Skip to content

Questions and answers

Perspectives on AI, EdTech, and the Future of Higher Education

Veton Krasniqi on student learning, faculty control, academic integrity, responsible AI governance, assessment, and the economics of education technology.

Who is Veton Krasniqi?

Veton Krasniqi is an AI and EdTech product leader whose work centers on implementing business strategy to impact education. Since 2016 he has worked in product, delivery, and quality roles across enterprise software, public-sector IT, and education technology, and he is currently AI Product Owner at Apporto. He speaks on AI and assessment, including at TxDLA 2026 and the University of Central Florida's Teaching and Learning with AI conference, and is pursuing an MSc in AI for Business Transformation at the London School of Innovation.

What does an AI Product Owner do in higher education?

An AI Product Owner connects educational needs, technical capabilities, and business objectives to develop products that institutions and their users can actually benefit from.

In EdTech, that requires understanding more than what an AI model is capable of doing. You need to understand how faculty teach, how students learn, how institutions operate, and where automation can introduce new risks.

My approach involves product strategy, requirements, model behavior, governance, usability, testing, and working directly with engineering and other stakeholders. The objective is to turn the capabilities of AI into reliable, scalable products that solve real problems, rather than building features simply because the technology makes them possible.

Are students still learning when they use AI?

It depends on what the AI replaces. When AI explains, prompts, and checks, it can support learning. When it does the thinking, students can turn in better work and keep less.

A quasi-experimental study of a math learning platform, still a preprint, found that over eleven quarters after ChatGPT's release, college students spent 26.9% less time on text problems that can be pasted into a chatbot. On proctored retention items, their odds of a correct answer fell 25% (Rismanchian et al., preprint). Those are population-level trends, not findings about any individual student.

The question I keep asking is when support becomes substitution, and I organized a January 2026 webinar around it: “AI is Here. Are Students Still Learning?”

If the AI disappears tomorrow, what can the student still do?

Read the related perspective on experience, reflection, and learning

How can faculty set practical guardrails for AI in their courses?

Set them at the assignment, not the institution. A chemistry problem set and a reflective essay need different AI behavior, so faculty should decide, assignment by assignment, what AI may do: explain a concept, give a hint, check work, or stay out entirely.

Put that permission where students can see it, ask for reasoning before solutions, and build on recognized external guidance such as EDUCAUSE's instead of writing rules from scratch.

Practical guardrails were the subject of my session at the University of Central Florida's Teaching and Learning with AI conference, “From AI Uncertainty to Practical Guardrails for Teaching and Learning.”

Read about environment-level exam governance

Can AI detection tools reliably identify AI-written student work?

No, not reliably enough to support an accusation on its own. OpenAI withdrew its own AI-text classifier in 2023 “due to its low rate of accuracy”; in OpenAI's own testing, that classifier had labeled human-written text as AI-written 9% of the time (The Decoder). A 2023 study also found that the GPT detectors it tested consistently misclassified writing by non-native English speakers as AI-generated (Liang et al.).

A probability score is not evidence a faculty member can stand behind. The better approach is process visibility: show how the work developed, then let the instructor follow up as an educator rather than a prosecutor.

This was the subject of my TxDLA 2026 session, “The AI Detection Crisis in Higher Education.”

Read the related analysis of AI detectors

How can educators see how learning happened, not just the final answer?

Look at the process as well as the submission. Drafts, revision history, time on task, and the reasoning a student shows along the way tell an instructor how the work developed. An essay revised across several sessions reads differently from one pasted in all at once.

That evidence is context for a conversation, not a verdict, and the decision stays with faculty.

I discussed moving past detection toward this kind of visibility in a June 2026 webinar.

Read about secure assessment and independent capability

What does responsible AI governance look like in higher education?

Responsible AI governance isn't simply having a policy that says AI is permitted or prohibited.

It requires clear expectations, faculty control, student privacy protections, appropriate technical guardrails, and a way to evaluate whether the technology is producing the educational outcomes it was introduced to support.

I approach governance at multiple levels. Institutions establish the broader policies and requirements. Faculty determine what assistance is appropriate within their courses and assignments. Product teams design systems that support those decisions rather than making them irrelevant.

Governance also needs to account for model limitations, inaccurate responses, bias, data handling, and accountability.

My view is that AI should be governed according to the decisions it influences and the risks it introduces, not simply because it carries an AI label.

Read the future of academic integrity in higher education

How should education leaders evaluate the value and economics of AI products?

I think too many AI discussions begin and end with what the model can do, without adequately examining whether the capability creates enough value to justify its cost.

An AI feature might reduce administrative work, improve the availability of student support, or help faculty provide more substantive feedback. But those benefits need to be evaluated against implementation costs, model inference costs, integrations, support, governance, and the actual behavior of users.

A successful demonstration doesn't automatically make a viable product.

I look at whether the feature addresses a meaningful problem, whether users adopt it, what it costs to operate at scale, and whether its outcomes justify the investment.

In education, there is another dimension we can't overlook: a feature can improve efficiency while undermining learning. That is not a trade-off institutions should accept without examining it carefully.

Read the two numbers every assessment leader should watch

What should assessment leaders watch as AI changes testing?

Watch the gap between what students produce with AI available and what they can do without it. If unproctored scores rise while supervised performance falls, the grade has stopped measuring the student.

Start with an audit. Sort assignments into AI-susceptible formats, such as text problems a student can paste into a chatbot, and AI-resistant ones that need interpretation or hands-on manipulation. Then see how much of the grade book sits in the first group.

I made this case in “The Two Numbers Every Assessment Leader Should Be Watching” in August 2026.

Read the two numbers every assessment leader should watch