A teacher used to sense something in the middle of a student’s essay, such as a voice, a rhythm, or a tiny hint of personality seeping through the syntax. Finding that moment now is more difficult. It may not even exist. And for the researchers present at Stanford‘s fourth annual AI+Education Summit in February, that loss—quiet and nearly imperceptible—is at the heart of a much more significant reckoning.
On February 11, 2026, educators, technologists, policymakers, and researchers gathered for the summit, which was organized by the Stanford Institute for Human-Centered AI and the Stanford Accelerator for Learning. The discussions were incisive and occasionally unnerving. A celebration of AI’s introduction into classrooms did not emerge. It was more akin to a collective acknowledgement that the tools came more quickly than anyone had anticipated and that the presumptions that underpinned education were no longer valid.
Professor Mehran Sahami of Stanford’s School of Engineering put it this way. For many years, teachers believed that a strong final product, such as a well-written essay, a finished problem set, or a persuasive exam, indicated something about the process that led to it. It is likely that a student who turned in excellent work also did excellent thinking. AI disproved that notion. Now, students can produce something that appears to be deep learning without actually doing it. Sahami is not arguing that students are dishonest or indolent. He estimates that between 70 and 80 percent of them are using it to avoid the hard work rather than to support it because the system gave them a shortcut before anyone built guardrails.
The disparity in how this manifests itself based on a student’s school location is striking. AI, according to Teach for All founder Wendy Kopp, is an amplifier that reinforces the educational foundation that already exists. AI truly helps in schools with excellent instruction, well-defined objectives, and a well-thought-out structure. It primarily serves as a distraction in underfunded schools, where those foundations are weaker. That point was further emphasized by Miriam Rivera of Ulu Ventures, who pointed out the difference between students who learn to create with technology and those who merely consume it. Kids with plenty of resources can code and build. Others copy and scroll. Although that disparity is not new, AI is making it wider.
Guilherme Lichand, an assistant professor at Stanford’s Graduate School of Education, examined how AI affected middle school students’ creativity in Brazil, and his findings may have been the most subtly unsettling. The majority of people would not have predicted his outcomes. It should come as no surprise that students who had access to AI assistance did well. However, their advantage completely disappeared when the tool was removed in the middle of the test. More worrisomely, students who utilized AI before losing access did significantly worse on a subsequent task—four times worse than what their prior advantage would have predicted. They also said they didn’t enjoy their jobs as much. Some had begun to think that AI was just more imaginative than humans. That’s more than a decline in performance. A child’s self-perception is damaged by that.
That finding might be the best way to describe the problem. A helpful tool had subtly persuaded children that they were inadequate without it and that they needed it. Seeing that dynamic develop in research data seems like something that should be read slowly, several times.
Within this mess, there are individuals working to create something better. A teacher at North Star Academy in Newark, New Jersey named Mike Taubman created what he refers to as a “AI driver’s license”—a structured curriculum that frames AI literacy around the actual teenage milestone of learning to drive. Pupils gain knowledge about the engine, steering, and when not to apply the accelerator. It’s an honest and useful metaphor. Because many children are currently sitting in the passenger seat, allowing the machine to transport them without fully understanding their destination.
It’s still genuinely unclear if schools are moving quickly enough to catch up. Classrooms already have the equipment. Whether the thinking has followed is the question.
