Four years ago, a handful of students in a typical university programming class were using AI assistants. Today, it's nearly all of them. That shift is forcing computer science educators to answer a question they can no longer dodge: what's the point of teaching someone to write code that a machine writes faster?
Ashley Juavinett, a neuroscience professor at UC San Diego, makes a compelling case in a new article for The Transmitter that the answer isn't "ban AI" or "ignore it" - it's to rebuild programming courses around the skills AI can't replace.
Debugging and Design Over Syntax Drills
Juavinett's argument boils down to this: if students will spend their careers collaborating with AI coding tools, then teaching them to memorize syntax is like teaching long division to someone who'll always have a calculator. The better use of classroom time is teaching students to design workflows, evaluate whether code actually does what it should, and debug outputs from tools they didn't write themselves.
In practice, that means swapping traditional "write this function from scratch" assignments for exams where students predict what code will output, trace through unfamiliar snippets, and explain the reasoning behind their project decisions. It's harder to test, but it targets the skills that actually matter when your daily coding partner is Claude or Cursor.
The Equity Problem Nobody Talks About
The most striking part of the piece isn't the curriculum advice - it's the data on who's actually using these tools. Research shows women use AI coding assistants less than men in educational settings, and a HEPI report found that students from vulnerable backgrounds are more hesitant to adopt AI tools, largely because they're worried about crossing academic integrity lines.
This creates a paradox: the students who could benefit most from AI assistance are the ones least likely to use it, because institutional policies are either unclear or overly restrictive. Juavinett's solution is practical - survey your class, establish shared norms about what's acceptable, and set different AI-use policies for different assignment types rather than blanket bans.
What This Means for the Tools Market
For anyone building or selling AI coding tools, this shift in education is a leading indicator. A generation of developers trained to collaborate with AI rather than compete against it will have very different expectations for their tools. They'll care less about raw code generation and more about explainability, debugging assistance, and workflow integration.
The professors figuring this out now are essentially designing the user expectations of 2030. Tools like Claude Code and Cursor that already emphasize understanding and iteration over pure generation are better positioned for that future than simple autocomplete engines.