AI Will Change How We Teach Computer Science—Here’s How

AI coding tools are increasingly automating routine programming, which could shift computer science teaching away from memorising syntax and towards deeper problem solving. Perplexity CEO Aravind Srinivas backed this view in March 2026, suggesting CS education may move closer to its roots in mathematics, physics and systems-level reasoning.

For years, computer science education has been shaped by a practical constraint: students needed to learn enough syntax to translate ideas into working code.

But as large language models become reliable at generating boilerplate, fixing bugs and navigating codebases, the constraint shifts. In March 2026, Perplexity AI CEO Aravind Srinivas endorsed a post arguing that LLMs are increasingly automating routine coding, which could change how computer science is taught — pushing it back toward stronger foundations in mathematics, physics, and logical reasoning. (ndtv.com)

This isn’t a call to “stop teaching coding”. It’s a signal that the most valuable outcomes may move upstream: from writing syntax correctly to defining problems, modelling systems, and making good design trade-offs.

What Srinivas endorsed

NDTV reports that Srinivas shared an X post from a physics and AI/ML student and wrote “Well said.” The student’s argument was that as AI handles more repetitive programming, software engineering will lean more heavily on foundational reasoning, problem solving and systems thinking than on syntax-heavy coding. (ndtv.com)

The post framed the shift as a “centre of gravity” move: away from manual code writing and toward theoretical thinking and mathematical insight. (ndtv.com)

Why this matters now

Most education debates about generative AI focus on assessment and cheating. Those are important — but they’re not the whole story.

The bigger issue is skills relevance: if employers increasingly expect engineers to work with AI copilots, then graduates need to be strong at:

• specifying requirements clearly

• evaluating outputs critically

• understanding algorithms, complexity and constraints

• designing robust systems and interfaces

• reasoning about edge cases, security and performance

These are harder to “autocomplete” than boilerplate code.

What should change in a modern CS curriculum?

1) Teach programming, but shift emphasis to reasoning and design

Students still need to understand how code works — but the assessment focus can shift from “write a linked list from memory” to:

• choose an approach and justify it

• identify failure modes and trade-offs

• reason about time/space complexity

• write tests and explain coverage

• interpret and improve AI-generated code

2) Make ‘review’ a first-class skill

If AI can write code, students must be trained to review code — for correctness, security, maintainability, and performance.

Practical exercises:

• “AI wrote this function — find the bug.”

• “This solution works, but it’s fragile — refactor it.”

• “Add tests that would catch regressions.”

3) Redesign assessment for an AI-present world

Instead of banning tools and hoping for compliance, design assessments that assume AI exists:

• oral defences (explain your choices)

• timed reasoning tasks (solve a new problem in front of an assessor)

• versioned projects with reflections (why you changed what you changed)

• emphasis on evaluation, not production volume

4) Strengthen maths, modelling and systems content

Srinivas’ point about returning to maths/physics is really about modelling and constraints: optimisation, probability, linear algebra, discrete maths, and systems design.

That doesn’t mean every student becomes a theoretician. It means programmes should make sure graduates can reason about why a system behaves the way it does.

What this means for enterprise L&D and hiring

If you’re hiring or upskilling engineers in 2026, the “AI changes CS education” debate is also a workforce planning issue.

Look for (and build):

• strong problem framing and communication

• system design competence

• code review and testing discipline

• security-by-design habits

• comfort working with AI tools while remaining accountable

Next steps

If your organisation is adapting skills frameworks or scaling AI safely into software delivery:

• Assess readiness: identify gaps in governance, skills and operating model maturity: https://www.gend.co/ai-readiness-execution-pack

• Build a delivery model: implement controls, evaluation, and adoption support for AI-assisted work: https://www.gend.co/ai-services

• Set governance early: define policies for approved tools, data handling, and accountability: https://www.gend.co/blog/ai-governance-evolving-board-strategies

FAQ

Q1. Will AI replace the need to learn programming?
No. Students still need to understand how software works. The shift is that more value moves to problem framing, system design and evaluation of AI-generated code.

Q2. What should CS students focus on if AI writes more code?
Core reasoning skills: algorithms, complexity, testing, security, systems design, and the ability to critique and improve AI outputs.

Q3. How can universities assess students fairly if they use AI tools?
Use assessment formats that require explanation and reasoning (oral defences, timed tasks, reflections, and review-focused exercises).

Q4. What is ‘computational thinking’ in the AI era?
It’s the ability to model problems, choose abstractions, reason about constraints and failure modes, and validate solutions — regardless of who (or what) writes the first draft of code.

Q5. What’s the biggest risk of AI-assisted coding for learners?
Over-trusting outputs. Without strong fundamentals, students may accept incorrect or insecure code and struggle to debug or reason about systems.