The year AI became more useful than a tutor for HSC feedback

I’ll be direct about something most EdTech founders won’t say: for most of 2024 and into early 2025, I didn’t trust AI to do what I was building.

I was using AI tools for content development in my work and they weren’t good enough. Not for anything that required nuance, consistency, or the kind of pedagogical precision that HSC marking demands. The feedback was generic. The hallucinations were real. The responses read like something that had absorbed a lot of text without understanding any of it.

So when people ask me why I built Assessment as Learning in 2026 and not earlier, the honest answer is: because earlier, I couldn’t have built it well. And I wasn’t willing to build it badly.

The threshold

Something changed. Not overnight, but crossing a line that I could feel in the work.

The moment that clinched it for me came from my corporate life, not my education background. In enterprise SaaS, I’d spent years building curated knowledge bases: structured, verified, domain-specific information that AI could draw on with genuine reliability. What I discovered was that when you eliminate the information problem, when the AI is working from a curated, authoritative source rather than the entire internet, hallucination drops dramatically and dependability rises to a point where you can actually trust the output.

That insight transferred directly to HSC feedback. NESA publishes marking guidelines. Past papers exist. Examiner comments are on the record. The knowledge base for HSC feedback isn’t just buildable, it’s already been built by NESA over decades. The question was whether AI had become conversational and nuanced enough to use it well.

By late 2025, the answer was yes, and improving faster than any human system could match.

Why tutors and teachers can’t maximise skills uplift

I want to be careful here, because I have genuine respect for what good teachers and tutors do. They build confidence. They explain difficult concepts in ways that land for a specific student. They provide accountability and human motivation that no platform can replicate. Your child has a teacher – value them. If your child has a great tutor, keep them.

But here is what teachers and tutors structurally cannot do: provide twenty targeted, objective, criterion-referenced feedback cycles on exam responses before October 13th. Not because they’re not skilled enough. Because there aren’t enough hours, and human feedback at that volume isn’t economically viable for most families. Skills improvement used to be capped by the available human resources. The landscape is different now.

This matters because exam technique, the specific skill the HSC rewards, isn’t maximised through explanation. It improves through repetition with correction. You have to write the response, get feedback on exactly where it diverged from what the marker wanted, adjust your mental model, and write again. That loop needs to run dozens of times to become instinct under exam conditions.

A tutor can run that loop three or four times across a term. AAL can run it every day.

The learning science behind why this works now

My Masters in Education at Macquarie University was in Adult Learning and Technology, and one framework from that work has stayed with me ever since is Vygotsky’s zone of proximal development.

The idea is not new, it’s simple and profound. Learning happens most efficiently in the space between what a student can do independently and what they can do with support. Too easy and there’s no growth. Too hard and there’s no foothold. The zone of proximal development is the precise gap where growth is possible, and the key to activating it is timely, specific feedback.

The word timely is doing a lot of work in that sentence.

When a student writes an HSC response and gets feedback three to seven days later from a teacher or tutor, the cognitive moment has passed. They’ve moved on. The feedback lands in a different mental context to the one in which the work was produced. It’s still useful, but it’s a fraction as powerful as feedback received while the thinking is still fresh.

What AAL does is land feedback in that fresh cognitive moment. A student writes a response. Within seconds, they receive examiner-calibrated feedback on structure, argument, use of evidence, question deconstruction and criterion alignment. Then, and this is the part we’re most proud of, they enter a guided chat where they can interrogate that feedback, push back on it, ask what a Band 4, 5 or 6 response would have done differently, and build understanding through genuine dialogue.

Three key distinctions about AAL versus free AI.

• A controlled knowledge base. These AI tools cost us money but give us the ability to ensure correct knowledge referencing. ChatGPT goes where it wants and you can’t control depth of response or its creativity (making things up). We can and we do. We chose a specific AI model for this purpose – ruthless precision.
• A controlled conversational interactive layer. We selected a different AI model used for our chat interface. This is what amplifies useful AI feedback from generic AI scoring. The chat is where the richest learning can happen. This one is selected for conversational language nuance, with a curated knowledge base. It is formative feedback in its most powerful form, not a grade, not a comment, but a thinking partner available at any time (even midnight the night before a trial exam).
• Tracking progress over time. By buildout out a progress dashboard per subject, and tracking over time, the student (and tutor, teacher and parent), can diagnose elements of skill requiring more practice. Short answers skills weaker than extended responses. Module B outperforming Module C. Modern History not as strong as Advanced English.

What this means for your Year 12

If your child is sitting the HSC in October 2026, they have roughly twelve weeks of serious preparation left. In that window, the students who improve their exam technique fastest will be the ones who get the most targeted repetition with the most immediate correction.

That used to mean paying for more tutor hours. Now it means using the right tool for the right job, and understanding that AI, in its current form, has crossed the threshold where it can do something tutors were never designed to do efficiently. We want to complement the work of teachers and tutors by providing a service to maximise uplift.

Assessment as Learning was built at this specific moment because this specific moment made it possible to build it well. Not as a replacement for the human support good tutors provide. As the missing piece that makes that support count.

You can try it free, three practice attempts, no credit card, no commitment, at assessmentaslearning.com.au.

---