AI in EdTech: The Readiness Gap No One Is Talking About
Every major EdTech conference currently has at least three sessions with “AI-powered” somewhere in the title. I get it. The power is real, and frankly, the numbers back it up. But after 13+ years in software and applications development, I’ve learned that the most interesting question is rarely about the technology itself. It’s about whether the ground beneath it is strong enough to hold. And in EdTech right now, the world is moving more than most presentations would have you believe.
Numbers Are Not Hype
Let’s be clear: the growing adoption of AI throughout education is not a fad. A 2025 study by the Higher Education Policy Institute found that 92% of students are using AI tools in their learning, up from 66% a year ago. That is not incremental change. That means a change in student behavior in twelve months.
On the market side, the global EdTech market is estimated at $187 billion by 2025 and is expected to grow at a CAGR of 10.8%, reaching $437 billion by 2033, while cloud-based deployments are expected to grow at a faster rate of 15.9% annually. [1]. These are not numbers from a niche segment. EdTech is now an important part of the global technology economy, and AI is accelerating its core. So yes, growth is real. But here is what is usually buried.
The Readiness Gap Nobody Talks About Enough
According to a RAND Corporation study published in 2025, while more than half of students and teachers now report using AI in school, teacher professional development, student training on responsible AI use, and school-level policies all fall short of that adoption rate. [2]. The part that really sticks with me is this: 76% of institutional leaders believe their users have received adequate AI training, but 45% of teachers and 52% of students report receiving 0 training. That’s not a small zoom error. That is a blind spot sitting in the middle of what should be a revolutionary story.
This readiness gap is not limited to education. It is seen in almost every field where the use of technology outstrips the organization’s readiness to support it. But education is more dangerous because the results reach the students, not just the workflow. From a development perspective, this is a common problem. It shows what happens when features are deployed without testing the actual user journey. The tool works in a controlled environment. It doesn’t work in the field. Failure is not in technology. It is in the assumptions made about how that technology will actually be used.
Where EdTech Platforms Really Fall Short
There is a particular pattern I keep seeing in the way AI is integrated into educational software. The team is building a state-of-the-art personalization engine. The benchmark results look solid. However, the platform is deployed in a school district where 40% of students access it on a low-end Android device with an unreliable connection. Recommendations are lacking. Dynamic testing is complete. The teacher-facing math dashboard requires three clicks to access and another two minutes to load. AI was not the problem. The infrastructure under it was the same.
Good educational software development starts with real use cases, not ideal cases. It designs degraded states as a core case, not an edge case. It does not take mobile first as a design preference but as a critical requirement, especially given that in many emerging markets, mobile learning already represents the first and sometimes only point of access for students.
This is more important than ever because the range of students serving EdTech is now wider than at any time in history. A platform that works well for a university student in a connected city but fails for a vocational student in a small town did not solve the problem of the readiness gap. It has just been optimized for its simplest version.
Three Shifts Worth Watching
Beyond the headline acquisition numbers, there are several structural changes that appear to be steady rather than cyclical. The first is that institutions are becoming more demanding consumers. They are still not impressed by the demo-room AI. They want evidence of learning outcomes. Studies have shown that students who use a well-designed AI tutor can achieve better performance benefits compared to those who use a basic AI chatbot, suggesting the quality of the implementation of larger issues, not just the presence of AI features.
Second is data management from legal concerns to strategic ones. Since only 10% of schools and universities have established policies for using AI according to UNESCO data, and regulators in many regions are tightening the requirements, platforms built with privacy as an architectural principle will surpass those that treat compliance as a checkbox.
The third is cooperation. Long-term contracts are increasingly on platforms that connect cleanly with student information systems, HR tools for business learning contexts, and third-party content libraries. The era of the standalone point solution in EdTech is rapidly diminishing.
What Makes AI Work in Education
The teams making real progress with AI in EdTech share some common practices. They invest heavily in understanding user context before specifying features. They create feedback loops that are strong enough to influence teacher decisions in real time, not just generate charts that sit on a dashboard. And they treat the quality of the underlying data pipeline as a first-order priority, because personalization algorithms are only as good as the behavioral data that feeds them.
There is also something to be said for restraint. Not all areas of the learning landscape benefit from the integration of AI. Institutions and vendors that think carefully about where AI actually adds value, and when it adds complexity without meaningful benefit, tend to build products that are more stable and reliable over time.
The 92% student achievement figure is impressive. But a more meaningful number might be one that tracks how many of those students feel that the AI they’re using is actually helping them learn, not just making them faster at completing tasks. That’s a rough estimate. This is also the right one.
