Beyond the specifics of edtech and teaching, I think the incubator chant of "does it/make it scale" has escaped into political arenas where it does some real harm. It got its cultural currency as a demand made of people making websites by the people lending them money, in an economic moment where the expectations of returns and the timescales to receive them more closely resemble mob bust-outs than any sort of infrastructure building. Websites are uniquely suited to blow up easily. They also routinely don't actually matter that much. The idea's escape into the world was mediated by people that ultimately hoped they could convince people that problems like poverty, crumbling infrastructure, and subpar educational outcomes could be solved by their expertise- making websites make money- and not by, say, taxing their website money to pay people with skillsets they openly denigrate. They are clearly both wrong and self- interested.
Everything scales if you're brave enough. There are about 75 million kids in the US- a little power-power-of-ten envelope math suggests that doing the most obvious things to maximize their learning- putting them in classes small enough to get personal attention (say, 10) and paying teachers enough to attract, develop, and retain competitive talent (say, $100K)- would cost about as much as the US's infamously excessive defense budget. Just as a thought experiment...
Over in the much-smaller-than-math niche of ancient language learning, we’re trying to think through similar issues. The legendary Latin and Greek instructor WHD Rouse said that the classroom is the worst place to learn a language, because you can’t talk about anything concrete in the classroom besides desks, chalk, and chairs.
But like you say, classrooms are nonetheless the places where people who want to teach meet the people who want to learn. So Rouse used the tech of his day (over 100 years ago) to ensure students got to see lots of images, so they could talk about them and “get outside” the classroom in order to practice using their target language.
I've been working on a chemistry ed-tech project for years, initially based on what I thought would be useful for my students when I taught community college and private school, and now also inspired by the way I wish we taught science. My project Chemiatria (https://chemiatria.org/) is mostly aimed at older and/or more independent students, but here's why I think it's human-centered in the way you mean:
It is carefully designed and written by a human for other humans, without shortcuts. Unlike many ed-tech offerings, it's a tight integration of content and code: if I want an interactive tool to teach "railroad tracks" for dimensional analysis, or to analyze and give useful feedback on a student's reaction equation, I build it. I try to be thoughtful about human cognition, encouraging the right kinds of mental engagement and friction, while minimizing the wrong kinds.
I want to honor students' curiosity. For example, my introduction to units and dimensional analysis tells a loose story about plastics: how much we make, what they're good for, how much we eat accidentally, and how we try to estimate whether the chemicals we're exposed to are bad for us. Along the way there are two fairly detailed, disturbing case studies of how the scientific method and peer review work in practice.
Several of the activities that introduce content are interactive "books" that tell a historical story: who discovered the facts or articulated the concepts and how, and what it all meant to people at the time, practically and philosophically. I hope that opening up big, eternal questions and showing why great thinkers of the past were passionate about science can be part of a solution to the crisis of meaning and pointlessness in education today.
Also, Chemiatria is free. With no army of salespeople pushing it, any student or teacher can try it on a whim and stop if it doesn't work for them (though I hope first they'll give me feedback so I can make it more useful!). I hope students who use it will understand that I built it for them, not to make money but because they (and their knowledge of the world) matter to someone.
Thanks, Dan, this is a really helpful and encouraging framing.
I've been looking at a related question in my PhD research: what changes in people's interest and curiosity in learning when another person is present vs. with just an AI? In a recent study, we found differences in both effort and intrinsic interest, even when the content itself was held constant.
The problem was never the technology. It was the sequencing: solution first, problem second. EdTech spent years convincing teachers that instruction is really just content delivery with better packaging, and the Khanmigo data is the natural endpoint of that assumption.
Katelynn Petersen’s 40/40/20 breakdown should be required reading for every product team in this space. Strip that out and you haven’t built a better tutor; you’ve built something else entirely.
The one layer I’d add: these misalignments are also leadership failures. Districts adopt tools without asking who owns the outcome when performance falls short. That ambiguity is what allows mandated usage and 15% engagement to coexist without anyone asking hard questions. Technology that serves humanity requires leaders willing to be accountable for the decisions they make around it. Would love to connect!
As a math specialist at a prek 5 school (now retired) I needed to design what I did to benefit as many teachers as I could. The first thing that needed to happen was to work out a math schedule for all the teachers where not all were teaching math at the same time. To do this I needed admin support. The next thing I needed was to meet weekly with every head teacher in the school. We did this by figuring out grade level meeting times and keeping to the time we had allotted. Then I needed to make my schedule one that would meet teacher needs. So that meant I rarely spent time in my "office" which actually was a desk in the lunch room where I was ,If there, available to anyone who needed me.
Finally I worked out a plan to work with staff. That at first took the form..new teachers or needy teachers first. Teachers I worked with were on a 6 week schedule. The first week I watched the teacher teach, took lots of notes and developed a plan for going forward. (It is important that anything I saw in a classroom stayed between me and the classroom teacher. I was not an evaluator) The second week I taught the class and the teacher watched and took lots of notes. The third week we taught together backing each other up. (all this time I wrote all the lesson plans). The fourth week with me still writing the lesson plans the teacher taught on her own knowing I would back her up any time still using my lesson plans. The fifth week the teacher wrote her own lesson plans and did her own teaching. Same with the 6th week as I might come and go and not be there all the time. To meet and talk about what we did we often tried to do it at recess with help for coverage from the staff. In the beginning we might meet every day but as time passed maybe 3 times a week. This was in addition to grade level meetings.
I usually did 3 teachers at a time. I know it sound hard but not impossible. I also know it worked. I had teachers who thought they could not teach math in the beginning become my best math teachers and the children loved math time.
100% - we should be using technology to support human connection, not intermediate it.
In terms of sharing what folks are working on, this is the thesis of Once (https://www.tryonce.com/) - use edtech to support grown-ups teaching kids to read, both to help parents at home who want to play a more active role in their kids education and to help schools train and coach paraprofessionals to deliver instruction. And you’re right about the need for better coaching - we do weekly 30 minute 1:1s with every instructor where we walk through recordings of their instruction, and people regularly share this is the best and most robust professional support they’ve ever received.
And as a result, we see very strong student results. Leaning into humanity may not scale as fast as rolling out a chatbot, but it works, and there are ways to use technology to expand access to excellent instruction to more kids and schools.
Just on the wacky politics, it's not just in education -- anti-AI, anti-Big Tech sentiments are forging new bipartisan coalitions to oppose the hysteria around building data centers. Americans are looking at the fractured country that Silicon Valley has played an outsize role in shaping and they are starting to revolt. I agree that some ed-tech companies will survive, but something quite powerful is happening politically.
The "infinite patience" piece feels akin to another misalignment I always notice in the design of AI Tutors in that they are designed/prompted to "never give the answer", which again portrays that the designers don't actually understand the relational dynamic system they're trying to mimic.
Would also love to hear more on your thoughts on coaching since EDSI is doing work on that.
I’m very interested in working on the caregiver support misalignment problem! The way I’m currently thinking about it is through creating undeniable proof points in the areas the caregiver cares about so they can see it. Part of this vision is removing the separation between work and school, collocating them and integrating projects. Would love to hear from others interested in this “misalignment” space!
I describe a "plan B" by addressing the issue that the most profitable question in EdTech (can anyone build a synthetic educator) is not the most important one (how can we reduce the stress, effort, and fragmentation that sits between students and teachers every day).
AI is coming for the classroom, the question is whether it further fragments the system or makes it more coherent.
Hey Dan! How's it going? Guess what -- I'm the guy you quoted who asked "What's Plan B"? I'm also the guy who pointed out that many tutoring programs don't seem to scale very well. And I'm posting to say that you have completely misinterpreted what I wrote.
First of all, I'm not "grieving" over Khanmigo. I'm mystified by how you came up with that, but you are simply delusional. I'm not an AI advocate, I don't use LLMs, and I am deeply skeptical about the introduction of AI into the classroom. I'm also not a big booster of Khan Academy, although I think it can be a valuable resource for some students. So I can hardly be considered to be any sort of advocate for AI in education, or for Sal Khan and his crazy ideas. But I'm not **actively pleased** to see that Khanmigo has not lived up to its promise. This was the weird thing about your original post -- there were so many people who were explicitly cheering its perceived failure and gloating over this. And that's just wrong! There are lots and lots of educational initiatives that I think are misguided or ineffective, but I don't openly express active delight when they don't work out. That's not "grieving", it's just common decency.
It's also bizarre that you can't understand that educational initiatives have to be able to scale. Yes! Even an individual school with a few hundred students has to contend with this. Tutors have to be recruited, trained, and integrated into the curriculum, on a continual, on-going basis. This is possibly doable for a small number of tutors, but for a project like this to be feasible you need a lot more than a few dedicated people; to make this work at the level of a school district is very, very hard. You think that this is a "strange standard for evaluating interventions in education", but it's actually one of the most important. Tell me -- if it's difficult to scale an intervention from a small pilot study up to an entire school district, why wouldn't that be an important consideration? If there are major obstacles to implementing a proposal, isn't that an argument against the proposal?
I realize that you and I have different perspectives, and you are welcome to criticize my viewpoint. But it would be helpful if you could accurately characterize my position. And spare me the condescending "grieving" bullshit.
"if it's difficult to scale an intervention from a small pilot study up to an entire school district, why wouldn't that be an important consideration?"
Can you name anything worthwhile we could do in education that ISN'T difficult? Isn't that just a given? Yes, it's difficult, everything is, now let's get on to the doing of it.
Hey Dan! I'm an Educational Technology Coach at Classmate — we work with K-7 teachers across a bunch of schools in BC, Canada, visiting classrooms a few times a week (https://classmate.team/chronicles/).
Most of what we do is try to move tech away from consumption and toward making. Podcasting, board game design, science project documentation, digital journals where kids tell the story of what they built and why. We lean heavily on recording student voice with audio and video.
Your point on coaching misalignment made reflect. What we've found is that the most useful version of our support isn't a model lesson or a walkthrough, it's sitting with a teacher on something they're already building and iterating on it with them. When we look to embed tech into their lesson, it tends to open other doors too: a natural spot for a reflection check-in, a way to make student thinking visible to the whole class, an opportunity for cross-curricular assessment. None of that happens in a one-off visit, it's often after a few conversations and time spent in the classroom with the teacher trying things out. Tech tends to be a trojan horse to better pedagogy.
Second comment since it's quite separate from the first.
Stephen Downes (Ed Tech researcher at National Research Council in Canada) says and publishes interesting things on this exact subject. He has been banging on the "Learning should be social and human-connected" drum for decades. His theory/framework of "connectivism" is really interesting to me. https://files.eric.ed.gov/fulltext/EJ1350294.pdf
"If that’s you, let me know what you’re working on in the comments."
This post outlines a successful edtech-math marriage with a human-first focus. If you are skimming, the news article from Blackfoot (paragraph 5, "read about the incredible impact right here") is particularly inspiring.
Beyond the specifics of edtech and teaching, I think the incubator chant of "does it/make it scale" has escaped into political arenas where it does some real harm. It got its cultural currency as a demand made of people making websites by the people lending them money, in an economic moment where the expectations of returns and the timescales to receive them more closely resemble mob bust-outs than any sort of infrastructure building. Websites are uniquely suited to blow up easily. They also routinely don't actually matter that much. The idea's escape into the world was mediated by people that ultimately hoped they could convince people that problems like poverty, crumbling infrastructure, and subpar educational outcomes could be solved by their expertise- making websites make money- and not by, say, taxing their website money to pay people with skillsets they openly denigrate. They are clearly both wrong and self- interested.
Everything scales if you're brave enough. There are about 75 million kids in the US- a little power-power-of-ten envelope math suggests that doing the most obvious things to maximize their learning- putting them in classes small enough to get personal attention (say, 10) and paying teachers enough to attract, develop, and retain competitive talent (say, $100K)- would cost about as much as the US's infamously excessive defense budget. Just as a thought experiment...
Yo!
I’m dabbling in the parent / caretaker support sector. It’s a first iteration to get started.
https://www.whenmathhappens.com/parents
Trying to not only support parents with math but also strengthening relationships with their students.
This is great, Dan. Edtech has to be for humans.
Over in the much-smaller-than-math niche of ancient language learning, we’re trying to think through similar issues. The legendary Latin and Greek instructor WHD Rouse said that the classroom is the worst place to learn a language, because you can’t talk about anything concrete in the classroom besides desks, chalk, and chairs.
But like you say, classrooms are nonetheless the places where people who want to teach meet the people who want to learn. So Rouse used the tech of his day (over 100 years ago) to ensure students got to see lots of images, so they could talk about them and “get outside” the classroom in order to practice using their target language.
His pedagogy was very innovative in a number of ways, despite him being personally quite grumpy about technology. His essay on the topic is great fun: https://antigonejournal.com/2024/11/machines-or-mind-loebs-rouse/
I've been working on a chemistry ed-tech project for years, initially based on what I thought would be useful for my students when I taught community college and private school, and now also inspired by the way I wish we taught science. My project Chemiatria (https://chemiatria.org/) is mostly aimed at older and/or more independent students, but here's why I think it's human-centered in the way you mean:
It is carefully designed and written by a human for other humans, without shortcuts. Unlike many ed-tech offerings, it's a tight integration of content and code: if I want an interactive tool to teach "railroad tracks" for dimensional analysis, or to analyze and give useful feedback on a student's reaction equation, I build it. I try to be thoughtful about human cognition, encouraging the right kinds of mental engagement and friction, while minimizing the wrong kinds.
I want to honor students' curiosity. For example, my introduction to units and dimensional analysis tells a loose story about plastics: how much we make, what they're good for, how much we eat accidentally, and how we try to estimate whether the chemicals we're exposed to are bad for us. Along the way there are two fairly detailed, disturbing case studies of how the scientific method and peer review work in practice.
Several of the activities that introduce content are interactive "books" that tell a historical story: who discovered the facts or articulated the concepts and how, and what it all meant to people at the time, practically and philosophically. I hope that opening up big, eternal questions and showing why great thinkers of the past were passionate about science can be part of a solution to the crisis of meaning and pointlessness in education today.
Also, Chemiatria is free. With no army of salespeople pushing it, any student or teacher can try it on a whim and stop if it doesn't work for them (though I hope first they'll give me feedback so I can make it more useful!). I hope students who use it will understand that I built it for them, not to make money but because they (and their knowledge of the world) matter to someone.
Thanks, Dan, this is a really helpful and encouraging framing.
I've been looking at a related question in my PhD research: what changes in people's interest and curiosity in learning when another person is present vs. with just an AI? In a recent study, we found differences in both effort and intrinsic interest, even when the content itself was held constant.
I wrote a bit about it here: https://caitlinmaking.substack.com/p/the-motivation-ceiling
The problem was never the technology. It was the sequencing: solution first, problem second. EdTech spent years convincing teachers that instruction is really just content delivery with better packaging, and the Khanmigo data is the natural endpoint of that assumption.
Katelynn Petersen’s 40/40/20 breakdown should be required reading for every product team in this space. Strip that out and you haven’t built a better tutor; you’ve built something else entirely.
The one layer I’d add: these misalignments are also leadership failures. Districts adopt tools without asking who owns the outcome when performance falls short. That ambiguity is what allows mandated usage and 15% engagement to coexist without anyone asking hard questions. Technology that serves humanity requires leaders willing to be accountable for the decisions they make around it. Would love to connect!
As a math specialist at a prek 5 school (now retired) I needed to design what I did to benefit as many teachers as I could. The first thing that needed to happen was to work out a math schedule for all the teachers where not all were teaching math at the same time. To do this I needed admin support. The next thing I needed was to meet weekly with every head teacher in the school. We did this by figuring out grade level meeting times and keeping to the time we had allotted. Then I needed to make my schedule one that would meet teacher needs. So that meant I rarely spent time in my "office" which actually was a desk in the lunch room where I was ,If there, available to anyone who needed me.
Finally I worked out a plan to work with staff. That at first took the form..new teachers or needy teachers first. Teachers I worked with were on a 6 week schedule. The first week I watched the teacher teach, took lots of notes and developed a plan for going forward. (It is important that anything I saw in a classroom stayed between me and the classroom teacher. I was not an evaluator) The second week I taught the class and the teacher watched and took lots of notes. The third week we taught together backing each other up. (all this time I wrote all the lesson plans). The fourth week with me still writing the lesson plans the teacher taught on her own knowing I would back her up any time still using my lesson plans. The fifth week the teacher wrote her own lesson plans and did her own teaching. Same with the 6th week as I might come and go and not be there all the time. To meet and talk about what we did we often tried to do it at recess with help for coverage from the staff. In the beginning we might meet every day but as time passed maybe 3 times a week. This was in addition to grade level meetings.
I usually did 3 teachers at a time. I know it sound hard but not impossible. I also know it worked. I had teachers who thought they could not teach math in the beginning become my best math teachers and the children loved math time.
100% - we should be using technology to support human connection, not intermediate it.
In terms of sharing what folks are working on, this is the thesis of Once (https://www.tryonce.com/) - use edtech to support grown-ups teaching kids to read, both to help parents at home who want to play a more active role in their kids education and to help schools train and coach paraprofessionals to deliver instruction. And you’re right about the need for better coaching - we do weekly 30 minute 1:1s with every instructor where we walk through recordings of their instruction, and people regularly share this is the best and most robust professional support they’ve ever received.
And as a result, we see very strong student results. Leaning into humanity may not scale as fast as rolling out a chatbot, but it works, and there are ways to use technology to expand access to excellent instruction to more kids and schools.
It was exciting to read about the LXD study today. Keep going, Once team.
Just on the wacky politics, it's not just in education -- anti-AI, anti-Big Tech sentiments are forging new bipartisan coalitions to oppose the hysteria around building data centers. Americans are looking at the fractured country that Silicon Valley has played an outsize role in shaping and they are starting to revolt. I agree that some ed-tech companies will survive, but something quite powerful is happening politically.
The "infinite patience" piece feels akin to another misalignment I always notice in the design of AI Tutors in that they are designed/prompted to "never give the answer", which again portrays that the designers don't actually understand the relational dynamic system they're trying to mimic.
Would also love to hear more on your thoughts on coaching since EDSI is doing work on that.
I’m very interested in working on the caregiver support misalignment problem! The way I’m currently thinking about it is through creating undeniable proof points in the areas the caregiver cares about so they can see it. Part of this vision is removing the separation between work and school, collocating them and integrating projects. Would love to hear from others interested in this “misalignment” space!
Here is a post about our AI design principals for elementary math: https://home.xtramath.org/blog/agency-over-agents
I describe a "plan B" by addressing the issue that the most profitable question in EdTech (can anyone build a synthetic educator) is not the most important one (how can we reduce the stress, effort, and fragmentation that sits between students and teachers every day).
AI is coming for the classroom, the question is whether it further fragments the system or makes it more coherent.
Hey Dan! How's it going? Guess what -- I'm the guy you quoted who asked "What's Plan B"? I'm also the guy who pointed out that many tutoring programs don't seem to scale very well. And I'm posting to say that you have completely misinterpreted what I wrote.
First of all, I'm not "grieving" over Khanmigo. I'm mystified by how you came up with that, but you are simply delusional. I'm not an AI advocate, I don't use LLMs, and I am deeply skeptical about the introduction of AI into the classroom. I'm also not a big booster of Khan Academy, although I think it can be a valuable resource for some students. So I can hardly be considered to be any sort of advocate for AI in education, or for Sal Khan and his crazy ideas. But I'm not **actively pleased** to see that Khanmigo has not lived up to its promise. This was the weird thing about your original post -- there were so many people who were explicitly cheering its perceived failure and gloating over this. And that's just wrong! There are lots and lots of educational initiatives that I think are misguided or ineffective, but I don't openly express active delight when they don't work out. That's not "grieving", it's just common decency.
It's also bizarre that you can't understand that educational initiatives have to be able to scale. Yes! Even an individual school with a few hundred students has to contend with this. Tutors have to be recruited, trained, and integrated into the curriculum, on a continual, on-going basis. This is possibly doable for a small number of tutors, but for a project like this to be feasible you need a lot more than a few dedicated people; to make this work at the level of a school district is very, very hard. You think that this is a "strange standard for evaluating interventions in education", but it's actually one of the most important. Tell me -- if it's difficult to scale an intervention from a small pilot study up to an entire school district, why wouldn't that be an important consideration? If there are major obstacles to implementing a proposal, isn't that an argument against the proposal?
I realize that you and I have different perspectives, and you are welcome to criticize my viewpoint. But it would be helpful if you could accurately characterize my position. And spare me the condescending "grieving" bullshit.
"if it's difficult to scale an intervention from a small pilot study up to an entire school district, why wouldn't that be an important consideration?"
Can you name anything worthwhile we could do in education that ISN'T difficult? Isn't that just a given? Yes, it's difficult, everything is, now let's get on to the doing of it.
Hey Dan! I'm an Educational Technology Coach at Classmate — we work with K-7 teachers across a bunch of schools in BC, Canada, visiting classrooms a few times a week (https://classmate.team/chronicles/).
Most of what we do is try to move tech away from consumption and toward making. Podcasting, board game design, science project documentation, digital journals where kids tell the story of what they built and why. We lean heavily on recording student voice with audio and video.
Your point on coaching misalignment made reflect. What we've found is that the most useful version of our support isn't a model lesson or a walkthrough, it's sitting with a teacher on something they're already building and iterating on it with them. When we look to embed tech into their lesson, it tends to open other doors too: a natural spot for a reflection check-in, a way to make student thinking visible to the whole class, an opportunity for cross-curricular assessment. None of that happens in a one-off visit, it's often after a few conversations and time spent in the classroom with the teacher trying things out. Tech tends to be a trojan horse to better pedagogy.
Second comment since it's quite separate from the first.
Stephen Downes (Ed Tech researcher at National Research Council in Canada) says and publishes interesting things on this exact subject. He has been banging on the "Learning should be social and human-connected" drum for decades. His theory/framework of "connectivism" is really interesting to me. https://files.eric.ed.gov/fulltext/EJ1350294.pdf
"If that’s you, let me know what you’re working on in the comments."
This post outlines a successful edtech-math marriage with a human-first focus. If you are skimming, the news article from Blackfoot (paragraph 5, "read about the incredible impact right here") is particularly inspiring.
What High School Math Gets Completely Wrong About "The Real World": https://substack.com/home/post/p-193435846