Picture this: you’re stuck on a problem — no teacher nearby, no classmate to ask. You open your textbook, and the textbook answers. Not with a canned hint — with a response aimed precisely at the spot where your reasoning went off the rails.
“A textbook that knows exactly what you didn’t understand” — that’s how Sal Khan described Khanmigo at TED in 2023.1
This is no longer a metaphor. In Khan Academy’s pilot programs, student engagement grew 2.3× compared to the control group.2 Not because the content changed — the mechanics changed: for the first time, the textbook started reading back.
By 2030 this dialogue mode will be a school standard across the CIS, not an experiment. The question isn’t whether it happens — it’s how fast the infrastructure catches up with what the models can already do.
Where we are now: school in 2025
But is the CIS school system ready for this — not individual enthusiasts, but the state and the market? By 2025 the answer became “yes” along three axes at once.
Regulators opened a policy window, the device fleet reached 1:1 coverage, frontier models matured in quality and cost. Three independent processes intersected right now — and only at that intersection is a mass-market school product possible.
The regulatory window. In 2024 Russia’s Ministry of Education approved methodological guidelines for AI in schools3 — the first official document opening the way to government procurement. Kazakhstan went further: in March 2025 the MDDIAI launched the national “AI in School” program with a 14 billion tenge budget4, turning adoption into a national-project priority. The two largest CIS markets legalized the tool simultaneously — that’s not a coincidence, that’s a signal.
Market and pilots. Russian edtech grew from 53 to 89 billion rubles in three years5: private capital has accumulated and is looking for a point of application. In parallel, 42% of Russian schools had already tested AI assistants by 20256 — not lone experiments, but a structural shift in how principals and methodologists across the country see the tool.
Devices caught up too: 38% of CIS schools reached 1:1 coverage by 2025 (UNESCO). Frontier models generate adaptive exercises in <1s at ~$0.003 per request — the economics finally work. Three processes ran independently. In 2025 they aligned for the first time.
The forecast: by 2030 the AI textbook is a mass-market product
Three conditions have converged. What’s left is to place a concrete bet.
That’s the lower bound, not an optimistic scenario. Here’s what backs it.
The market pushes from below. Global EdTech AI will reach $8.3B by 2028 (HolonIQ)7 — competition is already forcing vendors into the school segment at budget price points. Where the market matures, the product enters through pilots — and converts into a government contract in 12 months on average.8
The median time from a successful school pilot to a government contract in Russia is 12 months (Ed-Market Analytics). With scaling starting in 2026–2027, that produces a wave of government contracts in 2027–2028 — with room to spare before September 2030.
Demand is two-sided. 73% of teachers name task personalization as a priority tool (HSE)9. From the other end — 67% of parents in Russia and Kazakhstan are willing to pay extra for AI personalization (Sber Edu)10. The procurement officer ends up squeezed from both sides: by teachers in the classroom and by parents at the parent-teacher meeting.
The infrastructure threshold has been crossed. 38% of CIS schools reached 1:1 device coverage by 2025 (UNESCO)11 — above the critical mass needed to launch a mass-market product. An AI textbook is impossible without a device in every student’s hands; with one, mass adoption begins.
2025: pilots in select regions of Russia and Kazakhstan. 2026–2027: first government contracts based on the 12-month conversion cycle. 2028–2029: replication into neighboring regions via federal textbook registry analogues. 2030: ≥20% of CIS schools — the AI textbook in standard, not pilot, procurement.
Why this isn’t science fiction: the technological and market mechanics
The forecast above rests on a concrete mechanism — and it’s already running in production.
The student answers → the browser sends an xAPI event to the LRS → the model reads the personal profile from storage → generates an exercise at the right difficulty → the exercise appears on screen. The whole cycle takes under a second.
A system that adjusts content difficulty and sequence to the actual level and pace of a specific student — as opposed to a linear textbook that is identical for the whole class.
The central component of the loop is the LRS (Learning Record Store): an xAPI store accumulating every student interaction with the content. From these events the model builds a personal profile in real time: what was solved correctly, where the student got stuck, which task types produce systematic errors. The LRS is a mature industry standard, supported by Moodle, Canvas, and most enterprise LMSs. The infrastructure is already in schools — it just needs a model plugged in.
The main economic argument: 2025 frontier models generate adaptive exercises in under 1 second at ~$0.003 per request12. With 30 students and 20 answers per lesson — $1.80 per class. A set of methodological workbooks for one class costs $15–20 per school year and doesn’t react to a single answer.
The 2.3× effect Khan Academy recorded in the Khanmigo pilot13 is a direct consequence of this loop: the system readjusts on every answer without waiting for the final test. Inference speed and per-request cost are what turn a pedagogical idea into a product with measurable lift.
What could derail the forecast
The technology works. The demand exists. But the forecast is not deterministic — and an honest analysis requires naming three braking scenarios.
The main one — the federal textbook registry. A traditional school textbook in Russia takes 3–5 years to clear the Federal Textbook Registry14. No approval track exists for an adaptive digital product: the requirements were written for the paper format. Without a separate regulatory track for adaptive formats, every vendor goes through the full cycle — and some will simply leave the market before approval arrives.
The registry cycle (3–5 years), the children's personal data regime (xAPI/LRS require storing personal profiles), teacher resistance (retraining isn't budgeted in the standards). None of them technological — all solvable by regulation, not by engineers.
Children’s data privacy. LRS profiles record every student answer — legally sensitive territory. Neither Russia nor Kazakhstan has dedicated regulation for minors’ educational data. The first public leak incident could freeze government procurement for 2–3 years regardless of product quality.
Teacher resistance. Even among the 42% of schools that tested AI assistants by 202515, most stayed in pilot mode: the technology arrived, the methodological retraining didn’t. Without it, adoption becomes a burden instead of help.
Expectation calendar: 2025 → 2030
The risks are named — what remains is to fix concrete milestones against which the forecast can be verified or falsified.
2025: regulatory window open in Russia and Kazakhstan. 2026: first federal-level pilots in 3–5 regions. 2027: the key checkpoint — the first regional government contract. 2028: replication into 10+ regions, entry into the national registry. 2029–2030: mass procurement across ≥20% of CIS schools.
The median 12-month conversion cycle16 means every pilot launched in 2026 mathematically arrives at a contract by the end of 2027. Russia’s Ministry of Education has already opened the regulatory window17, Kazakhstan — the budgetary one18. The key difference between the tracks: Russia takes the market route (vendors → pilots → tenders), Kazakhstan — the programmatic one (the state funds from the top). The market route scales horizontally; the programmatic one guarantees passage through the budget cycle. If at least one track closes the cycle in 2027, the 20%-of-schools-by-2030 forecast stays on trajectory.
Every milestone is an independently verifiable event: open government procurement in Russia and Kazakhstan provides a public signal. If pilots convert into contracts by 2027 — the following milestones are realistic. If not — the forecast needs its timeline revised, not its thesis withdrawn.
-
Sal Khan, TED2023 ↩
-
Khan Academy, Khanmigo Pilot Report 2024 ↩
-
Russian Ministry of Education, Order №AI-2024/08 ↩
-
MDDIAI of Kazakhstan, press release 03.2025 ↩
-
Smart Ranking, Top-100 edtech companies 2025 ↩
-
Russian Academy of Education, school digitalization monitoring 2025 ↩
-
HolonIQ Global EdTech Market Report 2025 ↩
-
Ed-Market Analytics, edtech government procurement 2022-2025 ↩
-
HSE, Monitoring of Education Markets and Organizations 2024 ↩
-
Sber Edu, parent survey Q3 2025 ↩
-
UNESCO Digital Learning Landscape 2025 ↩
-
Anthropic/OpenAI pricing snapshot 2025 ↩
-
Khan Academy, Khanmigo Pilot Report 2024 ↩
-
Federal State Educational Standards, Federal Textbook Registry ↩
-
Russian Academy of Education, school digitalization monitoring 2025 ↩
-
Ed-Market Analytics, edtech government procurement 2022–2025. ↩
-
Russian Ministry of Education, Order №AI-2024/08. ↩
-
MDDIAI of Kazakhstan, press release 03.2025. ↩