AI Writes the Code, So Why Are Berkeley's CS Grades Slipping?

AI can write your code now. So here’s the strange part: at top American universities, computer science students’ grades and math skills seem to be heading the wrong way. The tools got dramatically better, yet human ability is sliding backward. Today, let’s unpack that paradox.

A quick note up front, because honesty matters. There isn’t a flood of fresh, hard data circulating in the developer community on this specific question over the past month. So rather than waving around a precise statistic and pretending it’s settled, I want to focus on the structural pattern that keeps surfacing — in classrooms and on Hacker News threads alike. The phenomenon matters more than any single number.

Coding Got Easier, the Foundations Got Shakier

The environment a CS student learns in today looks nothing like it did five years ago. Toss an assignment at a model and working code comes back in seconds. Data structure implementations, algorithm solutions, even the skeleton of a math proof — the AI drafts it all.

That’s where the trouble starts. The output is fine. But the process of getting there never runs through the student’s own head. Why the code behaves the way it does, why one algorithm beats another, why a single line of a proof holds — the stage where you internalize all of that simply vanishes.

Programming skill was traditionally forged in the act of getting stuck, debugging, and pulling your hair out. That painful grind was the actual learning. The moment AI handles the pain for you, you keep the convenience but lose the channel through which skill accumulates.

The Trap of Outsourced Learning

I’d call this outsourced learning. Just as a company farms out its non-core functions, a student hands the “thinking” part to a machine.

The danger of outsourcing is well documented. Send a function outside and your internal capability in that area quietly withers. It looks efficient at first — until you’ve lost even the ability to judge whether the outsourced work is right or wrong. Learning works the same way.

Math takes the hardest hit. With code, you can at least run the AI’s output and see if it works. But mathematical reasoning is invisible. Calculus, linear algebra, discrete math — copying down the answer does nothing for you. The intuition only comes from working it by hand, getting stuck, and grinding through again. Skip that loop, and the moment you sit in an exam room without AI, you collapse. The grade drop is just the symptom.

Why AI Can Be Poison for Learning Specifically

Don’t misread me. The argument isn’t that AI is bad. For an expert who already has the chops, it’s a powerful amplifier. The problem is what happens when a learner with no foundation yet reaches for it.

There’s a subtle trap here. AI-generated answers usually look plausible. So the student slides into the illusion of “yeah, I get this.” In reality they read it and nodded along — they couldn’t produce it themselves. Cognitive psychologists call this the fluency illusion. Looking familiar and actually being able to do it are entirely different things.

People raised the same alarm when the calculator arrived. But there’s a decisive difference. The calculator replaced one narrow chore: arithmetic. Today’s AI replaces the entire arc of thinking — understanding the problem, structuring it, designing the solution. The dimension of what’s being substituted is on another level.

So How Should You Actually Use It

The answer isn’t “don’t use AI.” That’s both impossible and a fight against the current. The key is sequence.

Struggle through it yourself first, then open the AI. Compare the answer you sweated over for an hour against the model’s, and learning detonates. Ask the AI before you’ve struggled at all, and the problem never becomes yours.

Classrooms have to adapt too. Some universities are dialing handwritten work back up on exams and leaning on oral assessments to verify whether someone actually understands — same logic. It’s a signal that they intend to grade the process, not the artifact.

AI is a tool that takes you farther. But climb into a car before you’ve learned to walk, and the moment you step out, you can’t take a single step. So ask yourself: are you using AI to build your skill, or to replace it? How you answer that may decide the next five years.