Teachers already knew how to guide learning.
Questioning, scaffolding, correction and productive struggle existed long before generative AI.
How and why we use AI
Good teachers have always guided learning through questions, interpretation, correction and carefully chosen next steps. What AI changes is not the value of that pedagogy. It changes its availability and scale.
The idea at a glance
Questioning, scaffolding, correction and productive struggle existed long before generative AI.
A modern language model can interpret a learner’s answer, rephrase, question and continue the conversation.
General tutoring guardrails work together with the educator’s question-specific SOTS route.
One carefully designed route can later reach another learner, and then another, without removing the educator from the design.
What changed
An educator could always write a worksheet, record a lesson, prepare hints or map out a sequence of questions. Those resources could preserve teaching, but they could not listen to the learner.
Modern large language models introduced a practical ability to reason through language and sustain a responsive conversation. The Over-the-Shoulder AI Tutor brings that conversational capability to the learning journey: it can interpret a response, notice an unexpected direction, ask the next question and adapt its wording.
What becomes scarce
For centuries, education and professional credibility often rewarded the person who could recall facts quickly and accurately. Memory still matters, because a learner needs enough knowledge to judge whether an answer is relevant, incomplete or wrong. But easy access to information changes what is scarce.
When facts, summaries and polished outputs are only a prompt away, the deeper educational question becomes different: can the learner question, judge, connect, explain, apply, debate, create, reject, improve or transfer what is available?
XVC therefore uses AI to make thinking more visible, not to hide it behind a finished answer. The learner must still show understanding through decisions, explanations, practice, correction and the ability to create something meaningful from knowledge.
Double pedagogy
The Tutor already operates under general tutoring guardrails. It is designed to conduct a learning conversation rather than behave like an ordinary answer engine.
A Pedagogical Engineer adds the educator’s intended route for a particular question, concept or misconception. The SOTS Prompt is itself a tutor design.
The teacher remains central
A SOTS Prompt does not ask a general system to invent the pedagogy from a blank page. The educator has already considered why the question matters, where learners commonly go wrong, which step should come next and what evidence would show understanding.
The Over-the-Shoulder AI Tutor then interprets that route dynamically. If the learner answers unexpectedly, the Tutor can respond conversationally while remaining shaped by both the general tutoring method and the specific teaching design.
This is human–AI complementarity in practice: AI augments the educator’s professional expertise, pedagogical intent and instructional oversight.
One learner at a time
XVC does not begin by asking how many subscribers can be placed in front of one general AI system. We begin with one learner, one question and the educational journey required to move from uncertainty towards understanding.
That work is slower at the foundation. A question-specific SOTS route may require an educator to consider the likely misconception, the order of the reasoning, the language used, what must be checked and what should not be revealed too early.
Once that route is sound, AI allows the educator’s work to reach another learner, and then another. XVC seeks scale through the careful reuse of good pedagogy, not through the removal of pedagogy.
Why training matters
Many people’s experience of AI is either a general chat box or a frustrating customer-service bot. That can create the impression that AI is either an impressive conversation partner or a badly trained support desk. XVC uses a different frame: the important question is not only whether a system uses AI, but how the system has been trained, what knowledge it may access, what tools it may use, what guardrails shape its behaviour, and when human support remains necessary.
A Large Language Model is the underlying language engine. A chat interface is one way to interact with that engine. A chatbot is a conversational application; some chatbots are simple scripted systems, while others are agentic systems that can retrieve information, use tools or take limited actions. An AI agent goes further by placing the model inside a structured loop where it can use tools, observe results and continue working toward a defined goal. None of these labels, by itself, guarantees educational quality.
For XVC, the value lies in the training and design around the AI. The Over-the-Shoulder Tutor is shaped as a tutor, not a general answer engine. SOTS Prompts add educator-designed routes for specific questions, misconceptions and learning moments. This is why Internet Learning Solutions focuses on careful training, guardrails and Pedagogical Engineer input: the AI supplies conversational scale, but the educational judgement must be designed into the system.
Parent-facing AI transparency
XVC does not want parents to be asked to accept a vague promise that “AI is being used”. A parent should be able to understand which tool is involved, what learning purpose it serves, how the learner is expected to use it, and what the AI is not meant to do.
This is especially important for PE Channels. The choice is not simply for or against AI. Parents may choose the educator, the learning style, the Channel, the resources and the level of AI-supported guidance that suits their child. The first opt-out remains simple: a learner does not have to enter or use a Channel, and a subscription can be cancelled through the normal route.
Where a SOTS route is offered, the explanation should be concrete: the Over-the-Shoulder AI Tutor supplies guided conversation, the SOTS Prompt supplies the educator’s question-specific pedagogy, and the learner must still participate, think, practise and demonstrate understanding.
Our preferred support route
Where two or all three routes form part of the learner’s service, none carries a separate individual charge and the learner may choose whichever support they prefer at any time. The sequence below explains how XVC prefers to use AI and prepared educator resources before limited human time is required.
Begin with an immediate, responsive conversation. The learner may bring a diagnostic weakness, a SOTS Prompt, a general question or an attempted solution.
Where available, use the educator’s stable visual explanation. Pause it, repeat it and return to the Tutor with questions about a particular step.
Use human support where the technology and prepared resources have not been enough. “Fallback” describes the practical sequence, not lesser access or an additional cost.
Responsible use
A confident response may be incomplete, misleading or incorrect. Learners should question, verify and use human help when the technology is not supporting them adequately.
It operates through algorithms, patterns and predictions. It does not possess morals, ethics, care or genuine understanding in the human sense.
The learner must still attempt, explain, practise, correct and demonstrate understanding. AI is used to strengthen thinking, not to create a false appearance of mastery.
Do not enter unnecessary identifying or sensitive information. Serious safety, health, counselling or emergency matters require appropriate humans and specialist services.
A living record
Artificial intelligence in education is developing quickly. XVC follows research, classroom experience and informed commentary—not to adopt every new idea, but to test our own approach against evidence, criticism and changing practice.
Inclusion does not mean full agreement with every source. Each reading note separates what the source argues from why it matters to XVC.
No reading notes match this search and topic selection.
