AI is breaking education before fixing it
Berkeley's CS 10 saw 35.3% of students receive F's in spring 2026, and CS 61A saw 10.6% fail. The thread connects this directly to AI usage eroding foundational math and coding skills. Students are using AI to complete assignments without actually learning the material, and when the AI gets stripped away for exams, the gaps show up hard.
This is the same pattern showing up in corporate settings. The Uber AI spend cap thread notes that heavy AI tool usage is producing output that looks like work but may not be building durable capability. The signal across both threads: AI is very good at completing tasks and very bad at building the underlying understanding that makes someone useful when the task gets hard.
The counterpoint in the Berkeley thread is worth noting: some commenters argue the failure rate spike is partly explained by more students who shouldn't be in these courses enrolling in them. That's plausible, but it doesn't explain away the underlying concern about skill atrophy.
So what?
If you are hiring junior engineers, the credential is becoming less reliable as a signal. A CS degree from 2026 onward may reflect less actual competence than one from 2022. You need to test fundamentals in interviews more aggressively, not just vibe-check portfolio projects that could have been AI-generated.