Why Failing CS Grades Are Rising: What Parents Need To Know About AI, Math Skills, And College Success
Last month, I met my old college roommate Sarah for coffee at the spot we used to sneak to between 8 a.m. lectures, back when we were undergrads. She ordered a decaf, stirred it three times, then slid a folded midterm grade report across the sticky Formica table. Her son Liam, a junior in computer science at UC Berkeley, got an F in Algorithms.
“He aced AP Calculus and AP Comp Sci in high school,” Sarah said, staring at the sugar lump dissolving at the bottom of her cup. “He had a 4.0 coming out of his first year. What changed?”
Liam tagged along that day because he had to pick up a new tire for his bike, so I asked him straight out. He shrugged, leaning back in the plastic chair. “Half the class failed. Our prof talked for 20 minutes at the start of the next lecture about it. He said failing rates have doubled in three years, and it’s not because kids are dumber. It’s because everyone uses AI to do problem sets, and no one practices the actual math anymore.”
I tutor high school kids applying to CS programs these days, so this wasn’t some out-of-the-blue shock. Last week, I sat with a 17-year-old named Mia who was prepping for her AP Computer Science Principles exam. She pulled up a practice problem on quadratic equations, grabbed her phone to open Photomath before I could even blink. When I asked her to solve it step by step on paper, she stared at the blank page for five minutes and couldn’t remember how to get from the problem statement to the first line of work. She gets A’s on all her homework, but her practice test scores are stuck in the C range, and she can’t figure out why.
After that coffee with Sarah, I looked up the public data from Berkeley’s CS department. Professors there confirmed that incoming majors now score 15% lower on baseline math placement tests than they did five years ago, before widespread AI tools became standard for students. The share of students getting D’s and F’s in upper-division core classes jumped from around 7% in 2019 to 17% last year.
It’s easy to turn this into a “kids these days” rant, but that’s not what’s actually going on. Liam isn’t cheating, exactly. He’s got a part-time internship to pay his rent, he’s leading a club project that takes 10 hours a week, and each week’s problem set takes 6 hours if you do every step by hand. AI cuts that down to an hour. It’s not laziness—it’s triage. Kids are overloaded, tuition is sky-high, and they’re just trying to keep up.
But the problem is, math and coding skills are built by working through the messy, frustrating steps where you get stuck. You don’t build muscle memory by copying a correct answer from a tool. When you skip those steps, you don’t develop the ability to solve problems from scratch when you don’t have AI to help—like on a closed-book midterm, or when you’re working a real job and have to debug a problem no one has solved before.
I’ve started doing a super low-effort check with all my tutees, and Sarah started doing it with Liam when he’s home on break, that makes a huge difference. When they finish a homework set, ask them to pick one problem at random, put their phone and laptop away, and walk you through every step of how they solved it, out loud. It takes 5 minutes, and you don’t need to understand computer science or advanced math to do this. If they can explain each step and why they did it, they’re fine. If they stumble and can’t remember why they took that step, that’s a clear red flag that they just copied the answer.
Another small change that works: have them do the first 20 minutes of any problem set session with no AI allowed. They have to write down every step of the first two problems by hand, before they use any tool to check their work or speed things up. We frame AI like a calculator: it’s a tool to check your work after you’ve done the hard thinking, not to do the hard thinking for you. Sarah had Liam try this for his current class, and he said it only adds 20 minutes to his homework each night, which he can make up by cutting an hour of TikTok after dinner.
For parents of high school kids, I also tell them to stop checking just whether the final answer is correct. Check if all the steps are written down. If your kid skipped three steps to get to the right answer, ask them to go back and fill those steps in, even if the answer is right. It takes two minutes, and it reinforces that the process matters more than the final result.
Last week, Sarah texted me that Liam got his last quiz back. He got a C+, which is still not the A he’s used to getting, but it’s a big jump from the F on the midterm. He’s still overwhelmed, he still uses AI to save time on problems he already understands, but now he’s doing the small amount of practice he needs to keep his math skills sharp.
UC Berkeley’s CS department is still figuring out how to adjust their classes and homework policies to this new reality, and no one has a perfect fix yet. This trend isn’t just at Cal, either. I see it in almost every STEM kid I work with, from kids aiming for Ivy League to kids starting at community college. AI isn’t going away, and it’s a useful tool when used right. But the small, boring checks we do at home, the 5 minutes a day of asking to see the steps, that’s the thing that’s working right now. No one knows what will change next, but for now, that’s enough.
Please indicate: Thinking In Educating » Why Failing CS Grades Are Rising: What Parents Need To Know About AI, Math Skills, And College Success