GitHub
Practical Prompt Engineering
3 hours, 43 minutes
CC
Course Description
Generate higher quality code from AI tools! Learn prompting techniques that work consistently across Claude, ChatGPT, Copilot, and Cursor. Transform vague project ideas into structure, actionable development plans. Stay productive by applying these future-proof prompting strategies as AI tools and models evolve.
This course and others like it are available as part of our Frontend Masters video subscription.
Preview
What They're Saying
Bri is an awesome teacher, very witty with examples to drill down concepts. I would say you would definitely be coming out of this course with practical understanding of LLMs, how it works and how to integrate it to your daily workflow, regardless of the usecase.
Ayman Yaqoub
Course Details
Published: October 24, 2025
Rating
Learning Paths
Learn Straight from the Experts Who Shape the Modern Web
Your Path to Senior Developer and Beyond
- 250+ In-depth courses
- 24 Learning Paths
- Industry Leading Experts
- Live Interactive Workshops
Table of Contents
Introduction
Section Duration: 5 minutes
Sabrina Goldfarb begins the course by sharing the origins of her interest in the science behind prompt engineering. The prompting techniques learned throughout the course will be used to create a client application for managing, rating, and exporting prompts. The application will be built completely from prompting—no code will be written.
Core Prompting Techniques
Section Duration: 37 minutes
Sabrina introduces LLMs, or Large Language Models. Prompt Engineering is the science behind craft prompts to return high-quality results from LLMs, which are non-deterministic or unpredictable. Transformer architectures are discussed, and Sabrina explains why scaling an LLM by 10x delivers 100x the capabilities.
Sabrina explains the controls available to make LLMs more or less deterministic. Temperature ranges from 0 (deterministic) to 2 (random). It controls how often the LLM will pick the next most likely token. Top P is an alternative to temperature and is used to remove potential answers from the dataset.
Sabrina highlights the importance of understanding tokens and the context window. Tokens are roughly .75 words on average. Since LLMs do not have a "memory", the full conversation is passed with each prompt so the LLM can "remember" what the conversation is about. Long conversations run the risk of filling the context window and can lead to hallucinations.
Sabrina introduces the "Standard Prompt," which is a direct question or instruction to the LLM. The quality of the question directly relates to the quality of the answer. Some examples of standard prompts are provided using Claude. Any AI/LLM tool can be used to follow along with these examples.
Building Better Prompts
Section Duration: 1 hour, 37 minutes
Sabrina walks through the tools and resources required to follow along with the course. All the prompts are available on the course website, which is linked below. Sabrina will be using GitHub Copilot inside VS Code. Other AI coding assistants like Claude Code, Cursor, Gemini, etc, can also be used.
Sabrina uses a standard prompt to create a Prompt Library application. The application was successfully created but it contained some unwanted features and functionality that didn't work. This demonstrated how AI agents can go beyond the scope and deliver results outside the original scope.
Sabrina introduces the zero-shot prompt. These prompts provide direct task requests without any examples. They work well for common tasks but rely entirely on the model's pre-training knowledge. A zero-shot prompt is used to recreate the Prompt Library application.
Sabrina shares some advice for selecting different models. Exploring alternative models can help improve accuracy or reduce costs in a application.
Sabrina compares one-shot prompting to zero-shot. With one-shot prompting, one example is provided with the request. The model learns the pattern, format, and style from the example, which establishes the format for future requests.
Sabrina writes a one-shot prompt to add a rating feature to the application. The prompt instructs the agent to analyze feature requests and provide the plan's technical requirements, code considerations, etc.. Once the plan is reviewed, a standard prompt is used to instruct the agent to implement the feature.
Sabrina introduces few-shot prompting. This technique provides two or more examples and edge cases. Models learn nuances and variations from a diverse set of inputs and outputs.
Sabrina asks Copilot to generate a few-shot prompt for implementing a notes feature. The examples provided ask for core requirements, implementation details, and deliverables. Once the prompt is generated, the agent is instructed to implement the feature.
Sabrina answers questions about generating few-shot prompts and managing the context window. She also shares a tip about typing "continue" to ask an agent to restart the output if it gets stuck.
Sabrina discusses why the placement of context matters. Providing context at the beginning and end of the prompt is much more effective than placing it in the middle.
Advanced Prompting Techniques
Section Duration: 1 hour, 21 minutes
Sabrina explains how to get a consistent format from LLMs by providing structured outputs. Specifying examples, templates, or schemas along with the prompt helps the model understand the desired output format. A few examples are demonstrated using Claude.
Sabrina prompts the agent with the structured output needed to implement a metadata tracking system for the Prompt Library application. Once the plan is reviewed the agent generates the metadata feature and the applicaiton is tested.
Sabrina walks through a chain-of-thought (COT) prompt example. chain-of-thought prompting asks the model to show it's reasoning step-by-step. This breaks complex problems down into intermediate steps and can be very effective when combined with the few-shot technique.
Sabrina uses the chain-of-thought prompting technique to implement and import/export features for the Prompt Library. After the code is generated, the features are tested. During the test, the wrong export files were selected, and Sabrina added an additional feature to support legacy formats.
Sabrina shares some advice for future-proofing prompts as models evolve. Documenting how prompts are used and the models where they are successful makes it easy to test them as new models are released. Also, recognizing that smaller models may work better with different prompting techniques than larger models helps identify how prompts should be adjusted for other models.
Sabrina shares research showing that LLMs can be enhanced by emotion. Emotional prompts caused the LLM to pay more attention to the more important parts of the original prompt, leading to more accurate results in the study. Sabrina also notes that this isn't universal across all models and can evolve.
Sabrina demonstrates how delimiters like quotes, dashes, XML tags, and markdown create boundaries and structure in prompts. This added structure allows LLMs to understand the prompt more easily and provides structure and readability to the output. Sabrina uses Claude to demonstrate using delimiters with a complex prompt.
Sabrina explains a technique for assigning a persona to a model. Personals instruct the model to identify with a specific role. They don't give the model extra capabilities but provide a perspective to steer the model toward a subset of data.
Wrapping Up
Section Duration: 1 minute
Sabrina wraps up the course by encouraging experimentation and exploration with prompting as new models are released.
Learn Straight from the Experts Who Shape the Modern Web
- 250+In-depth Courses
- Industry Leading Experts
- 24Learning Paths
- Live Interactive Workshops