AI is drastically changing the work of developers. But those who think ChatGPT and Copilot will make programmers obsolete are mistaken. “We’re moving from writing code to checking code,” was stated during a roundtable at Smals.
With the rise of generative AI tools, software development seems to be entering a new phase. Companies are introducing various programming assistants designed to simplify and accelerate coding work for developers. Think of Anthropic’s Claude Code, OpenAI’s Codex AI-agent or Mistral’s Codestral. Prompting is becoming the new programming, but is it really that simple?
During a session of the AI Competence Center of Smals, the influence of AI programming assistants on the competencies and shifting role of developers was highlighted.
Changing Role
“You won’t lose your job as a developer because of AI,” reassures Joachim Ganseman, AI researcher at Smals. “Less time will be spent on coding, and more time on checking the code generated by AI,” he adds. “AI tools make mistakes, and removing those isn’t always easy.”
Ganseman draws a clear parallel with earlier waves of digitalization: “we’ve been automating many administrative tasks for 50 years, but that has never led to a reduction in administration,” he emphasizes.
Dylan Cabal, Solution Architect at Smals, shares the same opinion. “The profession is evolving and so we must evolve with it, but that doesn’t mean the role of developers will completely disappear.”
Concept over Syntax
Will AI influence the competencies of developers? That’s a question that often comes up in AI discussions. According to Cabal, it’s mainly the type of skills needed that changes: “Developers need to focus less on syntax and more on how to give AI the right instructions. Prompt engineering becomes more important.”
Koen Vanderkimpen, AI researcher at Smals, builds on this. “We need to remain experts in what we ask AI to do.” Learning the syntax of certain languages will become less important, but understanding the concepts remains crucial. Those who use AI tools must also know when they’re wrong. Because those errors are inevitably there.
AI often generates syntactically correct but semantically incorrect code. “Small differences like a missing letter or an error in a greater-than sign can have major consequences. That’s why human oversight remains crucial,” says Ganseman.
Young Developers
Experienced developers will quickly catch AI agents making mistakes, but for young developers, this isn’t so obvious yet. “AI assistants aren’t mature enough to let young developers work without supervision,” according to Sacha Thommes-Alexander, Senior IT Project manager at the Federal Pension Service. Juniors risk blindly trusting incorrect suggestions.
According to Vanderkimpen, it’s important to give young developers the opportunity to build competencies. “Don’t immediately set them loose with AI but let them reason, debug, and find errors,” adds Fabian Petitcolas, IT Research Consultant at Smals Research.
Arsenal of AI
Developers today have a choice from a growing arsenal of AI tools, from simple chatbots like ChatGPT to more advanced solutions like agentic IDEs. The latter can not only generate code but also compile, execute, and test within the development environment.
“We’ve evolved from asking questions in a separate chat window to integrated AI that acts independently in the development environment,” notes Cabal. “In three months, there could be something better. It’s moving incredibly fast,” emphasizes Ganseman.
Currently, the gentlemen around the table agree that the added value of AI today lies in repetitive and small tasks such as tests or documentation. “For small things, it works really well. But as soon as the context gets larger, errors start to appear,” says Ganseman.
What about Old Software?
What can AI mean for existing or legacy applications that are built monolithically, in a language that is no longer common, or by people who are no longer present? “There’s no magical solution that can rewrite the program in a new language,” begins Vanderkimpen. “What you can do is look for small pieces of code that you then have AI analyze. This gives you more understanding of the code.”
He emphasizes that it’s important to hold AI’s hand. “Clearly state in which new language you want the code and what the architecture is. Here too, it’s important that you can identify the errors yourself.”
Cabal also believes you should be careful when rewriting old software. “You need to be sure that the output is verifiable and identical to your starting point.” According to him, knowledge is often lost in this process. “We don’t know exactly everything the software did before it was rewritten. So you can’t be certain that after rewriting, the software does everything it did before.” By everything, Cabal means EVERYTHING, including the errors.
read also
AI Coding Tools are more Popular than ever, but Trust in Results is Declining
He points out that we shouldn’t make a circular argument: “I’m going to use AI to rewrite the documentation and then rewrite the software based on the documentation. Then we’ve used AI to check AI. It’s important that there’s always a human factor present,” emphasizes Cabal.
Vanderkimpen draws an analogy with an archaeologist. “When you’re searching through and improving legacy code, you’re actually an archaeologist, only you have more tools than just a small shovel. You get a bulldozer, but you have to make sure that bulldozer doesn’t sweep away important elements in your code.” With this comparison, he shows that the interaction between human and machine remains important.
Remain an Expert
The shift from code production to code checking is not only technically but also psychologically profound. “Those who write themselves understand their code better. With AI code, you often lose the intuition about where a bug is,” says Petitcolas. And that can lead to errors that are harder to track down.
That’s why documentation, testing, and quality control are more important than ever. AI can help speed up repetitive tasks such as generating test cases or documentation, but the direction must remain with humans. Ganseman summarizes: “AI is a good tool for developers, but remain the pilot of your project yourself”.
AI offers developers the chance to work more efficiently and smarter. But without critical thinking and sufficient domain knowledge, that same technology leads to errors, dependency, and risks.
This editorial contribution was created in collaboration with our partner Smals.