Personalized learning is relational

Personalization has a production problem

A teacher can hold one learning destination in mind and still need several credible ways for learners to reach it. One student may need shorter task sequences and visible checkpoints. Another may need language support. Another is ready for a more complex application. Each path can require its own instructions, materials, presentation, formative checks, and assessment supports.

The need is not hypothetical. Statistics Canada reported that the disability rate among Canadians aged 15 to 24 rose from 13.1% in 2017 to 20.1% in 2022; mental-health-related and learning disabilities were the most prevalent types in that age group. That survey covers youth beyond secondary school and should not be read as a classroom count, but it does show a changing context around learning and participation. See the 2022 Canadian Survey on Disability release.

Across participating OECD education systems, TALIS 2024 found academically diverse classrooms were the norm, while modifying lessons for students with special education needs and excessive administrative work were commonly reported sources of stress. The report describes associations and teacher perceptions across different systems; it is not a direct estimate of Canadian curriculum-production time. Read The demands of teaching.

The relationship is part of the learning process

Teaching is not content delivery with a person standing nearby. Educators interpret hesitation, build trust, connect a task to a learner's history, decide when to challenge and when to scaffold, and make meaning with a group over time. Those judgments are social as well as instructional.

A 2023 preregistered review synthesized 24 prior meta-analyses representing more than two million pre-K–12 learners. It found meaningful associations between teacher–student relationships and clusters that included achievement, motivation, engagement, behaviour, executive function, belonging, and well-being. The underlying evidence varies in quality and is largely associational, so it does not prove that relationship quality alone causes each outcome. It does make the relationship too important to treat as decorative. Review Teacher-Student Relationships and Student Outcomes.

Online-course completion is a caution, not a verdict

The familiar claim that almost nobody completes a massive open online course is too blunt for serious use. Completion changes depending on whether the denominator is every registration, learners who became active, or learners who intended to finish. A 2024 comparative study shows how much those definitions matter. See Uncovering MOOC Completion.

MOOCs are not generative AI, and voluntary adult enrolment is not a secondary classroom. They cannot prove that direct-to-student AI will fail. A 2025 preprint reviewing computer-science MOOCs in K–12 settings is more relevant: it concluded that classroom teachers still played an important role in supporting and managing learners, while noting that the evidence base was small and varied. Read the K–12 computer-science MOOC review.

The responsible conclusion is narrower: content access does not remove the need for guidance, context, accountability, and relationship. Technology should be designed around that reality.

Inclusive design multiplies valuable work

CAST's Universal Design for Learning Guidelines 3.0 call for multiple ways to engage with learning, build knowledge, act, and express understanding. They also emphasize graduated support, meaningful goals, planning for challenges, and learner agency. That is a richer ambition than changing the reading level of one worksheet.

A 2023 systematic review and meta-analysis of 49 primary studies found positive aggregate effects for differentiated instruction, while also reporting variation by context and implementation. It supports differentiation as a serious instructional approach, not a claim that every differentiated activity works or that production cost disappears. See Does Differentiated Instruction Affect Learning Outcome?

This is the tension: the design work is educationally valuable, but it lands on finite educator time. If software creates three disconnected worksheets and leaves the teacher to reconcile the goals, sequence, assessment, and accessibility, it has produced more material and more management.

The better role for AI is teacher-led

There is early evidence that generative AI can reduce some preparation work. A 2024 randomized Teacher Choices trial studied ChatGPT used with a guide for Key Stage 3 science lesson and resource preparation. Participating teachers spent less time on the measured task, and a blinded expert panel did not detect lower resource quality. The trial was narrow, short, and focused on one preparation task; it did not establish long-term learning effects or prove that any AI product improves differentiation. Read ChatGPT in Lesson Preparation.

That is enough to support a direction, not a sweeping promise. AI is useful when the educator defines the destination and the system helps carry those decisions through the production work: the outline, unit sequence, lesson, presentation, assignment, rubric, guided path, independent path, and extension.

The distinction is between generating products for learners and strengthening the process through which an educator designs learning. Egora is being built for the second.

What that position requires from the product

Local-first file ownership is a privacy and agency feature, not a claim of offline AI. Generation requests still go to the remote model selected by the educator. That boundary should be visible before anyone uses the product.

The learner and educator outcomes belong together

A worthwhile education product should not force a choice between learner opportunity and educator sustainability. More responsive pathways can help learners access common goals. Less repetitive production work can give teachers more room for feedback, observation, conversation, and professional judgment.

That is the proposition behind Egora: create the scaffolding learners need to succeed, while reducing the workload required to make that scaffolding coherent. The AI helps with production. The educator leads the learning.