Last December I wrote about AI meeting education's crisis, the billions of students pushed online and the infrastructure that buckled under the weight. The question then was whether the investment would follow.
It did.
The money arrived
Through the first three quarters of 2021, global EdTech VC funding has already reached $14.9 billion, on pace to shatter every previous record. There are now 32 EdTech unicorns worldwide, with 61 mega-rounds exceeding $100 million in the past year alone. In the U.S., edtech companies raised over $3.2 billion in just the first half, already surpassing most full-year totals from the past decade.
The pandemic created urgency. The funding is creating infrastructure. But money alone doesn't guarantee we're building the right things.
What personalized learning looks like today
The pitch deck version of personalized learning is elegant: AI that adapts to each student, meeting them where they are, adjusting pace and content in real time. The classroom version is messier.
The best adaptive platforms in 2021 do adjust content based on individual performance. They identify where a student struggles and serve targeted practice. Some use NLP to understand written responses, not just multiple choice answers. That's genuine progress from even two years ago.
But there's a gap between "adaptive" and "personalized." A system that adjusts difficulty is adaptive. A system that understands how a student learns, what motivates them, when they're frustrated versus confused versus bored, that's personalized. We're firmly in the first category and still reaching for the second.
A friend who teaches middle school math told me about her experience with an adaptive platform this fall. "It's good at knowing when a kid gets a problem wrong," she said. "It's terrible at knowing why. Was the math hard? Was the word problem confusing? Did they just stop caring after twenty minutes?" The software adjusts the what. The teacher still has to figure out the why.
Where this is going
Three trends suggest where personalized learning is headed.
First, language models are making tutoring more conversational. Instead of branching decision trees, the next generation of AI tutors will be able to engage in actual dialogue: asking clarifying questions, explaining concepts in multiple ways, responding to "I don't get it" with something more useful than repeating the same explanation at a slower pace.
Second, multimodal analytics will go beyond test scores. Time-on-task, engagement patterns, how a student navigates between resources, all of these signals can help a system understand not just what a student knows but how they're learning. The tools to collect and interpret these signals are maturing fast.
Third, and this is the one that keeps me up at night, there's a real risk that personalized learning optimizes for the wrong metrics. Completion rates. Time on platform. Engagement scores. These are easy to measure and look great in a board deck. But they're proxies, not outcomes. A student who breezes through a module isn't necessarily learning. A student who struggles and takes twice as long might be learning more deeply.
The discipline ahead
The money and the technology are here. The pace of investment says the market believes in personalized learning. But belief isn't enough.
The question for 2022 and beyond isn't "can we personalize learning?" The tools say yes. The question is whether we have the discipline to measure what matters, not just what's easy to count, and to build systems that serve the student, not just the dashboard.