By Steve Grubbs

When the Brookings Institution released its January 2026 report, “What the Research Shows About Generative AI in Tutoring,” it posed a question that I believe will define the next decade of learning technology: under what conditions does AI actually improve learning—and when does it risk undermining it? That question matters because education is not merely an efficiency challenge; it is a deeply human endeavor. If AI tutors are to play a meaningful role in K–12 and higher education, they must strengthen learning without weakening the essential roles of teachers, sound pedagogy, and productive student struggle.

Constructive struggle honors learning’s elegance: students wrestle with ideas, make mistakes, reflect, and grow—building understanding that lasts because knowledge is earned, not handed over. I can remember sitting in my pencil-scratched desk at Frank L. Smart in Davenport, Iowa struggling with an essay on The Most Dangerous Game, trying to come up with something clever that has not already been said by Ms. Zamora. This was where true learning was earned.

Brookings did not begin with hype, nor did it assume AI was inherently good or bad. Instead, the researchers focused on evidence from tutoring research, cognitive science, and early deployments of generative AI. Their central concern was not whether AI can teach, but whether it can do so responsibly—without replacing human judgment, short-circuiting learning, or creating new equity gaps. In short, they asked whether AI tutoring could scale the benefits of great instruction while preserving human oversight.

A quick look at the most recent results from NAEP and it should be concerning for any educator: test scores in reading and math have hit historic lows. You can quibble all you want about the value of standardized testing, but this is a strong indication that students are learning less.

The findings from Brookings are refreshingly pragmatic. First, the research confirms what decades of tutoring studies already show: individualized tutoring is one of the most effective instructional interventions ever studied. When learners receive timely feedback, tailored explanations, and opportunities to practice at their own pace, learning accelerates dramatically. Generative AI, Brookings notes, finally makes it possible to deliver some of those benefits at scale.

“AI tutors are most powerful not as replacements, but as force multipliers. They give every student access to patience, repetition, and personalized explanation—things even the best teachers struggle to provide at scale.”

But the report is equally clear about the risks. Brookings warns against fully autonomous AI tutors that operate without guardrails, transparency, or educator involvement. Left unchecked, AI can provide answers instead of guidance, optimize for fluency instead of understanding, and unintentionally reinforce misconceptions. The researchers emphasize that learning depends on productive struggle—the process of wrestling with ideas, making mistakes, and refining understanding. Any AI system that removes that struggle may feel helpful in the short term while harming learning in the long run.

This is where human oversight becomes non-negotiable. Brookings consistently returns to a human-in-the-loop model: AI should assist instruction, not replace educators; it should adapt to learners, but within pedagogical frameworks designed and monitored by humans. Teachers, instructional designers, and institutions must retain control over goals, content, and evaluation. AI works best when it is embedded inside good teaching—not when it tries to become the teacher.

That framework aligns directly with how VictoryXR designed HoloTutor from day one. We did not set out to build an AI that replaces teachers. We set out to build an AI that scales great teaching. HoloTutor is intentionally structured around existing curriculum, textbooks, and instructional intent. It does not hand students answers. Instead, it teaches through analogies, Socratic questioning, and step-by-step guidance—preserving the elegant struggle that leads to mastery.

Equally important, HoloTutor is built for institutional oversight. Educators decide what content is taught, how it is framed, and how progress is monitored. Students receive personalized support, but within boundaries set by teachers and schools. That is precisely the hybrid model Brookings argues is most likely to improve outcomes while minimizing risk. In that sense, Brookings didn’t validate a specific product—but it validated an approach. And that approach is the one we’ve committed to.

The report also reinforces something I’ve seen repeatedly in practice: AI tutors are most powerful not as replacements, but as force multipliers. They give every student access to patience, repetition, and personalized explanation—things even the best teachers struggle to provide at scale. At the same time, they free educators to focus on higher-order instruction, mentoring, and human connection. That balance is where real transformation happens.

So what’s next for AI in education after the Brookings report? Expect clearer rules around transparency, privacy, and human oversight, alongside a shift from generic chatbots to purpose-built, pedagogically grounded instructional systems. The debate is moving from whether AI belongs in education to how well it is designed and governed. Brookings offers a balanced roadmap: AI can expand access to high-quality tutoring, but only when combined thoughtfully with human educators to preserve how real learning happens.

Steve Grubbs is CEO of VictoryXR, pioneering immersive learning with AI and virtual reality to expand access, accelerate mastery, and scale great teaching worldwide responsibly.