Vibe coding and low-code are making software development faster and more accessible than ever. At the ITdaily roundtable, we hear that the real shift isn’t in the technology, but in the people. Which profiles does the market still need, how do we train them, and how do we keep them afloat in teams that are becoming increasingly smaller and faster?
Seated around the table are Dirk Deridder, CTO at Smals, Pieter Vercammen, solution architect at Mendix, Ron Pooters, director of the Microsoft Innovation Hub in Brussels, and two representatives from Cegeka: Stijn Lambrechts (Global Delivery Lead Applications) and Bjorn Boisschot (Quality Engineering Manager).
The conversation begins with low-code and vibe coding, but quickly arrives at the question of what this does to the IT profession.
The profile is shifting
Everyone agrees that the developer’s work is changing. Ron Pooters of Microsoft primarily sees the nature of the work shifting. “The entire profile of a developer is going to change. You still need to know what you’re doing, but you can focus much more on solving problems and much less on writing repetitive code that you’ve already written five hundred times,” he says.
According to him, the role is shifting toward the business: “The developer is evolving into a kind of project manager who directs an army of agents. Each has their specialty, and that whole leads to better code.”
Pieter Vercammen of Mendix recognizes this shift. Where previously a business analyst sat next to the builder to explain what needed to happen, the developer is now becoming the person who understands what the business wants and knows how to direct his agents. Ron Pooters dares to go a step further.
“I think the developer and the business are going to grow much closer together. Over time, you might grow into a single profile: a product engineer who both knows what they want to build and has the tools to realize it.”
This new role primarily requires conceptual insight and critical thinking, Stijn Lambrechts adds. “The added value you still provide as an organization is determining for yourself when good is good enough,” he states. “Otherwise, you’ll over-develop and the application will soon cost far too much in maintenance; however, this cost must remain justified relative to the problem you are actually solving.”
How sure are you?
Vercammen: “The number one question you should ask an agent is: how sure are you of your answer? Sometimes they say: forty percent. And if we make decisions because an agent thinks they’re forty percent right, then I have my questions about that.”
Pooters puts it in the right context: we are still in an exploratory phase. “Scalability, logging, security, traceability—even a developer with twenty-five years of experience cannot simply make the right model choice today.”
Defining the specification
The crux remains the specification, Deridder believes. “How do you arrive at a good spec? A spec that is unambiguous, complete, and defines all exceptions in advance?” Vercammen adds laconically: “That has always been IT’s biggest problem.” The result, according to Deridder, is that the bottleneck shifts.
“We’re going to have developers driving a Formula 1 car to the next traffic light and then having to wait until the business has validated something,” he illustrates. Furthermore, the efficiency gain is partly eaten up by extra checks: “We’re going to have more guardrails, perform more verifications. Part of the gain you would have on the cost side is compensated by those additional checks with AI.”
For Lambrechts, this touches on a fundamental principle. “We like the separation of powers. Just as the police, the legislature, and the judiciary are separate, you want that in software development too. You don’t want to put writing specifications, developing, and testing in the hands of one person; otherwise, as an organization, you potentially create a quality risk that you’re not even aware of. Your entire service delivery cannot depend on one person if you want to guarantee continuity for your customer.”
Education is lagging behind
If the profile changes, the training must follow. That’s where the problem lies. Boisschot gives an example from his own home: “One of my children is in a management track. As part of her thesis, she had to transcribe interviews and was told by the school that they weren’t allowed to use AI, but had to type it out themselves. That is exactly something very simple and of little value to your studies. The school is missing the opportunity there to teach students how to handle AI responsibly.”
Boisschot shows understanding for the schools. “Structurally, it’s not illogical that they forbid it. A curriculum must be approved years in advance. I understand they can’t keep up with the speed of AI, but AI literacy will be crucial for their future.” Pooters therefore advocates for guest lectures and extra projects to challenge schools, “but it’s actually happening as we speak.”
Lambrechts sees the real challenge in the upskilling of those already at work. “The worst message we can give young people today is that IT offers no future,” he repeats. “But what training do you send someone to so they can keep up today? The courses currently available on the market are not yet at the level we actually want, so we have to create them ourselves and continuously adapt them to the new possibilities AI offers.”
“I have an intern in my team now, and seeing what they bring in terms of AI-first thinking is a wake-up call for my seniors. Juniors remain very important, but they shouldn’t come with old-school skills, but with an AI-first mindset.”
Smaller teams, but no shrinkage
The AI cost comes on top of the labor cost, and that forces organizations to make a choice: “Either we go to smaller teams for the same efficiency, or to more money for much higher efficiency.”
Smaller teams don’t necessarily mean shrinkage, Pooters counters. “Those teams develop faster, your lead time becomes shorter, so you can do more projects. And there will also be more demand for projects.” The legacy at insurers and banks, which no one dared touch for years, is now within reach. “That used to be unaffordable,” says Deridder, “but that is now starting to become an accelerator for industries that don’t even exist today.”
At the same time, Lambrechts warns of the human price. People who spent ten or fifteen years building a specific skill suddenly see it matched by someone with much less experience. “Their self-worth was built on that. Suddenly that skill becomes much less important, and people need time to process that,” he says. On top of that comes the cognitive load.
A thought process that an analyst used to take one, two, three days over, we now push through in four hours.
Stijn Lambrechts, Global Delivery Lead Applications at Cegeka
“A thought process that an analyst or developer used to take one, two, three days over, we now push through in four hours. Their brain is running at two hundred percent, and you can’t keep that up intensively for eight hours. We must be careful not to give our own people burnout.”
Vercammen extends the reasoning to the work organization itself. “Every company is trying to become more efficient and wants people to work ten times faster,” he says. “Sometimes I would really like to hear a company say: we’re only going for five times faster, and we’re giving our people half a day off a week.”
Yet enthusiasm prevails, and that is precisely what surprises Deridder. “I see enormous enthusiasm in that group to get started with AI. That is fundamentally different from before, when there was resistance,” he says. “I even see people taking personal subscriptions of two hundred and fifty euros a month in their free time, purely to experiment.” The output motivates: they are makers who see results immediately.
Not the Porsche, but the airplane
Those who only aim for speed are missing the point, according to the table. Lambrechts captures it in an image: “Previously you drove a BMW, now you can drive a Porsche.” The real leap isn’t in doing the same thing faster, but in processes you couldn’t even imagine before. You want to take the plane.
“Everyone says: we’re going to put money into AI,” Vercammen concludes. “No, you should think: how am I going to make my processes better? AI plays a role in that, but the goal is optimization, not the AI itself.”
The goal is always optimization, not the AI itself.
Peter Vercammen, Solutions Architect at Mendix
With that, a common thread emerges. The IT profile is converging toward a hybrid product engineer who is closer to the business. Vibe coding and low-code are powerful accelerators, but specification, governance, and guardrails remain what makes the difference between a fun prototype and a system that lasts fifteen years.
Education must become AI-first, in the classroom and in professional development. And while teams become smaller and faster, there is a risk that humans will no longer be able to keep up with the pace.
The technology can no longer be pushed back into the box; everyone agrees on that. The winners will be the organizations that take back control of their training programs and dare to rethink their processes. All while keeping their people, both young and experienced, on board. Those who see AI as a mere IT game to become faster and cheaper will, according to the table, pay the price sooner or later.