CT Education Reform
Education systems change slowly — except when they don’t. Between 2024 and 2026, AI literacy went from a niche concern to legislated curriculum in dozens of jurisdictions, and the old consensus that “everyone should learn to code” collided head-on with tools that write code on request. This page maps the reform landscape and the debate underneath it.
The AI literacy wave
United States. A 2025 executive order directed federal agencies to advance AI literacy in K-12, and the Department of Education made AI a grant-making priority. States moved faster than Washington: by early 2026, more than half had issued AI-in-education policy, and the 2026 legislative session brought a wave of binding mandates. Representative examples:
- Georgia made computer science — explicitly including AI — a high-school graduation requirement, phasing in by 2031–32.
- New Jersey requires districts to teach AI concepts, skills, and ethics across K-12.
- Idaho mandated a statewide AI framework, AI literacy standards, educator training, and data-privacy rules — while prohibiting AI from replacing human teachers.
- Ohio, Oklahoma, South Carolina, Massachusetts, and New York mandated district-level AI policies and teacher training.
International. The OECD and European Commission released the AI Literacy Framework for Primary and Secondary Education in 2025 — the first credible international standard, organized around engaging with, creating with, managing, and designing AI. China deployed a mandatory national AI curriculum from primary through secondary school, the most comprehensive anywhere. Finland extended its famed media-literacy program to cover AI, framing it as civic defense against disinformation. And in the clearest institutional signal available, PISA 2029 will assess AI literacy internationally — placing it alongside reading, mathematics, and science as a measured competency.
The coding-education debate
The awkward question underneath the reform wave: should children still learn to program, now that AI programs?
The case against, in its strongest form: AI now writes most routine code, so teaching syntax is teaching a task already automated — the 2026 equivalent of drilling long division for careers in accounting. Curriculum hours are scarce; spend them on prompting, evaluation, ethics, and domain knowledge.
The case for rests on three arguments that gained evidence through 2025–26:
- Verification requires understanding. You cannot evaluate code you could never have written. The vibe-coding literature itself shows this: studies find CS achievement predicts proficiency at AI-assisted development, and the loudest warnings — from veterans like Robert C. Martin, and from Andrew Ng’s objection to the “vibes” framing — are precisely that accepting unexamined AI output produces brittle, insecure systems.
- Programming is how CT becomes concrete. Decomposition and abstraction taught purely on paper stay inert. Code is the medium where fuzzy thinking fails visibly and immediately — that feedback loop is the pedagogical point, independent of whether the student ever codes professionally.
- Someone must build and govern the AI layer. A civilization running on systems nobody understands is a civilization that cannot fix them.
The emerging synthesis — visible in the OECD/EC framework and the better state standards — is neither “code more” nor “code less” but reweighting: less syntax drill and boilerplate production; more problem formulation, specification, testing, debugging, data reasoning, and critique of AI output, with programming retained as the laboratory in which those skills are exercised.
What changes in the classroom
Reform on paper meets three hard realities in practice:
- Teacher capacity is the bottleneck. Research on preservice teachers finds narrow conceptions of CT, low confidence teaching it, and difficulty spotting CT opportunities outside STEM. Mandates without sustained teacher education produce checkbox compliance. Nearly every serious 2026 state bill includes teacher-training funding for exactly this reason.
- Assessment must change or nothing changes. If the exam still rewards producing working code, students will produce it — with AI, learning little. The assessment-research community is responding with process-based and rubric-based instruments (see Research Frontiers); PISA 2029 will force the issue internationally.
- Scaffolded AI use beats prohibition and free-for-all alike. The consistent finding of 2025–26 classroom studies: GenAI improves students’ computational thinking when the pedagogy requires students to analyze, verify, and reconstruct AI output — and does little or nothing when students simply consume answers. The design variable is not whether AI is allowed, but what cognitive work remains mandatory.
What this means for you
- Educators: map your existing CT practice onto your jurisdiction’s AI literacy standard — the overlap is large, and you are likely closer to compliance than the policy language suggests. Then redesign assessments before redesigning lessons.
- Parents: the question to ask a school is not “do you teach AI?” but “what must students still do themselves, and how do you check?”
- Self-learners: the reform consensus is a decent personal curriculum — problem formulation, decomposition, data reasoning, verification — and the Hands-on Labs on this site are built to that spec.