WIP: Software Engineering Competencies in the Age of AI

Imagine the world of software engineering as a massive, bustling construction site. For decades, the blueprints for training new builders (university students) have been carefully drawn up by a group of strict architects called "accreditation bodies." These blueprints, like the SE2014 guidelines, told schools exactly what tools to teach: how to lay bricks, how to mix concrete, and how to follow the rules of the building code.

But then, a new kind of magic tool appeared on the site: Artificial Intelligence (AI). It's like a robot assistant that can instantly draft walls or suggest the best way to fix a leak. The problem? The old blueprints haven't been updated to include this robot. The paper by Vonderhaar, Towhidnejad, and Ochoa argues that if we keep training builders using the old maps, they will graduate ready to build with hammers, only to find the industry is now running on robot-assisted construction.

The Missing Piece in the Toolbox
The authors looked at the latest industry reports and job postings and found a glaring gap. While the big industry guidebook, called SWEBOK, has finally admitted that AI is a "foundational" part of the job (a huge change from its 2014 version which didn't mention it at all), the actual school curriculums and the competency model called SWECOM are still stuck in the past.

The paper suggests that universities are currently sending graduates into the "Golden Age of AI" without the right survival gear. It's like sending a sailor into the age of jet planes without teaching them how to read a radar. The authors aren't saying the old skills are useless, but they are suggesting that the current training is incomplete.

Two New Superpowers Needed
Based on their review of what companies are actually asking for, the authors propose that we need to add two brand-new "superpowers" to the software engineer's training manual. They aren't just suggesting random ideas; they are pulling these directly from what the literature says industry needs right now.

1. AI Literacy (The "Robot Whisperer" Skill):
   This isn't about building the robot; it's about knowing how to talk to it. The paper suggests that even if a student is just writing normal code, they need to know how to use AI tools like Large Language Models (LLMs) effectively.

   • What it looks like: It's like learning "prompt engineering"—the art of asking the robot the right questions to get a good answer.
   • The Catch: The authors note that if you use these tools without training, you might actually work slower. You also need to know how to check the robot's work for "hallucinations" (when the robot makes things up) and bias. It's about being a smart editor, not just a passive user.

2. AI Development (The "Robot Builder" Skill):
   This is for the students who want to build the AI systems themselves. The paper suggests this is a much heavier lift. It's not just about knowing a little bit of math; it requires a whole new set of tools.

   • What it looks like: This includes learning how to clean up messy data (like organizing a chaotic garage before building a car), understanding how to train models, and knowing how to test them.
   • The Twist: The authors point out that building AI is different from building normal software. You can't just plan the whole thing at the start because the AI might change its mind as it learns. This requires a different kind of testing and a focus on ethics that traditional software didn't always need.

What This Paper Is NOT Saying
It's important to know what this paper doesn't claim. The authors are not saying that every single software engineer needs to be a master data scientist. They are not saying that the old way of building software is dead. They are also not claiming that universities have already fixed this problem. In fact, they explicitly state that this is a "Work-In-Progress" (WIP) paper.

They are not presenting a finished, solved puzzle. Instead, they are holding up a mirror and saying, "Hey, look at this hole in our training plan." They suggest that the current guidelines (like SE2014) are outdated because they don't mention AI, and they argue that the competency model (SWECOM) needs to be updated to include these new skills.

The Next Steps
The authors are essentially saying, "We have identified the gap, and here is a suggested list of skills to fill it." They are suggesting that we add these two new areas—AI Literacy and AI Development—to the official list of what a software engineer should know.

However, they admit there are some big hurdles ahead. They note that AI changes incredibly fast, so keeping the curriculum up-to-date will be a constant race. They also worry that some schools might not have teachers who know enough about AI to teach it, and that adding these new topics might squeeze out other important subjects.

So, the paper doesn't offer a magic wand that fixes everything today. Instead, it offers a roadmap. It suggests that to survive in the modern workforce, the next generation of software engineers needs to be fluent in both the old ways of building and the new ways of collaborating with AI. The authors are proposing that we update the rulebook to make sure these students are ready for the job that actually exists, not the job that existed ten years ago.