Teaching Software Engineering in the Age of AI Coding Agents

Sun, 15 Feb 2026 00:00:00 +0000

Last weekend I built, shipped, and pushed a reasonably complex piece of software to production.

Several thousand lines of code. Live. Working.

Time spent: a weekend.

This would have been unthinkable not long ago.

With the rise of genuinely capable coding agents, Replit, Qwen Coder, Claude Code, OpenAI Codex, and their tight integration into tools like Cursor and VS Code, the act of writing code has become faster, cheaper, and increasingly commoditised.

And that has important implications for us as academics.

If software development itself is becoming easier, then our value in teaching cannot sit primarily at the level of syntax, APIs, or boilerplate implementation anymore. What matters far more is:

Teaching how software actually works at a conceptual level
How to design sound software architectures
How to think about security, safety, and robustness by design
How to build accessible, usable, and well-designed interfaces
How to reason about trade-offs, constraints, and failure modes
How to deploy responsibly in real-world environments

Even areas like cloud deployment and DevOps, once specialist skills, are now remarkably straightforward with the right tooling. The barrier has shifted.

This semester, I will be teaching a software engineering module, and I am actively thinking about how to integrate these principles into the curriculum, not by fighting AI-assisted development, but by designing learning around it.

Our responsibility, I think, is to prepare students not just to code, but to become AI-augmented software engineers who can design systems that are safe, secure, ethical, accessible, and genuinely useful to people.

The tools have changed.

The fundamentals, good thinking, good design, and good judgement, matter more than ever.

I am curious how others are adapting their teaching in this new landscape.

Future of Work | Dr Riasat Islam

Teaching Software Engineering in the Age of AI Coding Agents