From the "Jagged Frontier" to the "Fractured Frontier": The Real Challenge of AI in K12
Inside the "fractured frontier" of AI in K12: fear, worries, hope and possibilities.
When Ethan Mollick and his colleagues introduced the concept of the “jagged frontier“ in 2023, it was a genuine intellectual contribution. Their research showed that AI’s capability profile was deeply uneven — outperforming human experts on some complex tasks while stumbling on seemingly simple ones — and that this jaggedness would persist and shift even as models improved. It gave us a precise and useful language for technology’s strange and unpredictable output.
A lot has happened since then. What began as ChatGPT has expanded into a rich and rapidly evolving ecosystem — Gemini, Claude, Copilot, Perplexity, and dozens more — each powerful, each increasingly accessible, and each improving at a pace that outstrips most institutions’ ability to respond. Mollick himself has evolved his thinking, now arguing that the critical question isn’t just where AI excels or struggles, but where the human bottlenecks are — the institutional, organizational, and cultural friction points that determine whether AI’s capabilities actually reach the people who need them.
In K–12 education, those bottlenecks are everywhere. And they don’t just slow adoption. They fracture it.
This is what I'm calling the “fractured frontier” — a deliberate reframe of Mollick's “jagged frontier”. For me the biggest challenge today is not the unevenness of what AI can do, but the wildly uneven, often contradictory ways people understand it, use it, and use it for good, which includes teachers, school leaders, districts, and students – in K12 and beyond. It is a fracture that is dividing those who are becoming AI literate (or fluent) and use it as a “co-intelligence” tool (again, borrowing from Mollick) and those who are being left behind.
But as fast as AI is growing and developing, it is only some who are rolling up their intellectual sleeves and jumping in and learning how to use it as a superpower. Mindfully using it as a human co-intelligence, creativity, and thinking tool. And it is very few who see the incredible value of their colleagues and organizations to be fluent in the use of AI to better achieve their aspirations and pursue their visions and missions. As a result, we are increasingly seeing those that have, and those that do not. Those who are actively supported in learning how to use AI and those that are, in many cases, being told that they are not even allowed to use AI. Hence, the equity gap. And one growing, not closing, as a result.
We have seen this pattern before. When the internet arrived in the ‘90s, its potential was undeniable — but the real story quickly became how those early adopters jumped into its utility and the difference between those schools, districts, and communities who proactively sought out its utility vs. those that didn’t know how to embrace its utility. Some individual educators built it into their practice and others worked in systems that had no idea how to integrate into their systems. The gaps – as you can imagine – predictably fell along lines of wealth, race, and resources. We named that the digital divide — and we spent the next two decades trying to close it, but with mixed results. Even where we closed the access gap, we didn’t always close the learning gap, because access without pedagogical transformation and sustained support was never enough.
We cannot afford to repeat that story yet again with AI, sustaining the educational equity we already know exists and persists in our schools and school systems today, but carries a ripple effect for their learners’ future. Those not becoming AI literate may be left out of the economic mobility engine we hoped education could help to address and instead continue to exacerbate the equity gap. If not addressed, our schools and school systems will continue to perpetuate the social and economic divide that continue to be a wound in our communities and society.
The Fracture by Role: Who Is Doing What Right Now
The most immediate way to see the fractured frontier is through the people living it. At every level of a school system, two radically different realities coexist — not in different states or different kinds of communities, but often in the same district, sometimes in the same building.
What makes this fracture so costly is not the variation itself — variation is normal in any large system. What makes it costly is that the variation is not random. It maps onto existing inequities. The schools most likely to be frozen are often the ones already under the most pressure: under-resourced, under-staffed, navigating community skepticism, and lacking the organizational capacity to invest in anything uncertain. The students least likely to receive thoughtful AI guidance are often the students who most need every possible advantage. The fractured frontier is not just an adoption problem. It is an equity problem.
The Fracture by Domain: Where the System Is Breaking
Zoom out from individual roles and the fracture becomes even more visible — running not just through people but through the structures and conditions that shape what’s possible in any school or district. Even in systems where leaders genuinely want to move forward, these domains reveal where the friction lives and why good intentions so rarely translate into sustained change.
Why the Fracture Persists: A Knowledge Problem, Not a Courage Problem
It would be easy — and wrong — to reduce the fractured frontier to a story about courage. The educators and leaders in the frozen column are not there because they lack character or ambition. They are there, in large part, because nobody has shown them a credible, replicable, visible model of what getting this right actually looks like in a context close enough to their own to feel real.
This is one of the most durable findings in the diffusion of innovation literature. Everett Rogers identified observability — the degree to which the results of an innovation are visible to others — as one of the key determinants of whether a new practice spreads. If educators cannot see AI integration working in practice, in real schools with real students, in contexts that feel relevant to their own situation, adoption stalls regardless of how compelling the theoretical case is.
Compelling examples do exist. Gwinnett County Public Schools in Georgia opened Seckinger High School in 2022 as what many consider the nation’s first AI-themed public school — embedding AI across every subject through a three-tiered model that gives all students foundational AI literacy while enabling deeper exploration for those who want it. An external evaluation conducted by Georgia State University researchers in August 2025 documented what made it work: a shared AI Learning Framework that gave teachers a common language across subjects, sustained professional development, a culture that treated experimentation as a learning opportunity rather than a liability, and a deliberate commitment to ethics and student agency at the center. That is a replicable model — not just an inspiring story.
But Seckinger is one school — purpose-built and purpose-led. What the field urgently needs is a range of visible, documented examples across genuinely different contexts: the mid-sized urban district that built something from scratch with a modest budget, the rural superintendent who figured out how to bring a skeptical community along, the department team that started with one subject and scaled, the lone pioneer teacher whose practice became a school-wide model because someone finally paid attention. The examples need to be close enough to a hesitant leader’s own reality that they can say — not just “that’s impressive” — but “that could be us.”
Right now, those examples exist but are scattered — inconsistently documented, rarely packaged for transferability, and almost never organized around what a leader in a different context would actually need to act. The thoughtful district in Iowa has built something remarkable. The superintendent in Georgia has never heard of it. That is a knowledge diffusion problem wearing a fear costume. And it is entirely solvable.
What AI Can Actually Do — When Educators and Learners Use It Well
Before naming what needs to happen, it’s worth being precise about what’s actually at stake — because AI’s potential impact on learning is categorically different from any educational technology that came before it.
Previous ed-tech was fundamentally a delivery mechanism. A laptop delivered content. A smartboard displayed it. An adaptive platform sequenced it. None of those tools could think with a student, respond to where their reasoning broke down, ask the question that pushed their thinking further, or provide immediate personalized feedback at eleven o’clock at night when no teacher was available. They made delivery more efficient. They didn’t change the fundamental dynamic between a learner and the support available to them.
AI does something structurally different. Used well by a skilled educator, it can expand every student’s access to knowledge — allowing the curious student in a rural district with no AP courses to explore college-level content, or the English language learner to get explanations in their home language. It can genuinely personalize and differentiate learning at a scale no human teacher, however gifted, can sustain across thirty students simultaneously. It can provide continuous, precise, formative assessment — identifying not just whether a student got the right answer but where their reasoning broke down and what specific intervention would address it. It can surface the right resource at the right moment for the right learner rather than the same resource for everyone. And it can engage students in genuine intellectual dialogue — patient, responsive, Socratic — extending the thinking that a teacher initiates into the spaces and times no teacher can occupy.
Benjamin Bloom identified what he called the two-sigma problem in 1984: students who receive one-on-one tutoring perform two standard deviations better than students in conventional classroom instruction. That finding has haunted education research for forty years precisely because no one could figure out how to deliver that kind of individualized attention at scale. AI is the first serious candidate for a solution — not because it replaces the teacher, but because it extends what a skilled teacher makes possible to every student, not just the ones who raise their hand or can afford private tutoring.
And unlike every previous pedagogical innovation — project-based learning, differentiated instruction, competency-based education, personalized learning — AI is not fundamentally constrained by the limits of human execution. Those innovations were sound in theory but ceiling-ed in practice by how much any teacher can hold in their head and act on across thirty students at once. AI doesn’t have that ceiling. Which means that as models improve over the next three to five years, the distance between what’s possible for a well-supported learner and what’s available to an unsupported one could grow faster and wider than any gap the digital divide ever produced.
That framing is exactly right. AI doesn’t transform learning. Educators and learners using AI with skill, intention, and support transform learning. The technology is the accelerant. The human is still the fire. An AI that gives students answers isn’t expanding their access to knowledge — it’s shortcutting the cognitive work that produces it. An AI deployed without pedagogical intentionality is just a faster version of the same passive content delivery that ed-tech has always offered. The tool is only as powerful as the educator who knows how to wield it and the learner who has been taught to use it well. That is not a caveat on AI’s potential. It is the entire point.
Moving Beyond the Fractured Frontier
The path forward is not another report on AI’s potential, another policy framework from a state capital, or another conference session that inspires without equipping. What’s needed is something more deliberate and more difficult: a sustained, field-wide commitment to making visible what is actually working — at the classroom level, the school level, and the district level — with enough specificity and contextual range that a hesitant educator anywhere in the country can find themselves in the examples and know what to do next.
That means case studies built from practice, not theory — documented with the rigor of research and the accessibility of storytelling. It means frameworks developed with educators, not handed down to them. It means networks that connect the pioneer in Room 214 with teachers down the hall, and the innovative district in one state with the frozen one in another. And it means something more specific than loose “network” affiliations: it means creating and then supporting communities of practice. Small groups of five to seven educators or leaders, meeting every other week across five or six sessions, with a clear charge to try something, document what happened, and bring it back to the group. To document and curate what educators did – and the impact. To learn from one another and support one another. Then take the insights gained and lessons learned and share them with the rest of their community so that they too are “superpowered” in how to employ AI. Again, grounded in the real world application of AI in their context, with their learners, and with their aspirations.
The problem is not that we do not know how to do this. It’s that those in positions of leadership are not committing to creating and supporting the means for active and communal exploration and discovery across educators and leaders in their systems. Individual teachers might be learning a lot about how they can use AI to enhance the learning of their students, but that isn’t contributing to every teacher (or leader’s) AI literacy – in a way that could serve all of their learners. Absent an architecture for active exploration of how AI could serve their students’ learning, the fractured frontier remains fractured. Given this we individuals, schools, and districts committing to pursuing the possibilities with AI while the greater majority of educators and leaders are standing still — waiting for permission, waiting for a map, waiting for others to show them what is possible.
Mollick gave us the jagged frontier to understand how AI is shifting, growing, and changing over time, rapidly providing us with more and more possible uses. This is a given. What doesn’t have to be a given is the inertia in our schools and school systems to take advantage of these new tools. Absent proactive exploration of the potential utility of AI across educators in schools and districts, the few super inquisitive educators and leaders disposed to experiment with the utility of AI are learning a lot. And finding AI to be an incredible tool in service of the children they serve. While others are not acting, but waiting. In short, it is not the technology that is a bottleneck, but ourselves. Out of fear, many are not wanting to put their toes in the water. While a few are courageously swimming in the possibilities.
If you wish to add your story to this unfolding narrative, simply email me at cunger.neu@gmail.com and let’s set up a time to chat.
RESOURCES
For readers who want to go deeper, the following sources informed this piece and offer useful entry points into the broader conversation.
UNDERSTANDING AI’S CAPABILITIES AND THE JAGGED FRONTIER
Centaurs and Cyborgs on the Jagged Frontier
Ethan Mollick · One Useful Thing (Substack) · 2023
The post that popularized the “jagged frontier” concept — Mollick’s accessible explanation of why AI’s capability profile is deeply uneven, how humans can work alongside AI most effectively, and why the shape of that frontier matters more than its overall size.
Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality
Dell’Acqua, McFowland, Mollick et al. · Organization Science · 2026
The peer-reviewed research behind the jagged frontier concept, based on a field experiment with Boston Consulting Group. Finds AI assistance improves performance for tasks inside its capability frontier while creating risk of overreliance for tasks outside it — with significant implications for how organizations train people to work alongside AI.
THE STATE OF AI ADOPTION IN K–12 EDUCATION
Artificial Intelligence Policies in K-12 School Districts in the United States: A Content Analysis
Journal of Research on Technology in Education · March 2025
A peer-reviewed content analysis of 9,229 K–12 districts finding that as of May 2024, only 14.13% had any AI policy at all. Among those that did, policies focused primarily on academic integrity and responsible use — with far less attention to how AI can enhance learning.
AI Use in Schools Is Quickly Increasing but Guidance Lags Behind
Doss, Bozick, Schwartz et al. · RAND Corporation · 2025
A nationally representative survey of K–12 teachers, school leaders, district leaders, students, and their parents — the first to triangulate all five populations simultaneously. Finds rapid growth in AI use while professional development, student training, and district policy all lag significantly behind. Also documents a persistent equity gap: low-poverty districts substantially outpace high-poverty ones in teacher AI training, with RAND projecting the disparity will not close without targeted intervention.
More Than Half the States Have Issued AI Guidance for Schools
Fitzgerald · Stateline / States Newsroom · July 2025
A nonpartisan policy journalism report confirming that at least 28 states and the District of Columbia have issued K–12 AI guidance — and documenting the persistent gap between state-level frameworks and the practical support districts actually need to act on them.
State AI Guidance for Education
AI for Education · Updated 2025
A continuously updated tracker and summary of state-level AI guidance documents for K–12 education across the U.S. Useful for understanding what states have acted, what their frameworks emphasize, and the significant variation in specificity and usefulness across state approaches.
SCHOOLS AND DISTRICTS DOING IT WELL
What California Teachers Are Trying, Building, and Learning with AI Center on Reinventing Public Education (CRPE) with Silicon Schools Fund · August 2025
A research study of 18 California schools that piloted AI tools to address core instructional challenges — including learning gaps, low engagement, and time constraints — involving over 80 teachers and administrators across 30+ pilots. Key finding: when educators were given genuine support and autonomy, they quickly developed bold, creative uses of AI that deepened rather than displaced human relationships in learning.
An Ethical and Equitable Vision of AI in Education: Learning Across 28 Exploratory Projects Noakes, Shell, Murillo et al. · Digital Promise · October 2024 · Funded by the Bill & Melinda Gates Foundation
Documents how 28 district, edtech developer, and researcher teams tested AI approaches aimed at equitable outcomes for students who are Black, Latino/e, and from low-income backgrounds. Consistent finding across all projects: teachers and students must be at the center of AI development for tools to actually work in classrooms. One of the most grounded field-based portraits of what responsible AI implementation looks like across diverse school contexts.
The AI-Ready Pilot at Seckinger High School
Georgia State University researchers for Gwinnett County Public Schools · August 2025
An external evaluation of the nation’s first AI-themed public school documenting what made the model work: a shared AI Learning Framework across all subjects, sustained professional development, a culture of experimentation, and student agency at the center. Includes specific replication guidance for other districts.
TEACHING STUDENTS TO USE AI WELL
AI Competency Framework for Students UNESCO · September 2024
The global reference framework that national AI education strategies are built from. Defines 12 competencies across four dimensions — human-centred mindset, ethics of AI, AI techniques and applications, and AI system design — structured across progression levels from understanding to creating. Designed to guide policymakers, educators, and curriculum developers in equipping students to engage with AI responsibly, critically, and ethically.
Preparing K–12 Students With AI Literacy: Proposed Framework, Progression, and Task Design Principles Chakraburty, Ober, and Liu · ETS Research Report Series · November 2025
A peer-reviewed framework from Educational Testing Service — one of the world’s most respected measurement institutions — offering a theoretically grounded definition of AI literacy and a developmentally appropriate learning progression for K–12. Introduces practical task design principles that bridge instruction, assessment, and skill development, giving educators a usable roadmap rather than abstract goals.
Five Lessons for Schools to Prepare Students and Teachers to Use AI
Child Trends · November 2025
Drawing on research by Drs. Douglas Fisher and Nancy Frey of San Diego State, this report distinguishes three approaches: teaching about AI, teaching for AI, and teaching with AI. Offers a practical framework for districts moving from isolated classroom experiments to coherent, system-wide AI integration.
It’s Not Magic: How These Schools Are Teaching AI Literacy
Education Week · November 2025
Profiles of schools — including New Jersey’s Passaic district — building deliberate AI literacy programs for K–12 students. Documents specific classroom approaches, including an AI acceptable-use rubric designed as a structured conversation starter about responsible use rather than a restriction policy.
LEADERSHIP, PROFESSIONAL DEVELOPMENT, AND SYSTEM CHANGE
AI Early Adopter Districts: The Promises and Challenges of Using AI to Transform Education Center on Reinventing Public Education (CRPE) · August 2025
Examines how 27 "Early Adopter" school districts approached systemic AI adoption during 2024–25, drawing on surveys, focus groups, and interviews with district leaders. Finds most districts remain in early, fragmented stages of experimentation, with a small vanguard strategically embedding AI into broader transformation agendas. The most grounded empirical portrait of what district-level AI leadership actually looks like in practice.
K-12 Gen AI Maturity Tool CoSN (Consortium for School Networking) and Council of the Great City Schools · Updated 2024–2025
The most widely used practical leadership framework in the field. Covers seven domains — Leadership, Operations, Data, Technology, Security, Legal, and Academic AI Literacy — with a three-level maturity rubric guiding districts from emerging to mature practice. Actively referenced by state departments of education and Gates Foundation-funded capacity building initiatives. The most actionable self-assessment tool available for district leaders.
Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations U.S. Department of Education, Office of Educational Technology · 2023
The federal government's foundational policy report on AI in education, written to guide educational leaders, teachers, and policymakers. Addresses the urgent need for shared knowledge and responsible policy frameworks as AI becomes embedded in educational technology systems. Carries the highest institutional credibility for a K-12 practitioner audience and remains the baseline document most serious district-level policy conversations reference.
THE IMPACT OF AI ON LEARNING: CURRENT EVIDENCE
OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education OECD · January 2026
The highest-credibility international synthesis available on AI and learning. Finds that generative AI can support learning when guided by clear pedagogical principles — but when used without that guidance, it enhances task performance without producing real learning gains. The central challenge is not access to AI but the conditions under which it supports thinking rather than replacing it.
Learning Outcomes with GenAI in the Classroom Microsoft Research · October 2025
Finds that how students engage with AI determines whether it deepens or displaces learning. Students who iterate — re-prompting, editing, brainstorming — significantly outperform those who copy-paste. Students who received in-depth AI literacy instruction were 64.5% faster and improved a full letter grade on professional writing tasks. The evidence that AI fluency, not AI access, is the variable that matters.
The Effectiveness of AI-Supported Personalized Feedback on Students’ Learning Outcomes and Motivation: A Meta-Analysis Wang, Wang, Chen et al. · Journal of Educational Computing Research · 2026
A meta-analysis of 40 peer-reviewed studies involving 5,849 participants finding that AI-supported personalized feedback produces a moderate effect on learning outcomes and a strong effect on learning motivation. The first systematic meta-analysis to isolate AI-generated feedback specifically — making it the most rigorous available evidence for AI’s role in the formative assessment process at the heart of learning.
What the Research Shows About Generative AI in Tutoring Brookings Institution · February 2026
A synthesis of emerging empirical evidence finding substantial learning gains, greater knowledge transfer, improved motivation, and efficiency across multiple AI-assisted learning studies. Central conclusion: the optimal model is human-AI collaboration — teachers guide and monitor rather than cede the learning process. Authoritative and accessible bridge between academic research and practitioner application.
The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring Bloom, B.S. · Educational Researcher, Vol. 13, No. 6 · 1984
Bloom’s landmark finding that one-on-one tutoring produces learning outcomes two standard deviations better than conventional classroom instruction — shifting the average student from the 50th to the 98th percentile. The challenge of delivering that level of individualized attention at scale is the foundational problem AI-assisted learning is now positioned to address.