Vibe Coding Is Fine — If You Know Who You're Teaching

Derek Long from the University of Southern Queensland argues that GenAI programming education shouldn't be one-size-fits-all. For students who code as a secondary skill, like his agricultural cohort, embracing GenAI-assisted "vibe coding" from day one can break down barriers, lift pass rates, and build lasting learning autonomy. The catch: your assessments have to evolve with it.

"Vibe coding" has become a loaded term in programming education. For many computer science educators, it conjures the nightmare scenario: students prompting their way to working code they can't read, can't debug, and don't understand. And for a CS cohort, students whose professional identity will be built on deep programming capability, that concern is entirely valid.

But here's the question I want to put to you: what if your students aren't computer scientists? What if they're students who will write a script maybe a handful of times a year, to wrangle sensor data or automate a tedious analysis task? Should we teach them programming the same way?

My answer, after four years of teaching software skills to AgTech, agricultural engineering and mechatronic engineering students, is no. The right GenAI approach for programming education depends on the degree and the major. And for cohorts where coding is a secondary skill, leaning into GenAI-assisted coding, done deliberately, with scaffolding, isn't a compromise. I'd put it to you that it's the better pedagogy.

The cohort changes everything

I teach the software component of UniSQ's Agricultural Engineering program and the Bachelor of Agricultural Technology & Management, including AGR2202 Instrumentation and Automation in Agriculture — a first-year unit covering programming, electronics, and sensing for farm applications. My students come from agricultural engineering, and also plant science, animal science, and ag-tech degrees. Most arrive with little or no coding background, and many find the prospect of programming intimidating at best. In earlier offerings, finishing the course was met with relief rather than joy.

For these students, programming is a service skill. The goal isn't to produce software developers; it's to produce digitally literate agricultural professionals who can automate a data workflow, interrogate a sensor network, and — crucially — upskill themselves when the need arises years later. That last point matters: I can't teach them everything they'll ever need, so learning autonomy is the real learning outcome.

That framing changes the relationship with GenAI substantially. The "productive struggle" of debugging syntax errors at 11pm builds something valuable in a CS student. In an ag student, it mostly builds resentment and a lifelong belief that coding isn't for them.

Scaffolding the vibe: a SAMR approach

From 2023, I integrated GenAI into AGR2202 not as a tolerated shortcut but as a core tool, scaffolded using the SAMR framework so students unlock capability progressively:

Substitution: Students use GenAI as a tutor and research assistant. An early exercise has them answer the same Python questions three ways — GenAI chat, AI-summarised search, and traditional search — then reflect on which method suits which type of question. This builds an early, calibrated sense of what GenAI can and can't be trusted with.

Augmentation: Students use GenAI to help with small, contained subtasks — generating a subroutine to read a sensor pin, or decode a data packet — then inspect the output for unfamiliar concepts and learn from it retrospectively.

Modification: GenAI reshapes the workflow itself. In one task, students write a pseudocode plan for a weather station data script, get GenAI feedback on the plan, hand-code the solution, then get GenAI feedback on their code. The chat logs and a short reflection form part of the submission — making feedback literacy an assessed skill.

Notably, the first weeks of programming workshops are now GenAI-free, to build fundamentals before the tools are introduced. Vibe coding "done right" still starts with a foundation.

What happened: the good, and the warning

The results across three offerings tell a clear story — including a cautionary chapter.

In the first GenAI-enabled year (2023), I deliberately left assessments unchanged as a baseline. Student satisfaction jumped from 4.0 to 4.67 out of 5, and survey respondents uniformly agreed the GenAI integration benefited their learning. But the median grade jumped from 709 to 873 out of 1000. That wasn't a sign students had become brilliant programmers overnight — it was a sign my pre-GenAI assessments could no longer differentiate capability. The tools had absorbed the difficulty.

So in 2024 I redesigned the assessments: more complex, more authentic agricultural problems, and a shifted emphasis toward the things GenAI can't do for a student — justifying design choices, interpreting sensor data in context, validating and critiquing AI output. Grades returned to a healthy distribution (median 699), while satisfaction rose even higher. Harder assessments, happier students. The complexity was now meaningful rather than artificial.

The headline number, though, is completion. Over 2023–2025, AGR2202 averaged an 80% completion rate. Our university's general introductory programming course, still GenAI-prohibited, and equally a first touchpoint for novices, sat around 55% over the same period. Students who took both told us the GenAI-enabled, discipline grounded version was simply less intimidating and more engaging.

And the assessment tasks students now complete would have been unthinkable for this cohort pre-GenAI: parsing tractor machine logs, building automated pipelines for field sensor data, modifying multispectral image processing code. These are genuinely useful, industry-relevant tasks — not toy exercises. Several students have since chosen software-heavy honours projects, including an AI beehive monitoring system. That's what learning autonomy looks like.

How you might apply this

If you're weighing up GenAI for your own programming unit, start with the cohort question:

1. Ask what programming is for in this degree. Core professional identity, or applied service skill? Be honest — this determines how much productive struggle is pedagogically justified.

2. Scaffold the tools; don't just permit them. SAMR gave my students a staged pathway from GenAI-as-tutor to GenAI-as-collaborator. Open slather without scaffolding invites over-reliance; some students will offload rather than engage.

3. Teach source selection explicitly. My biggest early surprise was students trusting GenAI over library documentation and datasheets. Calibrated trust has to be taught.

4. Assume your assessments will break and plan the redesign. If grades spike, that's diagnostic information, not success. Shift assessment weight toward interpretation, justification, and validation.

Closing thought

The debate about vibe coding in education often assumes a single answer must apply everywhere. It doesn't. For computer science and software engineering students, guardrails around GenAI-generated code make sense. For my agriculture students, GenAI is the difference between programming as a locked door and programming as a tool they'll actually reach for in their careers. Same technology, different cohorts, different pedagogy, and that's exactly as it should be.

Dr Derek Long

University of Southern Queensland

6/8/26