The current hype around Generative AI (GenAI) has led many to believe it’s a mythical genie that grants wishes on demand. However, the reality of working with these models is far more nuanced. As a professional software engineer, I’ve come to see GenAI as a powerful tool that requires time, practice, and a methodical approach to master. In this post, I’d like to share my personal journey into the world of prompt engineering and the lessons I’ve learned along the way.
Why Prompt Engineering Matters
As humans, we often have a clear idea of what we want to achieve and the steps needed to get there, based on our experiences. For me, this usually involves writing code to solve problems—whether it’s creating API endpoints or automating workflows. However, when working with language models, every prompt you make in a new session starts from scratch. Unlike humans, these models lack historical context, making it essential to provide clear, detailed instructions every time.
Challenges in Prompt Engineering
One of the hardest parts of prompt engineering is defining exactly what you want to achieve. For example, creating a function with specific inputs and outputs sounds simple, but crafting a prompt that is flexible yet specific enough to apply to a wide range of situations can be surprisingly difficult. That’s where techniques like SMART goals and attribute-driven design principles come into play.
Key Attributes of Effective Prompts
To create reusable and effective instructions, prompts must meet the following criteria:
-
Accurate: Clearly state the desired outcome. -
Unambiguous: Avoid confusion or multiple interpretations. -
Concise: Keep it short and to the point. -
Actionable: Provide clear steps that the model can follow.
These attributes mirror the qualities of a well-designed programming language, making prompt engineering an interesting intersection of technical and linguistic skills.
My Goal: Automating the Software Development Lifecycle
My ultimate goal is to use tools like GitHub Copilot to automate the software development lifecycle (SDLC) from a technical architecture document to a deployable application. This includes:
- Selecting desired architectural services.
- Creating actionable tasks on GitHub Project.
- Writing Infrastructure-as-Code (IaC) based on tasks.
- Implementing actual business logic.
Tools and Approach
Here’s the setup I’m using to tackle this challenge:
-
Tools:
- A Confluence page or Markdown file as the source for the technical architecture.
- GitHub Project for task management.
- VSCode with the GitHub Copilot plugin.
- One software engineer (me!).
-
Approach:
- Define a desired output—ideally something familiar.
- Initiate a brainstorming session using self-prompting techniques.
- Set the model’s role to focus on the specific problem.
- Provide enough context to the model.
- Ask the model to generate a few initial ideas.
- Guide the model through the process until the desired outcome is reached.
- Generate accurate, concise, and unambiguous instructions based on the steps performed.
What’s Next?
Prompt engineering is still a growing field, and there’s much to learn. I’d love to hear your thoughts! Do you agree with the key attributes I’ve outlined? Have you used similar techniques in your work? Share your experiences in the comments below!
Top comments (0)