I’m a software engineer at heart. Consequently, a big chunk of my time revolves around writing code to solve problems. Lately, though, I’ve found myself putting aside the code-writing labor and everything else in text format. Still, I consider writing one of the most gratifying activities (especially once it is finished, of course). So what has changed? I’ve even lost interest in looking at poorly written new API documentation. I’m perceiving books, articles, and even tweets as too large and/or out of focus. Googling and reading Q&A threads in StackOverflow are less appealing than what they used to be. Definitely, something is happening to me, and it’s not just the usual slump. But at the same time, it looks like my laziness efficiency has increased considerably. Contradictory, uhm? After giving it some thought, I’m starting to understand what’s going on: I’m transitioning into a prompt engineer! Let’s think about it step by step.

Once upon a long time, software was a strange thing.¹ Even stranger was the people who build it: programmers. Or were they scientists? I’m too young to know exactly what they were called. Anyway, they were an unusual breed of people. Mathematicians, physicists, or pure engineers who mastered cryptic programming languages to tackle practical and scientific problems.² These individuals possessed advanced reasoning skills. They could understand complex systems written in COBOL (such as those for financial transactions), and implement algorithms like QuickSort in 5 minutes or less.

But that was a long time ago when software was still eating the world. Nowadays, software is everywhere. It already ate the world and digested it. Sometimes resulting in a mess that we still have to clean up. Anyway, its output is indispensable, just like salt for our food. In every industry, in every company, in every home, software is a necessary ingredient. So the crew of software cookers has grown at a rapid speed. But the world had not enough mathematicians, or engineers, to satisfy the demand for programmers.³ And as with any other industry, the higher the demand, the bigger the push humanity makes in order to make the production machinery more productive.

Over the past two decades, a new class of “programmers” has emerged and permeated the industry: software developers. The prerequisites have shifted dramatically. The average software developer doesn’t need to understand Fourier Series anymore to get a decent job offer. Indeed, it’s not even necessary to know how a computer actually works.⁴ It’s enough to master a particular framework, a library, or even a single digital tool (long live the Excel spreadsheet!). And that’s it. The wheels that keep software product lines keep going on.

But I’ve seen the panorama shifting in the last few months. What if I told you that knowing how to code is no longer necessary? Yes, that’s right Because now computers can generate code for us, automatically. I’m not talking about in-IDE code generation tools, such as templates or advanced refactoring tricks. I mean entire applications written by AI, from scratch. And not only code, everything that is digital can be AI-generated too. Even the boring stuff, like bug reports, code comments, and meeting minutes. The new era of automated text generation has come to us developers. And with it, there comes a new batch of fresh challenges.

Show Me the Prompt!

It seems that coding, as a profession, has become less complex nowadays. This is thanks to the rise of AI tools powered by Large Language Models (LLMs). These are Machine Learning (ML) models trained on huge amounts of crap text data. Once trained, the model can generate meaningful text responses to a given prompt. Interestingly, it appears that code generation is one of the most simple tasks for these models. So we developers are not as special as we thought, ouch. In other words, AI can write reasonably good code faster and cheaper than us. But this article is not about AI replacing developers (I wrote about it here). I believe that the price of every single line of code will decrease its value dramatically in the coming years. It’s supply and demand, there’s nothing we can do about it.

However, there’s a catch. A critical component is still essential to fuel the AI machinery: the prompts. Human expertise is necessary to properly craft the questions for the AI systems. Consequently, a new category of programmers has emerged. These are the individuals who specialize in creating effective prompts to tackle coding challenges. Let me introduce you to the prompt engineer.

The prompt engineer daily task is to ask fine-tuned questions. For that, sometimes it’s necessary to embody the inquisitive spirit of Socrates, or perhaps that of a 5-year-old child (depending on how much money you intend to make out of it). The job is fundamentally about crafting queries in natural language, for example: “I got the following log message, how to fix the problem?,” or “I have a list of items, write an algorithm to sort them?,” or “My boss doesn’t understand technical debt, what should I do?” The next step is straightforward, just taking the AI-generated response and integrating it into the appropriate context. Directly copy/pasting the response to the production environment is often just enough. Pure magic!

But perhaps this isn’t so magical after all. Doesn’t it look familiar to you? Yes, it’s the exact same thing we used to do every day when searching Google and looking at Q&A threads in StackOverflow. The difference now lies in the speed and quality of the responses we receive. We’re able to get answers without having to fully articulate our questions. It’s impressive. So, say goodbye to the fear of posting seemingly foolish questions on StackOverflow (RIP StackOverflow). There is no more shame because AI systems are still trained specifically for not trolling us anymore (thank you for that, Reinforcement Learning). What’s better, seems like AI is increasingly surpassing us in deciphering which our own challenges are.

Needless to say, this is the end of software engineering as we know it. I expect a huge impact on the tech workforce worldwide, as this technology scales very easily (consider OpenAI’s ChatGPT model, for example). What will I do with the extra time now that searching for answers is no longer necessary? Playing ping-pong with my colleagues? Or maybe I’ll just go back to the old days, when I used to read books and spend quality time with the family. A set of brand new human problems are coming to us.

On the other hand, the tide of public skepticism regarding the future of Computer Science (as a career) is rising. Formal Computer Science education is feeling the pressure. At the same time, the number of devs kicked out from leaving the industry is increasing. This is alarming to me.

But don’t worry. There is still some hope. The end is near but not here yet. We still have a few years left for us to adapt and redirect our skills. And maybe we can all keep our jobs after all. I’m talking, of course, about jumping into the new wave to become successful prompt engineers.

But first, let’s set the terms.

The Prompt Engineer Manifesto

The time has come for us, the successful prompt engineers to announce our manifesto.

Here it is:

“
We, the prompt engineers, are the new breed of programmers.
We are the experts in crafting prompts that empower AI to solve our problems.
We take all the credit from the solutions we derive.
We take the paychecks too.

We create value through innovative ideas.
We have regular meetings to brainstorm new ideas and create better prompts.
We break down complex problems into promptable smaller components.
We don’t write bugs, AI does.

We encourage a culture of lifelong prompting and skill enhancement.
We share our prompts with the community.
We have daily prompt reviews, where we blame AI for not understanding our problems.

We prompted an AI to help us write this manifesto.
We will keep improving it.
We will prompt a more powerful AI in the future to write it better.
”

Now that we have defined who we are, what we do, and what we don’t do, let’s explore how to become successful on the job.

The Successful Prompt Engineer

The successful prompt engineer has a clear goal in mind (you’re wrong, it’s not making more money). To reach the objective, we should really understand how AI actually works.⁵ This is necessary to craft the best prompts. The key skills to sharp involve language comprehension, understanding of the domain-specific keywords, and the context itself.

The successful prompt engineer knows how to apply the principles of clear communication. It’s about being concise, precise, and unambiguous. That way one can tweak the prompt to get the best results. Examples of techniques are summarizing, inferring, transforming, expanding, and iterating over an over the prompts until the AI gets it right. If in doubt, it may even involve calling another AI to write the prompt. Recursive prompting is called!

The successful prompter doesn’t give up easily. Sometimes one needs to experiment with different variations of prompts and evaluate the model’s responses. When the output is not so good, the successful prompt engineer knows when to flush the toilet.

The successful prompter leverages AI in favor of improving learning, coding, and creativity.

Prompting for Learning

The successful prompt engineer knows how to use AI to learn new stuff.

Here are five examples:

1. Use the Pareto Principle, which identifies the 20% of the topic that will yield 80% of the desired results, to create a focused learning plan for me.
2. Make a study schedule for all of the above in an appropriate number of weeks. I can study for X hours every Tuesday and Thursday. Include time for revision and testing.
3. Suggest various learning resources (like books, videos, podcasts, interactive exercises) for the above topics that cater to different learning styles to me.
4. List some beginner video game projects I could work on to help strengthen my C# coding skills. I’m a beginner interested in learning how to create games in Unity.
5. Guide me through a visualization exercise to help me internalize the term X and imagine myself successfully applying it to a real-life situation.

Prompting for Coding

The successful prompt engineer knows how to use AI to write better code.

Here are five examples:

1. Describe what the following code does. Explain it to me step-by-step in plain English.
2. Optimize this code to be more elegant, more readable, and more efficient.
3. Check for any hidden bug or vulnerability in the following code.
4. Write a JSON document for testing with 10 articles consisting of id, title, description, and timestamp.
5. Create an SQL script that inserts a row in the table users with content delimited by triple hashtags.

Prompting for Creativity

The successful prompt engineer knows how to apply the new stuff just learned to do something that cannot be done by prompting.

Here are five examples:

1. Make money value out of the following idea (e.g., “Can you provide a summary of the main points from the article X?”).
2. Create a new product or service based on the stuff just learned?
3. Humanize the stuff just learned (i.e., shaping it in a way for humans to understand).
4. Teach the stuff just learned to other colleagues so that they get to know how smart I am.
5. Write a blog post about the stuff just learned?

Prompts’ Wall of Fame

Here are some of my favorite prompts, most of them are inspired or modified from awesome-chatgpt-prompts:

Conclusion

The end of software engineering, as we know it, is here. AI has knocked on our door and entered our workplace. But AI still needs us. Prompt engineering is the brand-new coding mode for us developers. Not only to write software, but to solve many other issues. So in the era when data is gold, prompt engineers could be the miners. Stay tuned because we’ll soon start prompting with our voice, images, videos, everything. Websites will feature their own AI system, optimized specifically for their platform. The future is as uncertain as it is exciting.

External Resources

• awesome-chatgpt-prompts
• Prompt-Engineering-Guide
• awesome-chatgpt
• learnprompting.org
• ChatGPT Prompt Engineering for Developers

Footnotes