AI in Education: Who’s holding the pen? ✏️

Small Steps Vol. 118: AI is here, but teaching is still a human craft. This month, we dig into where the tools help, where they don’t and why it still matters who holds the pen.

This is the fourth edition of Giant Leap’s deep dive series, where our investment team explores the technologies and approaches addressing global challenges in climate, health and people with commercial solutions.

In this deep dive, we examine how artificial intelligence (AI) can help tackle key challenges faced by students, teachers and education systems, while also considering the ways it may hinder progress. For teachers and students, we explore the specific problems, the potential of AI solutions and their resulting complications. For the education system as a whole, we analyse the challenge of slow innovation and the market opportunities this creates for new learning models to emerge.

Introduction

Primary and high school education lays the foundation for lifelong learning - not just through academic achievement, but by developing critical thinking, resilience, creativity, ethics, and collaboration skills that students need to thrive in an increasingly complex world.

By nurturing these traits, schools help young people grow into capable, thoughtful individuals who are ready to shape their own futures and contribute meaningfully to society. Amid teacher shortages and administrative burdens, AI is being rapidly adopted with 77% of students and 60% of teachers reportedly using these tools.

“An overall view is that they are using [AI] for everything. So what we're doing now is always assuming that they're using it.” - Dr Natalie Day, literature teacher at McKinnon Secondary College.

This swift normalisation, while offering potential relief for systemic pressures, is giving rise to new tensions and challenges that educators must now navigate.

To understand this landscape, we’ve explored the impact through the distinct experiences of its core participants: teachers, students and education systems as a whole.

Teachers

The problem: Teacher workload

A significant challenge for educators is the growing imbalance between the time available for teaching and their administrative responsibilities, including manual paperwork, curriculum differentiation and resource creation, greater administration required for student wellbeing and staff recruitment.

💡 AI as a thought partner

AI platforms have the potential to act as a "thought partner" to streamline administrative tasks and improve workflow. For instance, it can provide a starting base for a new lesson plan or an assessment task, differentiate materials for various student needs, mark assessments, provide feedback and handle routine communications. This promise is compelling, freeing educators' time to focus on the inherently human aspects of their role such as building relationships, addressing complex student needs and fostering rich classroom dialogue.

AI complications: Quality and bias

While the initial focus was on managing student use, educators now face implications for their own practice. While AI offers the promise of efficiency, its value is entirely conditional on how it is used. AI output can be generic, inaccurate and low quality. It also isn’t neutral, potentially reflecting existing biases including pedagogical or teaching styles and racial.

For instance, research has shown that AI tends to default to 'teacher-centred’ lesson plans that focus on delivering facts, rather than modern approaches that encourage students to share ideas and take an active part in their own learning.

“Just because AI can generate a lesson plan quickly doesn’t mean it’s good. Without strong pedagogical knowledge, it’s hard to judge the quality of what it produces, let alone use it effectively in the classroom. The output is only as good as the prompts you give it, and those prompts are shaped by your understanding of teaching. AI can be a helpful starting point, but it still takes an expert to shape, refine, and enact it meaningfully.” - Dr Aylie Davidson, mathematics lecturer at Deakin University

This concern also extends to grading, which has shown signs of bias to learners from diverse ethnic backgrounds. This raises the critical question of who 'grades the AI' itself for fairness and accuracy. The risk is that these tools could inadvertently push teaching methods backwards.

Furthermore, there is a risk that an over-reliance on AI tools can erode the human connection essential to great teaching. If AI reduces opportunities for valuable classroom dialogue, it de-emphasises the teacher-student relationship that is paramount to effective learning.

“AI can give you a lesson plan, but it can’t bring it to life. Without the educator’s skill—the classroom management, the relationships, the ability to read the room and respond—it’s just a dry script. It’s the teacher who brings the energy, the nuance, and the connection that makes learning actually happen.” - Dr Natalie Day, literature teacher at McKinnon Secondary College.

For this reason, we believe the most pragmatic value lies in AI tools designed for targeted teacher support that address these core challenges of quality and trust. The opportunity is in developing solutions that respect a teacher's professional responsibility. A teacher needs to be able to trust the tool's output as a solid starting point, while always having ultimate control to evaluate and adapt the material for their own classroom.

Examples in this space range from platforms like Edexia and Mark My Words, which focus on streamlining feedback and assessment, and tools such as Teacher's Buddy and EasedUp aimed at alleviating content creation burdens, to more comprehensive or integrated platforms like Magic School, School AI and Kira that offer broader workflow support.

Students

The problem: The Two Sigma challenge

The conventional 'one-to-many' classroom structure, by its very design, imposes fundamental constraints on meeting diverse student learning needs. When a single instructor addresses numerous students simultaneously, the capacity to tailor pacing, provide individualised feedback and ensure engagement for each learner is inherently restricted.

Benjamin Bloom's paper in 1984 revealed that one-to-one tutoring, when combined with regular assessments and feedback, could elevate average student performance by two standard deviations - referred to as two sigmas, hence the term ‘Two Sigma Problem’. This result effectively placed these students above 98% of peers in standard classroom instruction. To no surprise, one-to-one tutoring is too costly and inefficient to implement.

💡 AI tutors for personalised learning

AI tutors have the potential to address the Two Sigma Problem. Research indicates that AI tutors can help students learn more material, with students self-reporting significant engagement and motivation to learn, where the AI tutor was tailored to behave like a seasoned instructor for the specific course.

AI complications: Over-reliance and intellectual passivity

There are concerns about students’ increasing over-reliance on AI platforms. The ease of task completion can decrease learning skills, undermine critical thinking skills and long-term learning outcomes, potentially creating 'apparent experts' where students who use AI to produce answers but haven't engaged in the productive struggle necessary for deep learning.

A study conducted by the University College of London showed that humans are hard-wired to take the path of least resistance. The study explored how people respond when a familiar action becomes slightly more difficult. Participants were asked to complete a repetitive task involving turning a handle. Over time, the direction that had been easier to turn was made slightly harder. Most people subconsciously adjusted their actions to continue taking the easier route, even when it was no longer the correct one. Only a small group recognised the increasing difficulty and made a deliberate decision to change their behaviour, even though it required more effort.

This finding highlights a deeper issue: awareness and deliberate effort are key to meaningful learning. If AI systems make tasks too frictionless, students may be nudged away from developing persistence and critical thinking, qualities essential for long-term learning success. In trying to optimise learning pathways, AI could inadvertently encourage intellectual passivity unless designed with intention.

We believe the real value lies in AI platforms built on sound pedagogical principles, designed to foster critical thinking and problem-solving, and capable of providing teachers with actionable insights into student understanding, not just their performance. We are particularly excited by companies building in areas such as Sigiq, Synthesis Tutor, Khanmigo and Squirrel AI.

Education Systems

The problem: Slow pace of innovation

A core challenge for our education system is its slow pace of innovation, a problem the Gonski 2.0 report attributes to Australia being stuck in a 20th century industrial model of schooling. This inertia is reflected in structures like the Australian curriculum's official six-year review cycle, creating a system that is not designed to incentivise schools to innovate. The result is a focus on static content, often geared towards standardised testing, which can under-prepare students with the resilient and adaptive skills required for the future.

Nowhere is this tension more visible than in the response to AI. While these technologies are evolving rapidly, schools are operating within a system that is not designed to keep pace. Yet many educators recognise that ignoring AI is not an option.

"We are proactively engaging with students, staff and families to develop the critical literacies needed to support responsible, ethical, and age-appropriate use of AI. Alongside this, we’re transforming our digital ecosystem and equipping staff with the skills to harness AI for greater productivity, and in the process generating meaningful insights that directly support student learning." - Kate Manners, Director of Strategic Initiatives at Camberwell Girls Grammar School

The opportunity: New learning ecosystems

This creates opportunities for founders to build solutions that operate both alongside and beyond the traditional system. One pathway lies in complementary programs, which work alongside formal schooling to provide critical skills development that is challenging to deliver or assess within the confines of a standardised curriculum.

While schools build foundational knowledge, these programs build applied capability: teaching students how to solve complex problems, work in teams, communicate with impact, and develop an entrepreneurial mindset. Our portfolio company HEX is a prime example, delivering future-focused skills for high school students and leavers, and university students by bridging the gap between classroom theory and real-world innovation.

A bolder path involves building new learning ecosystems entirely. It’s particularly interesting to see the growing demand for microschools in the US, which are small, flexible learning communities that operate as an alternative to traditional schooling. These models often prioritise personalised learning, giving students more agency over how they learn, with ventures like Primer and Kaipod leading the way. While this movement is currently more established in the US, it points to a global appetite for more personalised and flexible models and shows the ambition of founders wanting to redefine learning itself.

What does this mean for market opportunity?

Historically, the path for an Edtech startup to achieve venture-scale returns has been challenging. As we've noted in our pitch deck guide for edtech founders, hurdles like long sales cycles, a fragmented system and an Australian market that can be deceptively small for a niche solution have often made this a difficult landscape to navigate.

AI does not erase these go-to-market challenges. The user-vs-buyer dynamic persists, which means the people who use the software every day or users, namely teachers and students, are often not the ones who approve the purchase. That decision falls to the buyers, typically school leadership who hold the budget.

Instead, one of the major impacts of AI is its ability to lower the cost and complexity of building a product that could realise personalised learning for every student, and enhanced capacity for every teacher.

Eric Schmidt’s call to adopt AI faster to solve foundational problems like education, frames the immense and positive opportunity before us. Realising this opportunity requires a clear focus not on what technology can replace, but on what it can amplify: the impact and effectiveness of human teachers. AI remains a tool in service of the deeply human endeavour of teaching. Investing in this space means backing solutions that embrace this vision, aiming to build a future where technology empowers us to create an education system truly equipped to develop the creative and resilient minds our future demands.

If you’re a founder building in this space, or know someone who is, we’d love to hear from you. Get in touch with us here.

Hungry for more?

🤯 Lower brain engagement when ChatGPT writes your homework. A new study out of MIT Media Lab suggests students who lean on ChatGPT for school-style writing tasks show reduced neural engagement and weaker retention compared with peers who wrestled with the work themselves. The research is preliminary and not yet peer-reviewed, but it raises timely concerns about the cognitive effects of generative AI.

💡 Digital twins reveal why some brain struggle with maths. Stanford HAI researchers have built “digital twins” of students’ brains, pinpointing the neurological signatures of maths learning disabilities. They found too much neural activity, rather than too little, resulted in learning difficulties. This work opens a path to personalised learning plans tailored to each learner’s neural wiring.

🤖 Automation doesn’t just displace, it reshapes expertise. According to MIT economist David Autor, focusing on “automation exposure” alone misses the point. Analysing U.S. job‑task data from 1977‑2018, David found 64 % of routine tasks vanished while 76 % of new tasks were abstract, meaning some roles became more specialised and better paid (proofreaders), while others became less expert but more plentiful (taxi drivers).

Image source: Stanford Institute for Human-Centered AI, Assessing the Real Impact of Automation on Jobs

🔬 Reclaiming teacher’s superpower. Byron Scaf from Stile Education argues that AI should liberate teachers to focus on the messy, human parts of learning – mentoring, emotional coaching, real‑world projects – not replace them.

📚 Publishers are tapping into AI. Major publishers have announced partnerships with AI platforms, including Perplexity & Wiley returning answers anchored to citations for easy source‑checking; and Pearson & Alphabet creating personalised learning tools across K-12. One deal boosts trust and discovery; the other personalises practice. Together, they signal a future where AI elevates both the quality of information and the experience of learning it.

🧑‍🏫 Listen to a16z’s episode on the “State of AI & Education” to get their perspective on where they see AI in education today.

New paths

🔍 Trace is looking for a Sustainability and Customer Success Manager (Sydney).

🧘 Mindset Health is hiring a Senior Software Engineer - Mobile, AI (Melbourne, hybrid), Senior Software Engineer - DevOps) and Healthcare Account Executive (New York, hybrid).

⚡Amber is hiring a Software Engineer (Melbourne, hybrid), Senior Software Engineer (Melbourne, hybrid), Senior Full Stack Engineer (London, hybrid), Customer Service Specialist (Melbourne, hybrid), Business Development Manager (Melbourne, hybrid) and Technical Sales Expert (Melbourne, hybrid).

🚜 Artesian is hiring an Investment Associate in AgriFoodTech (Sydney).

🧻 Who Gives A Crap is looking for a Logistics Manager for a 12-month parental leave cover (London).

Save the date

📅 3 July: Investing in the Inevitable – Where are the Big Bets in Climate Tech? (Sydney). Interested in learning about investing in climate tech? Greenhouse is hosting a session with experts, founders and operators in the space. Get tickets here.

📅 10 July: The Secrets to More Sustainable Buildings (Melbourne). Secrets? Yes please! EnergyLab is hosting a panel to reveal what’s need for zero-carbon buildings to become a reality. Register here.

📅 10-11 July: AI Hackathon By Antler and MLAI (Melbourne). A 24-hour retreat-style Hackathon for 10 AI builders, engineers or researchers. Register here.

📅 24 July: LuminaX Accelerator Demo Night 2025 (Gold Coast). Tickets here to see what startups are building in the HealthTech space.

📅 6 August: Startup&Angels’ Impact Investors Pitch Night (Sydney). Ben will part of the panel - come say hi! 👋 Tickets available here.