Last year around this time, I made a bet with Alex Sarlin, the Global Edtech Lead of the venture capital firm ASU+GSV and co-host of Edtech Insiders, an edtech industry podcast.

Alex made this bet on LinkedIn:

I think by the end of [2025], it’s going to be hard to find an educator who doesn’t use Generative AI on a weekly basis.

I admired the specificity of that prediction, especially stacked against gauzy, non-falsifiable predictions like each of these, each one waving its hands and declaring 2026 “finally for real the year of the agent.”

I commented that I would take his bet and give him better than even odds. Not only wouldn’t we see anything close to 100% weekly usage, I said, but we wouldn’t see more than 80% AI usage of any type—casual, committed, or weekly.

Let’s check out both of those bets.

To Alex’s prediction of ~100% weekly usage, Gallup reported:

Only 32% of teachers report using AI at least weekly, while 28% use it infrequently and 40% still aren’t using it at all.”

Yikes! 32 is a lot less than 100, as far as numbers go!

Gallup also confirmed my prediction that fewer than 80% of educators would use AI in any capacity. You can triangulate Gallup with other sources too. Education Week has asked a national panel of teachers the same question every year for the last three years:

Which of the following best describes your current use of artificial intelligence-driven tools in your classroom?

The percentage of people who used AI at all in their 2025 survey (a little + some + a lot) was just 61%, almost identical to Gallup’s finding.

FWIW, Alex didn’t take me up on the bet I proposed. Nevertheless, I am bound by the Substacker’s oath to report to you that I would have soaked him.

What will happen in 2026?

AI use is certainly increasing among teachers, though that usage still seems confined to the dabblers in “a little” and “some.” The teachers using AI “a lot” are not exactly leaping off the axis. What kind of education technology is AI shaping into? High usage and high impact like slide software or an LMS? Low usage and low impact like the review games that teachers use every couple of weeks before a quiz?

I’m sticking with my predictions from early 2023 that AI will offer “quality-of-life improvements for teachers and students” but stop short of any “two sigma”-style changes to student learning or school staffing. Those predictions felt heterodox in 2023, but now I’d bet my house on any of them.

In 2026, you won’t see committed AI usage (“a lot”) rise above 10%. AI professional development will increase and you’ll see the dabblers swell in numbers—80% usage wouldn’t surprise me. But AI has not yet found its way to the core of teaching or learning. Chatbots and lessonslop still dominate the landscape. AI does lots of things faster and cheaper for teachers, but not enough things better. In 2026, you won’t find a room of educators where more than 10% of them would give up their slide software for MagicSchool. There are levels here.

What are the stakes of this bet?

More than I want to be right here (though I’ll take it) I want to be clear-eyed about the needs of teachers and learners, about the complex cognitive and social relationship they construct over the course of a school year, about the ways technology can and can’t support that relationship. The edtech industry is less sober than I have ever seen it, which turns your sobriety into a superpower.

My plan for edtech sobriety in 2025 involved substitute teaching monthly and helping a teacher out weekly. When 2025 rolled over to 2026, I immediately reset that goal. If you have perhaps misjudged the landscape of teaching and learning, if everything seems a bit blurry and you missed your predictions by 70%, I cannot recommend more highly that you shake your 2025 hangover by getting into a school. Try to teach something to someone who doesn’t really want to learn it, who doesn’t really think they can learn it. The experience will shape you and sharpen your senses and sober you up almost immediately.

Odds & Ends

¶ Eedi’s Chief Impact Officer Bibi Groot answered a question I raised in my review of their chatbot tutor study. Why did their chatbots help weeks later more than in the moment of tutoring? Groot speculates that the chatbot tutors drew more reasoning out of students than the human tutors, which supported a form of learning more amenable to transfer. I don’t know about that, but it sure is a thought! (I loved the graph above, indicating the human tutors eventually start cutting the chatbot tutors off, saying, “Hey—we’re done here.”)

¶ What innovation has re-energized this teacher?

“I feel like I’m coming home every day less exhausted, and I just have more energy to reinvest in the relationships with students,” Hesseltine said. “I’ve been teaching for over 20 years, so I really cannot stress this enough that this has made the biggest impact on education that I’ve ever seen.”

These are the kinds of results AI guys have been promising for three years now. So what is the teacher talking about and what stops us from getting it to more teachers?

¶ Just before the new year, Sal Khan wrote an op-ed in the New York Times considering employment in the age of AI. His solution to the problem of job loss is to pass a tin cup around to the companies profiting from AI and ask them to kindly donate 1% of their profits to programs dedicated to reskilling workers for AI fields. I can only imagine who Khan imagines would put those donations to work and lead the re-skilling effort.

The fund could be run by an independent nonprofit that would coordinate with corporations to ensure that the skills being developed are exactly what are needed. This is a big task, but it is doable; over the past 15 years, online learning platforms have shown that it can be done for academic learning, and many of the same principles apply for skill training.

In any case, the New York Times readers are absolutely scathing:

We have a mechanism for managing change equitably. It’s called the government. We don’t beg corporations for 1%. We tax them appropriately and use the tax funds to improve the lives of citizens and keep production and consumption within the limits that the biosphere has established.

¶ “Friction-maxxing” has entered the chat.

Friction-maxxing is not simply a matter of reducing your screen time, or whatever. It’s the process of building up tolerance for “inconvenience” (which is usually not inconvenience at all but just the vagaries of being a person living with other people in spaces that are impossible to completely control) — and then reaching even toward enjoyment. And then, it’s modeling this tolerance, followed by enjoyment and humor, for our kids.

¶ That article connected me to a recent post from Dylan Kane where he concluded:

That’s a tough reality of teaching: often the path of least resistance is the path of least learning.

1997

In 1997, education historian David Labaree observed that we love to fight about education and that our fights about pedagogy and policy really mask a fight about purpose. He identified several purposes for education, two of which have been locked in combat for the last several decades. From my perspective, we can confidently declare a winner and loser in 2025. Here’s Labaree:

From the democratic equality approach to schooling, one argues that a democratic society cannot persist unless it prepares all of its young with equal care to take on the full responsibilities of citizenship in a competent manner.

The social mobility approach to schooling argues that education is a commodity, the only purpose of which is to provide individual students with a competitive advantage in the struggle for desirable social position.

2025

The dominant education trends of 2025—including AI, personalized learning, increasing student absenteeism, education savings accounts, school choice, microschools, the kidsmaxxing of Alpha School and its partners on X—all indicate an American public that is giving up on democratic equality and is trying, instead, to maximize the social mobility of their children however they can.

Traditional public schools are one of the few remaining places where we gather very different people together in the same room to learn and be nice to one another. If you’re interested in fostering democratic equality, that work is extremely important and it isn’t cheap. Teaching students to learn math and learn to be nice to one another across their widening social, academic, and financial differences represents a huge challenge for teachers. Additionally, learning math and learning to be nice often takes more time than learning math alone, because learning more stuff generally takes more time than learning less stuff.

In 2025, we watched more and more parents say, “We really care a lot about the learning math part, but the costs of fostering democratic equality seem quite high.” Consequently, we watched more and more parents opt out of the relative heterogeneity of public schools and enroll their students into more homogeneous learning environments.

Not every parent group, though. Abigail Francis and Joshua Goodman studied enrollment rates in Massachusetts last year and found that enrollment declines in public schools were largely driven by wealthy parents. “The highest income 20% of districts have lost more public school students than the other 80% combined,” they found, “with these lower income districts having largely recovered.”

I think this is a dismal turn for our democracy, one that has been decades in the making, one that seems likely to continue in 2026. As the project of democratic equality provides fewer and fewer material benefits to people outside of schools—e.g. basic goods are very expensive; wealth inequality has only grown over the last 30 years; jobs are hard to find even for people who did all the things grown-ups told them to do like go to college and learn to code—people have, I think quite expectedly, prioritized that project less and less inside of schools.

Democratic equality has not lifted all the boats as promised. The ship of state is taking on water. People with the means to secure a lifeboat for their kids are doing exactly that.

2026

For 2026, as a citizen, I am most interested in projects that are committed to social democracy, committed to the truth that we have enough resources for everyone to have a lifeboat and more besides to fix the damage so none of them are necessary.

For 2026, as an educator and technologist, I am most interested in supporting democracy in schools, in working with teachers towards classrooms where every student feels valued, where every student feels like their ideas are needed, where every student feels like this class and its learning would have been less had they been absent today. I have several projects planned for 2026 and each of them will try to make it much easier than it is now for teachers and students to learn together.

Hey folks - this is where I write on special Wednesdays about edtech and math education. Today I’m going below the surface of some recent high-profile research that investigated how AI might offer students more help in math.

The Headline

The Reality

The Study

AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms

What They Did

The researchers embedded three kinds of support inside Eedi, a math software platform1.

• Support #1: Static pre-written hints.

• Support #2: A chat interface for accessing support from a human.

• Support #3: A chat interface for accessing support from an LLM, with a human approving the messages.

The study was nicely controlled. Every condition got tested against the other. Support #3 is the focus of the study. “In sessions with LearnLM, a supervising tutor reviewed each message that LearnLM drafted. They could either edit the message, completely re-write it, or approve it without any changes.”

Their outcome measures were whether students corrected a mistake, resolved a misconception, or transferred their knowledge. (See p. 9 for more precise descriptions of those outcome measures.)

Can I See a Picture?

What Happened?

Both of the dynamic supports clobbered the pre-written hints, as you’d probably expect.

There wasn’t much difference between “human alone” and “human + AI” on mistake remediation or misconception resolution. The big headline: the human + AI saw a 5.5% increase over a human alone on knowledge transfer. The study reports the odds that this is a legit finding at 1.3.

Finally, we directly compare the two tutoring conditions. We estimate that receiving support from LearnLM improved a student’s odds of success by a factor of 1.3 [0.9, 1.7] relative to human tutors, corresponding to an ATE of +5.5% [–1.4%, +12.4%]. Based on this posterior distribution, we find a strong probability (93.6%) that LearnLM elicited greater knowledge transfer than human tutors alone.

If, like me, you aren’t particularly familiar with the Bayesian methods here (the odds ratio) and if you are likewise concerned that the confidence interval contains 0 and the odds intervals contains 1, you can read this response from one of the study contributors, who seems to say, “Yeah, the odds aren’t 100% but they’re a lot better than chance.”

The study stored the original AI responses and the edits from human tutors (a fun source of data) which led to the finding that 76% of the LLM responses were accepted in full by the human tutor.

LearnLM proved to be a trustworthy source of pedagogical instruction, with the supervising tutors approving over 76% of its messages without changes or with only minimal edits (changing one or two characters; e.g., deleting an emoji).

Reasons to Be Happy

• It’s a responsible study. Humans supervising the LLM. Lower risk than just setting the LLMs onto the kids directly.

• “No differences” is still good. I think we can chalk something of a win for AI even when there isn’t a significant difference between the human tutor and human + LLM. The LLM operates under different constraints than a human. It can work faster and support more students simultaneously. The mere fact that there wasn’t a degradation in student learning is reason enough to be happy.

Reasons to Be Real

• It’s a small practical effect in a short intervention. The total time here was seven hours—one hour per week over seven weeks—for an increase of 5.5% on some proprietary measures. I would put this study in the basket of “point solutions that don’t congeal in any obvious way into a larger plan for school improvement.” Point solutions aren’t bad but we need to be real about the scope of the intervention here.

• This study is trying to do the wrong thing the right way. If the best tutoring interventions have kids a) working with trusted community members (parents, paraprofessionals, etc) b) in person for c) multiple times per week, d) doing work that is aligned to their curriculum, this intervention has none of those hallmarks. In general, getting tutored by a human through a text messaging interface shaves off a huge amount of the value of getting tutored by a human while also playing to a text chatbot’s strengths. It’s as if we asked Michael Bublé to play the kazoo and then supervise a robot purpose-built to play a kazoo. I’m not sure we’d be surprised by the results. I’m also not sure we’d feel good about saying “AI+Michael Bublé Match Quality Of Michael Bublé Alone.”

Reasons for Surprise

• It’s honestly quite strange to me that the Human + LLM showed its strengths with “Knowledge Transfer.” I would have expected to see the greatest difference on follow-up tasks that were closer to the tutoring moment. Wouldn’t you? Kids didn’t fix their answer on the next try (“mistake remediation”) or the try after that (“misconception resolution,” I guess) at different rates between the human and human+AI conditions. But apparently if you look beyond those moments, to a question in a different unit the next week, the effect of the AI really kicks in?

• This isn’t what’s commonly called “Knowledge Transfer.” Doing the next topic is doing the next topic. Knowledge transfer is typically understood as taking your learning on a topic and transferring it to a different context. i.e. “Okay you learned about slope in the context of an airplane landing. Can you calculate it in the context of the unit price at a grocery store?”

• I’m still surprised by this. If I were cynical, I’d wonder if the researchers found no significant differences on those first two much more intuitive measures and then went looking around at other measures, a frowned-upon process known as p-hacking. But thankfully I am not a cynic—a sweetie, in point of fact—and I will just process my cognitive disequilibrium on my own.

Implications

In these kinds of text-based tutoring environments, it seems possible AI could do some good or at least not do harm. Even if the LLM did fine in 76% of exchanges, though, I suspect there is some real frustration in the long tail of the distribution. For example, one human tutor said they had to overrule the LLM’s tendency to question students to death:

Tutors often found it necessary to step in when LearnLM’s Socratic questions, while pedagogically sound, persisted longer than a student’s patience. One tutor described a common scenario where “[LearnLM] will go, ‘Okay, you’ve got the answer. Let’s dig a little deeper about why you’ve got that answer.’ And the child is just like, ‘No, I’ve got it. I know what I’m doing. Can I go now?”’ (T1).

If the questions persist past their usefulness, I don’t think we can call them “pedagogically sound.”

Odds & Ends

¶ Over on LinkedIn, Shelbi Cole posted the image above and wrote:

The only reliable way to create strategic groups for Tiers 2 & 3 math instruction is to understand what children are doing.

The kid gets one out of six problems correct for a score of 17%. But “one out of six problems correct” is a pretty limited understanding of the child’s understanding and actively unhelpful for prescribing an intervention for the child. That’s a challenge for digital systems that reduce student thinking down to a binary correct / incorrect mark and that don’t make room for student handwriting.

¶ I was a guest on the last EdTechnical podcast of 2025. Libby Hills and Owen Henkel have a great vibe, know the space backward and forward, and keep a brisk conversation. We chatted about discovery learning v. explicit instruction and about AI Guys, how they’re made and how they limit their own effectiveness in edtech.

¶ Dylan Kane describes High Accountability Teaching in a post I liked quite a lot. Here’s the tease:

The core of high accountability teaching is to get as many of the swingers on board as possible.

Lately, I think a lot about “mattering” in classrooms. Do kids feel like they matter in their class? Do they feel like their ideas matter? Do they feel like the work they are doing matters? When kids complete work in class that a) isn’t seen by other kids, b) isn’t seen by grownups, c) isn’t commented on, d) doesn’t count for a grade, when they start to wonder, “Does this work actually matter to anyone?” that spins up a pretty negative cycle for learning. Dylan’s post added a lot to my thinking here.

¶ University of Washington freshman Luran Yang wrote a beautiful op-ed in the Seattle Times responding to the near-constant use of AI on her campus. “Not all hope is lost, however — I’ve been in one class, and only one class, with a near-zero rate of AI usage.”

¶ Education Week reports on Austin ISD’s use of artificial intelligence to support the annual creation of their master schedule, a task that historically requires lots of manual shuffling of teachers and classrooms and students.

… it saved [principals and their scheduling team] at least 50 hours. And the feedback that we got from those schedulers was that it increased their job satisfaction. They were so excited they got their summer back. They felt like they could actually manage master scheduling, and the master scheduling wasn’t managing them.

I don’t hold up “time savings” as a first-order goal for teachers, administrators, or schools. There are plenty of ways to save time that result in a worse education for kids. But I am much more optimistic that AI can support back-office operational tasks like this rather than the in-classroom relational work of teaching.

Hey there - this is where I write on special Wednesdays about edtech and math education. Today that means:

• Letting you know how the AI + education debate in Washington, D.C. went on Monday a/k/a 🤓 nerd fight night 🥊.

• Sharing two of the fantastic ideas for teaching slope you all shared in response to last week’s newsletter.

On Monday, four of us—

1. Shanika Hope, Google

2. Alex Kotran, AI Education Project

3. Jake Tawney, Great Hearts Academies

4. Me

—argued about the motion, “Maximizing School Improvement by 2035 Means Integrating AI into Classrooms Today,” in Washington, DC, at an event hosted by the American Enterprise Institute. We all had three minutes for our opening remarks. Here is what I prepared.

Look—I’m not made of stone. I love the motion and wish it were true, this idea that we made a massive technological breakthrough three years ago and now we can improve schools like never before. I wish it were true. There are a few reasons why I don’t think it is.

The first is the work we have asked schools to do in 2025. We have asked schools to help kids read, write, do math, and be nice to each other all at a time when hormones are flying around the room like pinballs, in a country that has some of the highest child poverty rates of the developed world, at a time when kids are walking to school scared that they or their friends or their loved ones are going to get jumped out on by some dude in a mask dragging them to God knows where. It’s a hard job at a hard time. I tried it last week. I tried to teach slope of a line to eighth graders in East Oakland, wonderful kids in the Fruitvale neighborhood. Ask me how well that went. Kids have a lot more on their mind than math right now.

The second is to understand the tight resources we have given schools to do that job. Financial resources are tight, obviously, but I’m talking also about time. We give schools 180 days to turn 3rd graders into 4th graders. We give schools four days for teachers to get out of class, get together, and learn something new. And perhaps most importantly, administrators have limited reserves of social capital. You can’t go to your school board, your parent community, ten times a year and say, “Okay, we’re going to try something new, something big, everybody get on board.” You can do that once, maybe. Opportunity costs are high. Saying yes to one thing means saying no to ten other things by definition. So that one thing needs to have absolutely bulletproof evidence that it’ll help your school help kids learn to read, write, do math, and be nice.

So my third reason is that in 2025, after three years of trying, the evidence that AI can help schools do that work is extremely weak. I know you have seen the same happy headlines I have or the profiles of schools doing something novel with AI. But scratch even a millimeter beneath the surface of those studies and stories and you’ll find some very weak evidence. You’ll find weak controls, where the control group gets nothing, as it should, and the experimental group gets AI, as it should. But they also get something extra—extra time, extra tutoring, extra rich parents. Elsewhere, you’ll find positive results for AI but they’re in a basket full of point solutions that don’t congeal in any obvious way into a school improvement plan. Still other studies run the other direction, like the MIT study this summer, where kids who used AI to support their essay writing were effectively de-skilled, experiencing diminished cognitive activity and unable to recall the arguments they’d made, compared to students who didn’t use AI.

For those reasons—hard job, limited resources, high opportunity costs, weak evidence in favor of AI, my recommendation is we wait. We take a beat. We take the next bus. We wait for stronger evidence and then move. For now, if you want to maximize school improvement, it’s an easy call to wait rather than integrate AI.

My comrade Jake Tawney argued that these tools would inhibit the formation of novice learners and novice teachers alike. In their arguments for the motion, Alex Kotran and Shanika Hope argued that:

• Kids will need AI skills for full US economic participation

• AI is being used outside of school, often poorly, and schools need to help kids learn to use it well

• AI will enable teachers to do better work than ever before, planning, differentiating, personalizing, preventing burnout, etc.

The debate was decided by a vote from the audience before and after the debate. The winner was whichever side most increased their share of the vote.

We won, which was fun of course, and the conversation kept everyone dancing, trying to stay on message, trying to answer the question we wished we’d been asked without seeming too obvious about it.

Jake and I counter-argued that many of the claims about the value AI offers teachers or students were dreams about the future rather than descriptions of the present. It is also a particularly unfortunate time to suggest that “we know what kids will need for economic participation in ten years,” given we are in the weakest labor market that software development has ever seen, ten years after telling every 12 year-old they needed to learn to code. That effort has gifted the private sector an oversupply of software developers who now have double the unemployment rate of art history majors.

If I had to guess what persuaded the audience to join us against the motion, it was our focus on “maximizing school improvement today.” In my closing argument, I said there were a bunch of motions I thought we’d all agree on, including the need for responsible experimentation, the need for dreaming expansively about what technology might offer kids. But maximizing school improvement today means taking seriously what a school most wants to improve, which is a kid’s ability to read, write, do math, and be nice, and applying the research we know works today. I suspect that approach gave some audience members permission to think AI is quite neat but also that 2025 is not the year that AI comes off the bench to maximize school improvement.

Mathematical Mailbag

I expressed difficulty last week keeping students thinking on a conceptual level about slope when the operations (“subtract, subtract, divide”) are so close at hand. Here are a couple of your very interesting responses.

Linda, offering an anchor activity that students can refer back to time and again as they try to make sense of new, related ideas:

1. I asked [students] if they realized that if you build stairs in your house that are too steep, it would be “illegal” because it could be dangerous. And that less steep stairs can be inefficient because they might take up too much space in a house.

2. I then gave each pair of students 12 identical Lego bricks.

3. I asked them to make 3 stair cases, using 4 bricks each, of different steepness.

4. After checking their stairs, I had them lay the 3 cases sideways on paper and had them trace it.

5. I then had them determine and record the rise and run of each of the stair cases. I checked those too.

6. Finally, they were to connect the top to the bottom and draw that line with a ruler. At that point, I asked them to record, “What do you notice about the lines you just drew and the slopes of the stairs?” We had conversations including the fact that some people’s stairs sloped downward to the right and others up. So, we also made that connection to positive and negative slope. Some noticed that larger slopes signified steeper stairs (and lines). In following days, when students were working on slope we would refer back to our “Lego stairs” to conceptualize what was going on.

Michael Hayashida, via email, illustrating how to help students understand the need for the measure known as slope:

Here’s something that seems to work in my freshman (somewhat remedial) math classes. All the kids have portable whiteboards.

“Draw me a set of axes and then draw a line that’s really steep.”

“Hold up the whiteboards - anyone have a line with the same steepness as someone else?”

“Ok, draw me a line that’s only medium steep.” They compare again.

“Ok, for mathematicians, this is a problem. They want precision - so that everyone draws a line with the same steepness when I give the instruction. To do that, you need to be able to put a number on steepness. Like... if I call for a line with a steepness of 5, then everyone draws a line with the same steepness. For the next minute, talk to your partner and see if you can come up with a way to measure steepness with a number.”

We share out. Sometimes kids will come up with an angle measurement method - kinda cool.

“Ok, this is actually a solved problem - mathematicians have already agreed on a way to measure steepness. Check this out - everyone put your marker on a lattice point - doesn’t matter which one. Now go over 1 and up 2 and put a point there. Do it again. Do it again. Connect the points. Hold them up - does everyone have lines with the same steepness? Ok, mathematicians call this a steepness of 2.

<Do the same thing, but with a slope of 0.5. Then 2.5.>

“This is how mathematicians decided to measure steepness - it’s what the y is changing by as x goes up by 1. They use a different word than steepness though - they call it slope”.

Hey—great to see you again. You’re looking well. I’m Dan Meyer, and I’m back in your inbox after an eight-month hiatus. I took a break to draft a manuscript and create some new technology for teaching math. I plan to share more about both projects soon.

In the past, I emailed you every Wednesday about whatever interested me most that week. My new plan is to email you only on Wednesday, and only when something has interested me about both a) math education, and b) edtech. You will never ever receive this newsletter on any day other than Wednesday and those Wednesdays might be months apart. That’s the Dan Meyer promise.

First, some plugs.

Here are a couple of time-sensitive announcements.

A debate about AI in education in Washington, D.C. Next Monday, December 8 at 6:30PM Eastern, I will be debating the motion “Maximizing School Improvement by 2035 Means Integrating AI into Classrooms Today.” I will be arguing against the motion (surprised??) with Jake Tawney, a leader in a classical school network. Jake and I will debate Alex Kotran, founder of the AI in Education Project, and Shanika Hope, who works on education initiatives at Google. The American Enterprise Institute is hosting the debate and I hope all of us will walk away a little smarter. You can attend in person or watch online.

A prediction contest about AI in education. The EdTechnical duo of Libby Hill and Owen Henkel are inviting you to put down some concrete predictions about AI and its future impact on education. There are five tracks with a top prize of $2,500 in each track. I am happily serving as a judge in the “Teaching Profession” track. We’re inviting predictions for the question: “By the end of 2028, what percentage of non-interpersonal teacher activities (lesson planning, grading, and parent communication) will teachers routinely delegate to AI systems?”

Yesterday’s humiliation.

Richard Rohr prays for one good humiliation daily to help him see the difference between how he thinks things are and how things really are. I beg people to let me take over their classes for roughly the same reason.

It’s dishonest work—building tools for teachers. Figma designers do their design work in Figma. Cursor devs develop software with Cursor. Google employees use the Google Suite. They all use their own tools regularly. They all regularly feel the same pain as their customers. For a few good reasons and a lot of bad ones, it’s hard to find edtech developers “dogfooding” their own tools1. This is why our team here at Amplify ritualizes classroom visits. This is why I beg.

I was teaching kids to calculate slope from two coordinates yesterday here in Oakland, CA. Operationally, this is very easy. Subtract, subtract, and divide. Conceptually, this is a topic that’ll bloom for at least the next five years. What does the slope mean? Why do we divide the vertical change by the horizontal? Where do the graph and the algebra connect? What happens when the horizontal change becomes very, very small? What is the slope of the slope?

The core challenge.

How do you “press” 30 kids to develop important conceptual knowledge when the easy operational knowledge is right there? I found myself doing a lot of desk work, trying to figure out what ideas students were bringing to the table, offering them questions to help them develop their intuition about slope.

• Okay which quadrant is this point in? Is this other point to the left or right? How do you know?

• Here are some points. Pick a couple of them you think would make a downward sloping line.

• How high did I have to climb from this point to this other one?

• Let me offer a wrong answer. How can we know it’s for sure wrong?

• Etc.

It would have hastened the whole interaction to direct them towards the easy operational knowledge. Subtract, subtract, and divide. I don’t know how to shortcut the conceptual work here. It’s possible to do some of it from the front of the class, but most of it requires serious, individual engagement with kids and their ideas. I don’t know how to do that across 30 kids. And when you turn that work over to a computer, it will happily encourage the kids to subtract, subtract, and divide.

I need new ideas for doing conceptual work at classroom scale. Let me know what you have on hand.

Odds & Ends

I have eight months of great links to catch you up on. Let’s start with five.

¶ I appreciate the skeptical posture Michael Pershan adopts towards just about everything. Recently, he directed it towards the overly tidy explanations our pundit class has offered for America’s declining test scores.

Here’s the situation: Americans are getting dumber...well, mostly not. But our lowest performing students seem to be losing ground. And, simultaneously, our adults. Some international tests show a similar decline happening in other countries. On other exams, America is on its own. What gives?

¶ Similar from Freddie deBoer: Was the United States Once a Global Leader in Educational Metrics? Have We Fallen From Those Lofty Heights? No and No

So here’s the nut of the whole thing: commentators often point to the recent decline in test scores (post-2020) as proof of a unique American failure. This ignores the global context: almost all developed nations have seen declining scores, especially in the context of the Covid-19 pandemic, and in fact in many countries these declines have been more pronounced than in the United States.

¶ Justin Reich’s podcast The Homework Machine concluded its run during my hiatus. It was extremely well-reported. Great interviews. Great perspectives on teaching, learning, and technology. In its final episode, a San Francisco English teacher named Sara Falls is asked why AI isn’t really helping her and she proceeds to describe the core challenge of teaching in a way that still haunts me months later. You should listen to her tone in that moment, but if you must, you can read what she says:

… this kid who just brought me this candy cane … this sweet kid … I’ve struggled … the fact that she brought me a candy cane is a big deal because at the beginning of the year she hated me. You know … and I’ve stuck with it … I’ve stuck with her and she just brought me a candy cane. This is the work.

¶ While I was out, OpenAI, Anthropic, and Google, all released “study mode” versions of their chatbots. These new versions won’t just throw you the answers when you ask. They’ll make you work for those answers. (The marshmallow test goes to college!) I enjoyed reading Mashable’s Chase DiBenedetto’s reports from test-driving each one, especially how the directive to not give the answer ever can drive a student to madness:

But here’s the unsettling part of using Claude: As conversations continued, it increasingly felt like I was trapped, in the sense that lessons seemed to go on forever. Because Claude wants to really make sure it knows where your head is at, the bot doesn’t provide you a clear stopping point for the lesson, sometimes making conversations hard to follow, and the learning experience frustrating. During our math session, Claude fervently lauded me when I got the answer right, but sent me approximately 100 follow-up questions. Even when I asked if we were done with the problem — AKA can I go home now? — it gave me more work to do.

¶ Related: I loved reading RAND’s survey of students and their use of AI and I don’t think I saw anyone report on it. One interesting question asked students “so what tools are your teachers recommending?” and then “what are you actually using?” I went ahead and plotted both sets of data simultaneously and you can get a preview of the popularity of these study modes. The max teacher recommendation was for Khanmigo—a Socratic tool. Kids went ahead and used vanilla ChatGPT anyway.