GM. This is The Great Unlock, the newsletter for people who build, buy, and run K-12 learning and refuse to pretend the machine is fine.

Today's deep dive: Ray Kurzweil says every information technology gets better on an exponential curve. I've been reading The Singularity Is Nearer, and one idea kept snagging on the book I've spent three years writing. It publishes tomorrow. So this week is personal.

Let’s unlock it.

Kurzweil is known for the Law of Accelerated Returns. It says that once a domain becomes information, progress stops being linear. Each generation of the technology gets used to build the next one, so the curve bends upward instead of climbing a straight line. The printing press did it. The internet did it. Chips did it. Genome sequencing did it. Search did it.

So, the more I’ve been reading The Singularity is Nearer, the more I’ve been wondering if LOAP is going to drive hockey stick innovation in K12 education. My answer is yes but it’s going to hit a roadblock: regulations.

Every interaction a student has with an AI tutor is data. Every wrong answer, every hint pulled, every three-second hesitation before a student picks the wrong denominator. In The Great Unlock I call the engine that reads those signals Root Cause Analysis, and its whole job is to trace a surface failure back to the buried gap that caused it. A kid fails Algebra 1. The real problem is fractions back from Grade 5.

That data doesn't help one student and then evaporate. It sharpens the diagnosis for every future student who trips on the same root. The system that takes six days and a human to spot a fractions gap this year takes one second next year, because it has seen a thousand more kids fail the same way.

More usage, more granular data on where learning breaks, better diagnosis, better outcomes, more adoption, more usage.

That's a flywheel. It's the same loop that made internet search engines improve with scale. And it's a real, defensible reason to believe AI can make a mastery-based classroom get better every year instead of staying flat. I believe it. I built HierLearning on the basis of it.

What I’m testing is where Kurzweil's curve holds inside a school, and where it slams into a wall he never had to think about. Both things are true at the same time.

The way I see, here’s the wall:

The bottleneck in American education has almost never been the algorithm. It's the law.

In The Great Unlock, there's a scene with a State Director of Education named Elena. She's looking at a list of her own state's statutes, and one of them is Statute 14.B: a math credit requires 120 hours of seat time. 120 hours of sitting in a chair. The clock counts attendance, and attendance is the only thing it counts. A student can master the material in 40 hours and the statute still says she owes the chair another 80.

Kurzweil's curve does nothing to that number. The diagnostic model can get 100 times sharper and Statute 14.B doesn't move a millimeter. Licensure rules don't improve because the AI improved. Funding formulas that pay per seat, per day, don't recalculate themselves because a tutor got smarter. The software races up an exponential curve while the system it lives inside sits on a flat line drawn in 1893.

I named those constraints the Three Locks: personalization, motivation, and time. Every one of them is structural and legal before it's ever technical. That's the part Kurzweil's law can't touch. You can have the fastest engine ever built and a parking brake welded on by statute.

The clarity on the problem is nearly universal. The path is the only thing missing. A faster algorithm is not a path.

Not everything in a classroom sits on the same curve, and this is what I want to stress test with district leaders and founders.

Competence and pacing are computational problems. Give the model more data on how kids fail at fractions and it genuinely gets better at teaching fractions. That curve bends up, exactly like Kurzweil says.

Motivation and relatedness are a completely different animal. Whether a kid feels like a whole person instead of a data stream does not obviously improve because the model has now seen a million students. In the book I put a rough ratio on it: motivation is about 90% of the outcome, and the tools, the platform, the AI tutor, all of it, is the other 10%. Trust has a flat curve. You can't compound your way to a child caring.

Next, there's a second flattening worth point out. A lot of the gain in a Studio Model (discussed in The Great Unlock) comes from a one-time structural change. Closing the feedback loop from six days to one second is a single step up. Real, large, and finite. Cutting seat time frees the calendar once. Those are enormous wins. They're also finite, and I won't pretend the classroom rides Kurzweil's curve forever when a chunk of the value is a single door swinging open.

So my honest take on it is that the diagnosis compounds, the architecture is a step change, and the human part stays stubbornly, wonderfully flat.

If you lead a district, or you're building the tool a district buys, these are the four I'd ask before you believe anyone's acceleration story:

Sources: Ray Kurzweil, The Singularity Is Nearer (2024) · Benjamin Bloom, "The 2 Sigma Problem" (1984) · Karim Kuperhause, The Great Unlock (2026).

The claim: the more students use it, the more the good version spreads to everyone.

It's a lovely story. It's also backwards about how flywheels behave.

Data flywheels concentrate power toward the center. Every turn pulls advantage inward. The platform with the most students collects the most failure data, which builds the sharpest diagnostic model, which wins the most students, which widens the lead. That's the physics of the loop working exactly as designed.

Play it forward and you don't get great teaching for everyone. You get one or two vendors with a model nobody can catch, and a new version of the oldest inequity in education: the good stuff goes to whoever signed the contract first. Last time it was which family could afford a tutor. This time it's which district picked the winning vendor in 2026. Same gap, new logo.

A flywheel is a moat with better marketing. If you want great teaching distributed, you have to design for that on purpose, because the loop on its own pulls the other way.

Elena had spent 30 years believing in the system, and she had the office to prove it. State Director of Education. The person who could actually change a rule.

The pilot data was sitting in front of her and it was not ambiguous. The pilot students had mastered the math. Faster, deeper, with the gaps closed instead of papered over. The AI had done exactly what the vendor promised. The flywheel had spun. The model was sharper than anything her state had ever run.

Then she looked at the other list. Her own statutes.

Statute 14.B: a math credit requires 120 hours of seat time. Her studio students had earned the math in 40. On paper, in the eyes of the state she ran, they hadn't finished. The best diagnostic engine in the country had produced kids who were, legally, incomplete.

She felt the two plates grind against each other. On one side, a technology climbing a curve Kurzweil would recognize. On the other, a number written into law in an era of factory bells, sitting there, immovable, waiting for the chair time it was owed.

The engine had done its job. The problem sat in Elena's hands, printed on state letterhead.

The whole point of The Great Unlock is that leaders like Elena are the ones who open the last lock. The software can't do it. The statute can't do it. A person with the authority to change the rule does it.

 ________________________________________________________

It's out tomorrow, Monday, August 3rd, on Amazon.

Fourteen years in a classroom, a decade building EdTech, one argument I couldn't stop making at dinner parties: the K-12 system isn't broken, it's obsolete, built in the 1890s for a world that needed factory workers, and AI is the first real tool we've had to build something better instead of patching the old machine.

If you've been meaning to grab it, tomorrow is the day that counts. Amazon ranks by the hour, and sales concentrated on launch day carry more weight than the same copies spread across a month. I'll send the link Monday morning, before it goes anywhere else.

That's all the unlock for today. Tune in next week.

Stay awesome, you unlockers!