach is not only do you get many examples to learn from, but this builds you an ever increasingly valuable prompt library of your own as your chats are saved in your account history for later exploits!

And I encourage you to follow along in your own account, either performing the same prompts as myself, or you can always grab the prompt link from the resources section and continue the conversation in your own account.

A zero-shot prompt provides no prior examples to ChatGPT or any other AI tool, and it's expected to understand the task solely from the prompt. A zero-shot prompt may fall outside its training data. The AI tool is expected to generate a meaningful response or perform a task without being necessarily trained on that particular prompt. It leverages its understanding of language and general knowledge to generate relevant outputs, making inferences or providing answers based on its pre-existing knowledge. Zero-shot prompts enable the model to generalize and perform tasks beyond its explicitly trained capabilities.

Let’s prompt ChatGPT for examples of Zero-Shot Prompts!

Zero Shot Chain of Thought (Zero-shot-CoT) prompting involves a slight tweak to a simple zero-shot prompt, such as appending the words "Let's think step by step." to the end of your zero shot prompt. This guides ChatGPT to generate a chain of thought to answer your question. This zero shot chain of thought approach helps extract more complete and thorough answers without the need for specific prior training.

So ‘let’s think step by step’ with ChatGPT

The model is given just one example of the results you want in a one shot prompt, which is explored in this lesson.

The opening goal: spawn four agent teams that read eight specification documents, build four production tools in parallel, then a QA team, summary agent, and Canva agent that produces a branded LinkedIn carousel. This is the picture of what a real agent team does, before any code.

I dump the full Claude Code session into Claude on the chat side and ask it to find the false assumptions and the moment the quality fell off a cliff. This is the retrospective discipline that turns a frustrating session into durable lessons learned, before the next agent team gets spawned.

Cowork is Anthropic's bridge between Claude chat and Claude Code. I show how it tracks progress, holds context, and lets me drop back into chat mid-build for assessment.

The Canva API will accept long generation prompts and time out. The fix is batching: max five slides per generation call, max 90 characters per slide description, content injected separately. The quality cliff was self-inflicted by ignoring the API limits.

The Canva API can list and apply brand kits but cannot create or modify them. You build brand kits inside Canva's UI first, then reference them by ID from Claude Code. Skipping this step guarantees inconsistent results. Up to 100 brand kits are possible on a Canva Pro account.

Walking through the Canva outline review flow: audience settings, style options, brand kit selection. The connector returns a structured prompt with the brand kit ID embedded.

When Canva generation times out on a 17-slide presentation, what does an agent team fall back to? An HTML or PowerPoint export path so the work isn't lost. Four-tier retry logic: structured generation, smaller batch, unstructured fallback, HTML fallback.

Chain of thought prompting with Canva: go back and forth between Claude and Canva, creating one batch, selecting from candidates, injecting content, then moving to the next batch. This is how the human at the helm maintains taste and quality control across an agent team's output.

Why I write skills, Agent Team Requirements Documents, and design docs even when agents could write them too. Durable rules survive the next model upgrade. The agent's job is to execute taste, not to invent it. Notes on what separates a craftsman from a prompt-engineer.

One Claude Code prompt spawns the entire team: read the eight specs, build the four production tools in parallel, run QA, write the executive summary, and generate the Canva carousel. The one-liner is the surface; the Agent Team Requirements Document framework is what makes it repeatable.

Inside the build: the Writer phase generates content against the requirements, running on my Mac Studio with local compute. Token cost surfaced in real time. What it actually looks like when an agent team works in parallel on a single machine.

Closing the loop: review the work, edit where taste calls for it, commit the changes. The token cost of the run is shown so the economics are concrete, not abstract. The takeaway: agents do the volume, the human owns the taste.