Audrey Woods, MIT CSAIL Alliances | May 22, 2026
As AI tools reshape what it means to write software, a hard question is emerging for educators: what do you teach the next generation of computer scientists in the world of AI?
Armando Solar-Lezama, Associate Director and COO of Massachusetts Institute of Technology Computer Science and AI Laboratory (MIT CSAIL) and Professor in MIT’s Electrical Engineering and Computer Science (EECS) argues on the CSAIL Alliances podcast that the answer starts with understanding what AI will and won't be able to do.
"The models are very good already, and they're going to get way better at anything that can be quickly and easily checked because it's very easy to go and train a model to get better at pretty much any task that you can very quickly look at the answer and say, 'Yep, that's a correct answer.'" The skills that will matter going forward are the subjective ones, like design, scalability thinking, and architectural judgment.
The irony is that computer science education has historically optimized for exactly the kind of work AI is about to automate. "When we teach, especially when we teach introductory classes, we tend to focus a lot on the things that are easy to grade, and the things that are easy to check if the student did the right answer, because we have lots of students and we want to grade all of them quickly," Professor Solar-Lezama explains. "I think we will have to see a transformation in how we teach our students, focusing much more on those skills that are difficult to test right away and where you require some of that experience in order to see that something maybe is not the right thing."
Another complication of an AI-fueled economy is what level of readiness graduates need to have on day one. "Before, we just had to get people to the level where they could do some of the simple tasks, and hope that they learn the more complicated things on the job," Professor Solar-Lezama says. "I think now we're going to have a responsibility to make sure that by the time people graduate, they have more of this experience. And they've built more of these intuitions to really be able to work effectively at these more supervisory levels that people will now have to work on from day one."
So what experience, exactly, do graduating programmers need? Professor Solar-Lezama says AI tools will be most powerful in the hands of those who can tell when they're wrong. "The people who are able to make the best use of this technology, and the people who get the biggest productivity boosts from the technology, are the people who already have a very solid foundation on the basics. The people who can very quickly tell, 'Nope, that thing that the tool is telling me, that doesn't look right.'" It's not enough to know the material, it has to be instinctive. "It is very important to build that foundation, not just of knowledge for the students, but to really push them to the point where that knowledge is at their fingertips, and it's intuitive, and they can recognize when something is right or something is wrong. Even on the things that the models can mostly do or can do really well, if you want to be a partner to the model as opposed to just an observer who hopes the model does it right, then you have to be at its level. You have to actually understand those things that the model understands."
Educational institutions around the country are already taking note, offering new courses, programs, and even majors such as MIT’s new “artificial intelligence and decision-making” program, which is now the second-most-popular undergraduate major. Driven by the popularity of AI tools, student interest in “future-proof” degrees, and massive industry investments from tech giants, universities are rapidly launching specialized AI departments to meet soaring demand.
While the ultimate change of computer science education remains uncertain, the trajectory Professor Solar-Lezama describes points toward less emphasis on tasks machines can check and more on judgment, design sense, and long-horizon thinking. In other words, things humans are still better equipped to do.
Professor Armando Solar-Lezama is the Associate Director and COO of MIT CSAIL and a Professor in MIT EECS. Listen to his conversation with Kara Miller on the MIT CSAIL Alliances podcast for more.
