The Future We're Building

The Evolution of EdTech — Part 13 of 13

By Clarence Stephen | May 2026

2,300 words | 9 min read

We've traveled a hundred years in twelve chapters — from a wooden box with four keys at a psychology conference in 1924 to an AI-powered reading platform generating illustrations in real time in 2025. Along the way, the same arc repeated: a new technology appears, promises to transform education, partially delivers, and settles into modest utility while the next wave builds.

Teaching machines. Television. Personal computers. The internet. Learning management systems. MOOCs. iPads. AI chatbots. Each was supposed to be the revolution. None was. And yet, cumulatively, they changed education profoundly — not through any single breakthrough but through a gradual accretion of tools, capabilities, and expectations that no one generation can fully perceive.

The question for this final chapter is whether the current moment — the convergence of generative AI, local computing power, and open-source models — is genuinely different. Whether the cycle will repeat. Or whether we are, after a century of false starts, at an actual inflection point.

I believe we are. Here's the case.

The Convergence

Three developments are happening simultaneously that have never coincided before.

AI that understands language. Large language models — GPT-4, Claude, Qwen, Llama, and their successors — can read, write, analyze, summarize, translate, and converse with a facility that was science fiction five years ago. This isn't keyword matching. These models interpret ambiguous text, infer meaning from context, adjust explanations for audience, and generate original content that is coherent, relevant, and frequently insightful.

For education, this means the adaptive, conversational, infinitely patient tutor that Pressey and Skinner envisioned is now technically feasible. Not perfect. Not human-equivalent. But functional, and improving at a pace that makes year-over-year comparisons almost meaningless.

AI that creates visual content. Image generation models — SDXL, DALL-E, Midjourney, and successors — produce detailed, coherent images from text descriptions in seconds. This capability didn't exist in any meaningful form before 2022. It transforms every text-based experience into a potential multimedia experience. Any book can be illustrated. Any concept visualized. Any abstract idea made concrete.

For education, this means the visual dimension of learning — consistently shown by research to be critical for comprehension, retention, and engagement — can be generated on demand, for any content, at any scale, at near-zero marginal cost.

Local, open-source AI. The third development is the most consequential and the least discussed. Until recently, state-of-the-art AI required massive cloud infrastructure and corporate API subscriptions. The most powerful models were proprietary, expensive, and controlled by a handful of companies.

That's changing fast. Open-source models like Llama, Qwen, and Stable Diffusion run on consumer hardware — a single GPU costing $1,000 to $2,000. Local inference means no data leaves the device, no privacy concerns, no per-query costs, no dependency on corporate infrastructure.

For education, this means AI-powered tools can be deployed in schools, libraries, community centers, and homes without ongoing subscription costs, without robust internet requirements, and without the privacy concerns that have legitimately plagued cloud-based EdTech.

These three capabilities — language understanding, visual generation, and local deployment — have never existed together. Their convergence creates possibilities that weren't just difficult two years ago. They were impossible.

Where We Go Next

The next five years will see AI integrated into education at every level, in every subject, in every modality. Here's what I expect:

AI tutoring becomes standard, not exceptional. By 2030, AI tutoring will be available to every student in the developed world and increasingly in the developing world. These systems won't replace human teachers — nothing will — but they'll provide individualized support that has historically been accessible only to the affluent. The 2 Sigma gap won't close, but it'll narrow meaningfully. Khanmigo is the beginning; what follows will be more capable, more specialized, and more widely deployed. The competition among Khan Academy, Duolingo, MagicSchool, and dozens of startups is fierce, which is good for students.

Multimodal learning becomes the default. Text-only education will increasingly feel as dated as blackboard-only education feels today. AI-generated illustrations, diagrams, animations, and interactive visualizations will be woven into reading platforms, textbooks, and lesson plans. Students will learn from content that adapts not just to their level but to their preferred modality.

This is where what I'm building with Readify points toward something larger. The technology that illustrates a book can also visualize a lecture, animate a concept, depict a historical event. Visual enrichment isn't a niche application. It's the future of content delivery.

AI agents absorb the administrative burden. Teachers spend an estimated 40 to 50 percent of their time on non-instructional tasks — grading, planning, communicating with parents, documentation. AI agents that automate or streamline these tasks can return hours per week to actual instruction. According to a 2025 Microsoft report, 86 percent of education organizations are already using generative AI — the highest adoption rate of any industry. This isn't a projection. It's the present.

The equity question becomes urgent. Every technology in this series widened the gap before narrowing it. AI will follow the same pattern. In the short term, affluent schools and students will adopt AI tools faster and deploy them more effectively. The digital divide will manifest as an AI divide.

But open-source AI offers a different trajectory. When powerful models run on a $1,500 computer instead of a $100,000 cloud subscription, the economics of equity change fundamentally. Local AI doesn't require broadband. It doesn't require recurring payments. It doesn't require corporate partnerships. It requires hardware, open-source software, and someone willing to set it up. That's not a complete solution to structural inequality. But it's a different kind of problem — and a more tractable one — than the barriers that have defined EdTech for decades.

Readify's Roadmap

I want to share where I'm taking Readify, not as a promise but as a direction.

Readify Aspen — the cloud tier — will bring the core experience to any device with a browser. No GPU required. This is the tier designed for broadest reach: students, adult learners, libraries, anyone who wants visually enriched reading without dedicated hardware.

Readify Sakura — the laptop and desktop tier — will use SDXL Lightning for fast, efficient image generation on consumer GPUs with 8-16GB of VRAM. Local, private AI at an accessible price point.

Readify Sequoia — the workstation tier — remains the flagship: dual GPU, full SDXL with refiner, maximum quality and speed for institutions, power users, and research environments.

Beyond the tiers:

Why This Decade Matters

The AI in Education market is projected to reach $41 billion by 2030, growing at 43 percent annually. Market numbers don't automatically translate to impact. But they represent an unprecedented concentration of capital, talent, and attention at the intersection of AI and learning.

The question is where that capital flows. If the pattern of the last century holds, most will flow toward the already-served: college students, corporate trainees, affluent districts. The 130 million adults who can't read comfortably will remain invisible. The 69 percent of fourth graders not reading at grade level will continue to fall behind.

But the pattern doesn't have to hold. The tools available today — open-source language models, open-source image generators, commodity GPU hardware — make it possible to build education technology that is powerful, private, and affordable. The technology isn't the bottleneck. The bottleneck is will.

A Personal Note

I started this series as a researcher tracing the history of a field I've studied for years. I want to end it as a founder, speaking directly.

I built Readify because I believe reading is the most important skill in human civilization, and the technology to help struggling readers has been possible for years — nobody built it. Not because it was impossible, but because the market incentives pointed elsewhere.

I have a physics degree from Yale and experience at Tiger Global and Morgan Stanley. I could be optimizing portfolio returns. Instead I'm optimizing image generation pipelines on RTX 4090s. The reason is simple: the reading crisis is the most important problem I can work on, and the tools to address it have never been better.

"Every story deserves to be seen" is not a tagline. It's a conviction. Every book — whether The Great Gatsby or a children's picture book or a biology textbook — contains visual information that most readers never access because it exists only as text. AI can make that information visible. For struggling readers, that's not a feature. It's a bridge — from text to meaning, from confusion to comprehension, from giving up to keeping going.

The Last Hundred Years

A hundred years ago, Sidney Pressey stood at a conference with a wooden box and a radical idea: that machines could help people learn. He was right. The machines just weren't ready.

Today, the machines are ready. Large language models understand and generate text with superhuman breadth. Image models visualize any concept in seconds. Open-source communities have made these tools available to anyone. Consumer hardware runs them locally, privately, at marginal costs approaching zero.

The question is no longer whether technology can help people learn. The question is whether we'll direct that technology toward the people who need it most.

Pressey dreamed of freeing teachers from drudgery. Skinner dreamed of personalized instruction for every student. Papert dreamed of children creating with computers. Khan dreamed of a free education for anyone, anywhere.

None of those dreams has been fully realized. But each moved the field forward. Each contributed tools and insights that made the next advance possible. And each was driven by the same basic conviction: that education can be better than it is.

That conviction drives Readify. That conviction drives this series. And that conviction — tested by a hundred years of hype, disappointment, and incremental progress — is about to be vindicated.

The tools are here. The need is here. The only thing missing is the decision to act.

Let's go.

This concludes "The Evolution of EdTech," a 13-part series tracing educational technology from 1924 to 2026. Thank you for reading.

Try Readify at app.ireadifybooks.com. Follow the journey at ireadifybooks.com.

Sources

Clarence Stephen is the founder of Readify, an AI-powered literary platform. Yale B.Sc. Physics. Previously: Tiger Global, Morgan Stanley. Learn more at ireadifybooks.com

What could AI do for your classroom?

Readify uses local AI to adapt reading material to each student's level — generating illustrations, simplifying vocabulary, and tracking comprehension in real time. No student data leaves the building.

Bring Readify to Your School Try the Platform

Liked this article?

Get the full Evolution of EdTech series delivered to your inbox, plus weekly insights on AI in education.

Subscribe to Readify Branches