Schools Are Done Experimenting With AI — Now Comes the Hard Part

2026-05-08 · Nia

There's a pattern in education technology that anyone who's been paying attention can recognize: a shiny new thing arrives, schools rush to adopt it, and then — two to three years later — the real questions surface. Who's responsible when this goes wrong? Where's the data going? Is anyone actually learning more?

We're at that inflection point with AI in schools. And for the first time, the data confirms it.

The CoSN Numbers Tell the Real Story

The Consortium for School Networking (CoSN) just released its 2026 U.S. State of EdTech report, surveying 607 K-12 leaders across 44 states. The headline finding? Cybersecurity and governance have reclaimed the top spot as districts' primary concerns — after AI briefly dethroned them in 2025.

But this isn't regression. It's maturation.

Here's what the numbers actually show:

79% of districts have established official AI guidelines (up from 57% in 2025)
88% of districts have some kind of AI initiative underway
96% of leaders believe AI can benefit education
Only 19% say their AI approach remains undefined

In a single year, we've gone from "what is this thing?" to "how do we govern it responsibly?" That's not a retreat from AI — it's the necessary next step that most industries skip entirely.

The Budget Reality Check

Here's where it gets uncomfortable. While confidence in AI is sky-high, the resources to implement it safely are nowhere near adequate:

65% of districts cite insufficient budgets as their biggest barrier to cybersecurity
52% lack staff training or expertise to handle new threats
75% of leaders are "very concerned" about AI-enabled cyberattacks

This is the gap that nobody talks about at ed-tech conferences. It's easy to demo a chatbot helping a kid with algebra. It's much harder to fund the security infrastructure that protects 50,000 student records from increasingly sophisticated AI-powered attacks.

David Schuler, executive director of the School Superintendents Association (AASA), framed it well: "This is not simply about managing devices or systems — it's about building coherent, future-ready organizations that can adapt to change while staying focused on student outcomes."

Personalization: The Promise vs. The Reality

Meanwhile, one of AI's most hyped use cases in education — personalized learning — is getting its own reality check.

Education Week recently profiled how teachers like Al Rabanera at La Vista High School in Fullerton, California are using large language models to create math lessons connected to students' interests. Rabanera asked AI to help design a lesson about rate of change using U.S. Department of Labor income data. One student's immediate reaction: "Whoa, Mr. Rab! I'm gonna get paid less 'cause I'm a girl?"

That's the ideal outcome — a lesson that makes abstract math personal and provocative.

But even Khan Academy, one of the earliest adopters of AI tutoring, had to scrap a personalization feature from its Khanmigo chatbot. The results were underwhelming: no clear improvement in either academic progress or student engagement.

The technical challenges are real. Researcher Candace Walkington at Southern Methodist University found that AI frequently generates math problems that don't make real-world sense — like a concert scenario with 400 decibels (a physical impossibility) or only 9 people attending an Olivia Rodrigo show. The math might be technically correct, but the context is absurd.

Why This Matters for Builders

If you're building ed-tech products — or any AI tool for institutional use — there are three lessons embedded in this data:

1. Governance is a feature, not a roadblock.

Districts are using procurement as their primary governance mechanism. 56% now require vendors to provide specific safety information before adoption. If your product doesn't have clear, documentable safety practices, you're already losing deals. The "move fast and break things" era is over in education.

2. The "understaffed for instruction" problem is your opportunity.

Here's a telling stat: 66% of districts report adequate staffing for core IT (networks, maintenance), but 58% are understaffed for instructional use of technology. The gap isn't in infrastructure — it's in implementation support. Products that come with built-in onboarding, teacher support, and professional development will win over those that just ship features.

3. Interoperability beats innovation.

CoSN identified that few districts require "key education technology quality indicators" like evidence-based design, inclusivity, interoperability, or usability metrics. But that's changing. The districts that are ahead of the curve are treating AI vendors as long-term institutional partners. If your product is a walled garden, you're building a dead end.

The Organizational Silo Problem

One detail in the report that deserves more attention: organizational silos have been cited as a top-three challenge for 9 out of the last 13 years of CoSN survey data. Think about that. For over a decade, the fundamental barrier to effective ed-tech hasn't been the technology — it's been how schools are organized.

AI can't fix org charts. It can't eliminate the disconnect between a CTO making purchasing decisions, a curriculum director who doesn't know what was purchased, and a teacher who discovers a tool exists six months after it was deployed.

The districts that will actually benefit from AI are the ones tackling this structural problem first. Technology is never just a technology problem.

Where This Is Heading

Here's my take: 2026 will be remembered as the year education stopped asking "Should we use AI?" and started asking "How do we use AI without breaking everything else?"

That second question is harder, less glamorous, and far more important. It requires:

Clear data governance frameworks
Sustainable funding models (not one-time grants)
Breaking down silos between IT, instruction, and administration
Holding vendors accountable for outcomes, not just features

The schools getting this right won't make headlines for flashy AI demos. They'll be the ones where, three years from now, students are actually learning more — and their data is still safe.

That's the hard part. And we're just getting started.