AI in the Classroom, Part 3: Where It Actually Earns Its Place | Ben Forta

This is Part 3 of a 3-part series. Here are links to Post 1 and Post 2.

I've spent the last two posts on what's actually happening in classrooms and on the fears worth taking seriously. This one is about the part that actually gets me excited: what happens when this stuff is used well.

Individualized Instruction, Finally Attainable

Personalized, one-on-one instruction has been the dream of education forever, and rightfully so. Ask any teacher what they'd do with unlimited time and a class of one, and they'll describe exactly what tutoring provides: pacing that matches the student, feedback the moment it's needed, meeting a kid exactly where they are instead of where the syllabus says they should be. That's not a novel idea anyone needed convincing of. It's the thing teachers have wanted since the day they took the job, and never had the staffing, the time, the budget, or the autonomy to actually deliver.

In 1984, researcher Benjamin Bloom put a number on that gap: students who got individual tutoring performed two standard deviations better than students taught in a normal classroom, a gap so large it became known as the two-sigma problem, and for forty years it stayed exactly that: a problem, because there was never going to be a tutor for every kid. Not enough tutors exist, and even if they did, no school system could afford them.

That math is starting to change. A 2026 randomized trial in UK classrooms found students working with an AI tutor achieved median learning gains more than double the group taught in-class, with higher engagement to go with it. I want to be honest about the caveats here, because I spent the last post arguing against overclaiming. Khan Academy's own Khanmigo doesn't yet have that kind of independent proof behind it. Sal Khan has said the early version undersold what everyone hoped for, and the first real randomized trial of Khanmigo specifically isn't due until later this year.

To be clear, AI in and of itself does not solve the two-sigma problem, And no, we won't (or shouldn't) be seeing AI or robot tutors any time soon. But AI can free up teacher time, it can throw up the Bat Signal to spotlight areas of concern, it can direct attention where needed, and it can help craft custom corrective action plans. AI does not solve the problem, but it absolutely can assist educators in doing so.

The direction is clear even where the specific product isn't proven yet: one-on-one, adaptive instruction at a cost that scales to every kid, not just the ones whose parents can afford a tutor, has gone from a fantasy to an active engineering problem. That's an enormous shift, and it happened in about three years.

Removing the Bottleneck Between Idea and Output

The benefit I actually care about most isn't the test-score version of personalization, though. It's what happens when AI removes the gap between having an idea and being able to get it out of your head.

I've made this argument before, about coding specifically, in Does Vibe Coding Change My Answer?:

No more blank canvas? No more writer's block? No more "I want to build something but I have no idea where to start"? Good! That's the friction that's been killing kids' enthusiasm for decades.

That's not just a coding argument. It's the argument, full stop, and it shows up everywhere AI touches the classroom.

This one is personal. My handwriting is so bad that I've been asked, more than once, if I'm a doctor. My spelling isn't much better. I struggled to memorize spelling lists as a kid, and never really got good at it. Once upon a time, those would have been real handicaps. Instead, I've published forty books with over a million copies in print, because a word processor and a spellchecker meant the mechanics of getting words onto a page stopped being the bottleneck between me and the actual writing. That happened well before AI. AI is just the newest, much more powerful version of the same trade: separate the idea from the mechanics of producing it, and let the idea win.

I saw the same thing recently, in miniature, from a room full of first graders. They were showing me what they had created in Adobe Express, and one young man proudly showed me the slideshow he'd made about reptiles: slide after slide, images he'd found or generated himself, put together with a level of polish a six-year-old had no business having access to twenty years ago. What struck me wasn't the content. It was the pride. This kid wouldn't let me move on to engage his classmate, he wanted to show me one more slide, because for the first time, what was in his head and what he could actually produce were the same thing. And it wasn't just this one young reptile fan; I was treated to presentations, documents, and websites on puppies, family vacations, sloths, pandas in tutus (thank goodness for generative AI), and yes, more puppies. That kind of confidence compounds. Kids who feel like...