Software has been eating the world and now AI is eating software. Rix Groenboom wonders how ICT education programs can still fulfill their role. Since the Covid pandemic, everybody knows what exponential growth means. If R is above 1, something serious is about to happen, much faster than most people expect. That’s what we’re seeing with generative AI. Every few months, the models can do things they couldn’t do before, most recently with a massive impact on security and reverse engineering. For ICT education, this produces an assignment straight out of Tom Cruise movies: Rebuild a four-year program for a profession that reinvents itself every six months. Make sure graduates find work while their starter jobs dry up. And convince a 17-year-old that software engineering is still worth studying. According to Randstad, Dutch starter vacancies in tech are down roughly 35 percent since early 2024. After twenty years in software quality and testing, I know what we gave juniors to cut their teeth on: the bug fix, the CRUD screen, the unit tests nobody wanted to write. An agent now does that in minutes. We train HBO ICT students for the bottom rung of the ladder, while that rung is being cut off. Then there’s the image problem. De Volkskrant declared hand-coding programmers dinosaurs, De Groene Amsterdammer put a robot eating its own programmers on its cover and De Morgen had experts predicting that a fifth of all knowledge jobs will be gone within four years. Parents read it, teachers repeat it, and applications drop. Fewer students come in, fewer junior jobs wait at the other end Fewer students come in, fewer junior jobs wait at the other end and in between, we have to keep up with a technology whose R is well above 1, in institutions that change linearly at best since we ship curricula in waterfall. So, the gap widens every year. If the bottom rung is gone, graduates have to start one or two rungs higher: breaking a problem down, specifying it well enough for an agent to build, checking what comes back, testing, integration and owning the result in production. Requirements, test design and architecture, the subjects we pushed aside in the name of Agile, are becoming the job description. There’s a catch. Judgment is the result of doing the hard work. Nobody learns to spot a race condition by reading prompts; you learn it by writing one and losing a night hunting it down. That used to happen in the junior years, on the employer’s payroll. If those years disappear, education has to provide them. However, there might be a way upward: We can borrow ideas from the chip industry: the cleanroom. In Collabspace in Groningen, we’re building a software cleanroom where MBO, HBO and WO students write serious software with professionals, with security and compliance built in from day one, with and without AI. It runs on real projects, not fixed course descriptions, so it can change every semester. The next step is getting companies to bring us their real problems. For industry, that’s a great deal: We take over part of the junior phase. Graduates start one rung up, having already debugged real systems and reviewed AI-generated code against a spec. At the same time, we have a compelling story for a 17-year-old: Already in your first year, you’ll work on real systems for real companies. So, my proposal is two-fold. Curriculum designers: Stop debating which tools to add and start teaching the rung above. Industry: bring your real problems, your engineers as mentors and enough trust to open up your code. Embedded and safety-critical software doesn’t ship on vibes, and your seniors of 2035 have to start from somewhere. Shout it from the rooftops, at open days and in the papers: You still need programmers. Together, the mission isn’t impossible.