Has AI killed programming - or merely changed what it means to be a programmer?
by Marcus Birkenkrahe
I am currently teaching computer and data science at a small liberal arts college in the US. AI definitely has not replaced programming at my school or at any other school that I know of. I have experimented with AI in the classroom in many ways. My overall experience is that students who are genuinely interested in computer and data science have no interest in replacing their coding skills with AI. They are happy to take the occasional hint or accept some help.
At one point, I thought that they perhaps were giving up too early, but I no longer think that’s true. They seem quite conscious of the fact that AI could shorten the path to a complete solution, but they do not want to take that shortcut. For these students, AI is more of a continuously available, unloved tutor. For other students, it replaces all their own efforts — but these students simply should not study computer or data science. They’re not in it for the learning but only for the winning, and they do not mind if the machine wins for them.
As for the world outside of school: all my former students report using AI in production. One said that he cannot remember what he did without it. These were good students, and they learned to program “the old way” (without an early AI exit). But AI has not replaced their jobs; it has only altered them. They got these jobs not because they could program, but because they could identify, understand, and solve problems using programming. They still do this, but now they use AI on the side, or as part of an agentic workflow, depending on the task.
None of what I hear from industry sounds like programming has been killed. At most, it has culled the playing field for the better: students who choose to learn programming now know that their jobs will look different, and that they may be asked to read, understand, and debug more code than they write. But writing code has been a team effort for a long time. Large projects are rarely created without a legacy foundation or multiple legacy components. There’s a new team member now: AI. It cannot always be trusted, and it tends to overreach itself — not unlike any other junior developer. Will it grow up and mature? Probably not. So you may save some time and spend it with your loved ones — or, more likely, with the messed-up code of your AI teammate.
As for me: I use AI a lot inside and outside of class, and I often show students how to use it effectively, especially how to avoid deskilling too rapidly or deluding themselves into thinking that they know something when actually only the machine knows it. This requires not more, but different, techniques to stay ahead of the deterioration of one’s mind. But I trained as a physicist, and I learned during my years in school and later at work that no tool comes without a price.
My own prediction is that wider adoption of AI in its current state of development will lead to demand for far more software engineers and people with programming skills, not fewer. We’re entering the age of personal software, where everyone can build their own applications without having sufficient skills to keep them safe, keep them running, and keep them up to date. If these skills shift even further toward more advanced AI, then the skills needed to maintain this software — the software that does the job for others — will remain in demand. Additional layers of abstraction do not eliminate human guardians; they only shift their work to higher levels of abstraction. They do not eliminate the lower layers.
AI is changing programming workflows without eliminating the need for human expertise, though it is changing the type of expertise that is needed. The new expertise lies less in producing code directly than in judging, supervising, integrating, and taking responsibility for increasingly autonomous systems.