EXECUTIVE NOTE · READING GUIDE
AI education is leaving the single-model story. Timetables, school rules, teacher work, family devices, public resources, and learning evidence now constrain one another. This report keeps publication, adoption, continued use, and outcomes separate, then gives readers internal paths to inspect the underlying topics, answers, guides, weekly analysis, and news.
Core judgment
AI becomes educational value when it can enter procurement, curriculum, teacher tasks, and an evidence chain.Features open the door. Systems sustain adoption. Evidence tests the result.01 · Scope
What this report asks
This report treats AI education as a system made of policy, schools, teachers, learners, families, content, evidence, and infrastructure. It separates an announced capability from a usable classroom practice, a pilot from sustained adoption, and participation from learning outcomes. The Chinese edition links each claim back to a topic, answer, weekly issue, guide, or news record. Start with Issue 05 for the editorial standard.
The observation boundary is 11 October 2026. The live site showed 906 collected records, 891 entries in the news list, 114 product leads, and 581 public sources. The numbers are a capture snapshot: dynamic pages can change as items are published, merged, corrected, or reclassified. The product radar export is a separate 20-product and 6-event snapshot, so it should not be added to the news total.
Throughout the report, facts identify a source, date, and actor; analysis proposes a mechanism across several facts; judgment offers a conditional decision rule. Where a source is inaccessible or only a secondary account is available, the uncertainty stays visible. This is a map of public evidence, not a census of every private contract or classroom.
02 · Executive view
The unit of competition is becoming an adoption system
The next phase of AI education will be decided by the conditions around a feature. A model can answer, generate, recognize, or coach, yet a school still needs an approved use case, a teacher who can place it inside a task, a learner who retains independent thinking, a family that understands the data boundary, and a supplier that can support mistakes and exit.
Policy, curriculum, teacher workflow, product access, and evidence now move together. The largest visible category in the live inventory is policy and governance, followed by companies and industry, research, and teacher or learning tools. This density shows where public change is being recorded; it does not establish which category has the greatest commercial value.
A practical opportunity map has three layers: compliance and implementation infrastructure with an identifiable budget owner; workflow products embedded in curriculum and teacher work; and learning-outcome claims that require stronger controlled evidence. The first two can begin with reversible pilots. The third needs a baseline, comparison, process measures, and follow-up.
03 · Policy
Policy is moving into age, lesson, and responsibility boundaries
The Chinese policy path combines national resources, local implementation, and school-level procedures. The China pathway topic follows how a broad direction becomes lesson hours, teacher development, platforms, and local adoption. Jiangsu’s full coverage of general AI literacy classes makes time on the timetable an observable commitment; Beijing’s age-banded guidance turns child protection into a procurement and classroom rule.
New York offers a different boundary. Its K–8 pause on student-facing generative AI sits beside limited high-school critical-thinking work. The policy does not answer whether AI is good or bad in the abstract. It separates a young learner’s access to an answer engine from the teaching of judgment, attribution, and responsible use.
For suppliers, policy raises explanation and service costs: data collection, retention, permissions, incident response, audit logs, family communication, and a workable exit path. Sales cycles lengthen because a pilot must satisfy more than a feature demo. This creates room for governance and implementation services alongside software.
04 · Curriculum
More classes do not automatically produce more capability
AI literacy includes understanding generation, checking uncertainty, describing a task, protecting data, and knowing when to refuse or ask for help. A curriculum that only teaches a chat interface creates visible activity while leaving transfer and judgment unmeasured. The curriculum topic and the K–12 literacy guide provide a longer reading path.
Vietnam’s national framework illustrates the implementation question: planned yearly lessons and high student exposure coexist with gaps in school guidance and safe-use knowledge. The relevant outcome is not the number of lessons announced, but whether students can explain a source, reject an unsafe output, and solve a new task without the tool.
A scalable curriculum product should provide task design, permission rules, examples, rubrics, process evidence, and teacher support. It should make the student’s own contribution visible. One-click completion may improve speed while making understanding harder to observe.
05 · Teacher workflow
Teachers remain the system’s critical interface
Teachers propose tasks, review outputs, interpret context, and carry responsibility. The teacher workflow topic records why training cannot be reduced to a product launch. A study of public universities in Guangxi linked technology pressure with professional identity and found that self-efficacy and organisational support matter.
A useful tool reduces hidden work. Generated lesson plans still need checking; automated grading needs an explanation and a correction path; a risk score needs a next action. The teacher tools guide is a practical procurement checklist: evidence, editability, handover, privacy, and the ability to return to a human process.
Evaluation should observe preparation time, edit rates, task completion, feedback adoption, and second-week use. Login counts are weaker than repeated, intentional use in a real lesson. School-level plans must budget for training, support, and migration instead of treating them as free after-sales work.
06 · Tutoring
Personalisation is about the learning process
AI tutors are moving from instant answers toward learner models, error histories, and next-step prompts. Long-term memory can support a sequence, but it can also preserve a wrong diagnosis. A learner profile needs teacher visibility, correction, retention rules, and an understandable boundary for families.
A Maryland randomised trial of a GPT-4o course assistant reported lower final grades, lower platform participation, and relatively low actual use. It does not prove that every AI tutor fails. It shows that access, task design, usage depth, and learning outcomes must be measured together. The personalisation topic collects related evidence.
Better measures include hint levels, independent answers, next-day transfer, error correction, and active help-seeking. Conversation turns and time-on-tool are useful operational signals, but they cannot stand in for learning.
07 · Market
The market is splitting by workflow and responsibility
The visible supply now spans curriculum platforms, teacher assistants, assessment tools, family hardware, voice tutors, skills programmes, and vocational training. The product ecosystem topic is a better starting point than a leaderboard because it asks who pays, who uses, who approves, and who bears the cost of failure.
Schools buy a service system: accounts, permissions, content, training, support, evidence, and an exit plan. Families buy convenience and reassurance, yet may receive less control over data and long-term cost. Public scholarship programmes widen access but still need completion and transfer data.
The strongest near-term propositions make a low-risk task measurably easier while preserving human review. Products that promise broad learning gains without a credible evaluation design face a higher bar. The outcomes and evidence topic keeps the claim proportional to the design.
08 · Assessment
The question is what remains when the tool is removed
Assessment should distinguish tool-assisted performance from independent transfer. A student may finish homework faster while losing practice in retrieval, explanation, or error correction. The assessment topic connects classroom rules with evidence and feedback.
A useful pilot records the task, allowed assistance, student process, teacher time, error events, independent checks, and a follow-up task. It can then show whether the tool supported understanding or simply substituted for it. A vendor-reported accuracy or engagement number is a starting claim, not an outcome evaluation.
This also changes product design. Systems should expose sources, show uncertainty, preserve drafts and revisions, and give teachers a way to inspect the path to an answer. These affordances support learning and make a later audit possible.
09 · Decision rules
A reversible adoption path is more valuable than a bold promise
A school can begin with a narrow task, a named owner, a short cycle, a baseline, and explicit stop conditions. It should decide in advance which data are necessary, who can access them, what an incident looks like, how a teacher takes over, and how student work leaves the system. The school adoption topic maps these decisions.
A procurement review should ask: what changed from the previous workflow; what evidence is independent; which learners are excluded; what is the full support cost; and how does the school exit? A product that answers these questions clearly may be easier to adopt than a product with more visible features.
The report can be extended as a monthly series. Each revision should preserve the capture date, evidence boundary, changed facts, and the reason a judgment was narrowed or strengthened. Readers can return to the topic index, knowledge base, and latest daily report to continue the evidence path.
10 · Limits
How to read and update this report
The site’s public inventory is broad but selective. Private school data, undisclosed contracts, classroom logs, and long-term retention remain unknown. Counts can describe observation density; they cannot on their own describe market size, quality, or learning impact.
The report uses public pages and topic answers captured on 11 October 2026. The public product export was read on that date, while most product records still carry an earlier verification date. This distinction is kept in the Chinese report’s method section so a later reader can recheck both freshness and provenance.
The internal links are part of the evidence design. News links preserve an event record, topic links explain a durable question, answer pages provide the current synthesis, guides turn findings into action, and weekly issues show the reasoning in context. Following those paths is the fastest way to test a conclusion or find a counterexample.