Tech in EdTech
Tech In EdTech improves the dialogue between education leaders and the innovators shaping edtech. This is your go-to show for actionable ideas and solutions that make digital learning not just possible, but effective, practical, and inclusive.
Tech in EdTech
If Everyone Has the Same AI, What's Your EdTech Advantage?
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Textbooks have always been static, with the same page providing the explanation. It never takes into account who's reading. Dakshay Mehta, Head of Technical Staff, Stealth AI, joins Eric Stano to talk about what happens when a book becomes a conversation instead. They dig into how AI tutors can adapt to different learning styles and how sequential curricula need to take a step back in modern learning.
00:02.49
Eric Stano
This is Tech in EdTech. I'm Eric Stano, and I'm joined today by Dakshay Mehta, the founder of PantheonAI. And today we're going to be talking about what happens when the textbook turns into a thinking partner, or more appropriately, more accurately, what comes after the textbook. Dakshay, for our listeners, I'd like you to introduce yourself and walk us through your journey into education. But, as you do that, I'd really like to hear a little bit about the fact that you've built Pantheon AI, your company, around the idea that knowledge should be interactive and conversational, not just delivered as static content. And I'd love to hear a little bit about as you talk through your journey. What led you to that view in the first place? And secondly, you know, a lot of education companies still talk about AI as an add-on to existing products. When you look at the market, what part of this shift do you think people are still underestimating the most? So, if you could talk us through your journey, and maybe reflect on a couple of those questions, that'd be great.
01:08.47
Dakshay Mehta
Yeah, first of all, thank you for having me. Appreciate doing this, and yeah, I'll walk you through. So Pantheon AI started as a podcasting company back in 2024, kind of when…
01:24.79
Eric Stano
So you've done this before.
01:31.35
Dakshay Mehta
Yes, so this is nothing new to me. So yeah, we were, at that time, Google just launched NotebookLM, and this was this new tech tool a lot of students were using, and we were like, you know what, like we can build something better. But just focusing on education. So that's what we built initially. And then we were like, okay, if we're doing podcasting, then what else are we missing? Then we asked the questions: oh my God, like courses or textbooks. I mean, that's kind of how we kind of like initiated the idea of, okay, we can make Talking Books, and we need to build a ground-up experience rather than add on to just like slapping on AI to everything, you know?
02:21.00
Eric Stano
Right, right.
02:21.91
Dakshay Mehta
So, that was our initial starting point into Talking Books. And then, once we kind of had that epiphany, and we were able to, you know, build some MVPs in terms of, okay, how does it even position? How do we make it not look like an add-on, and how do we make it from a ground-up experience? And that was like our whole goal that we were running after. So, then, we were like, okay, what does an educational experience what matters most? And then we went back to the drawing board and asked these questions. It was usually the common answer was the teacher, right? It's either the teacher or the tutor or the person actually writing the book.
03:07.20
Eric Stano
Right, right.
03:13.59
Dakshay Mehta
And then, we were like, okay, let's just connect these dots together. And how do we put this in an amazing experience that not just feels like work or like, man, I got to read this book now. Kind of looking at it from the aspect of let's just have a conversation. So that's when we got to the point that we have to create an experience where you have a conversation and learn at the same time while you, kind of, have this interactive media between you and the author, or the book in this matter. So that's how it all started. And that's where the Talking Books came from. And of course, the main origin of this idea was actually, back in 19- might be missing the date- 1986, or somewhere in the 80s. Steve Jobs went…
04:07.93
Eric Stano
So somewhere, well, but well before you were born, I presume.
04:14.17
Dakshay Mehta
Yeah.
04:16.70
Dakshay Mehta
I think two decades before I was born. And he went to this conference, and he talks about, like, hey, I would have loved to have had this Aristotle. I would have loved to ask Aristotle a question. And because we lived in different generations, that's what he's saying. And I'm paraphrasing this: he can't ask Aristotle a question. And can’t have him as a tutor. So that's where I was like, wait, why hasn't anyone done this before? We're living in 2025. And why hasn't anyone done this? I was like, okay, those were the questions we were asking. And then this kind of is how it, kind of, transformed. It's like, yes, you can have a book, you can come and listen to it, which Audible also does, but then you can listen to it in the author's voice. That's one. The second is you can come and listen to it, but then it's only in one language. Then we're like, okay, ah sure, let's just add a couple of 70 different languages because then like that democratizes the access. And then slap on the experience of having the ability to have a conversation. So, we put all of this together and boom, we have our product.
05:32.22
Eric Stano
Yeah. And yeah. And I've heard you, you know, describe books as thinking partners. It's a term that I really like for a variety of reasons. And, you've begun to, kind of, broach it and talk about the features. But, can you get into that a little bit more? You know, what does a traditional textbook or coursework model really stop doing well enough to remain at the center of the learning experience?
06:00.02
Dakshay Mehta
Yeah, so current textbooks are something that we print, right?
06:06.12
Eric Stano
Right. Right.
06:09.02
Dakshay Mehta
It's kind of something that is static. It's not adapting with the person. Not everyone is the same. Not everyone learns the same. Not everyone speaks the same language. A lot of moving parts here. So it's about, hey, should we just build this one textbook? And it's very static. The concepts are the same, right? We want everyone to be learning these concepts. But then there's adaptability that hey, I would want to learn more visually. And Eric might want to do something different, and he's more of an auditory learner. So everyone's different. So that's where these books can become a conversation rather than, you know, just being static. So maybe you want to talk to Albert Einstein while you're learning theory of relativity. If you're a student or a kid, you can just ask dumb questions. Like that's the beauty of this, right? You can be like, what does relativity mean?
07:03.39
Eric Stano
Right.
07:05.90
Dakshay Mehta
I'm five years old. Could you explain? I mean, you can also do that to ChatGPT, but then you'll get this verbose, kind of, cold answer rather than it sounding like Albert Einstein. And, you know, explaining that, hey, so-and-so kid, like, this is how we do this, and you should be not worrying about and giving these examples. So it's like very, very like pertinent to the person. That's something we're looking to do, basically.
07:38.30
Eric Stano
Yeah, and now you've talked a bit here really about, sort of, the almost, the adaptivity that your approach enables. It speaks to different learning styles, or can speak to the questions that a student or learner might have on the fly. And, as you were talking, it strikes me that somebody could potentially go down a rabbit hole and begin to ask lots of questions, which is great. They're demonstrating their engagement.
08:05.46
Dakshay Mehta
Which is great, yeah.
08:06.62
Eric Stano
Yeah. But you know, I'm thinking about that in the context of what publishers and institutions have traditionally done and still do. And actually speaking as a former textbook editor myself. You know, publishers and myself, we always relied on structure and sequence and editorial control, really for good reasons. To ensure that a student was marching towards the objective, you know, of that learning moment. In an AI native world, and given that the kind of product you've created, are there parts of that model that are still, you think, essential for a learner?
09:08.05
Dakshay Mehta
I really don't think so because, I mean, if this person is going in a textbook with an intent of learning something, that could be in chapter 1 and chapter 10, and maybe you want to put it all together. But if you were taking a sequential approach to things, then you might be missing a lot of context. That's just talking about a textbook.
09:28.33
Eric Stano
I see.
09:30.38
Dakshay Mehta
But let's say you're in a nonfiction book, and you probably want to learn, like, how does this author think, or what is the intent of what is he talking to me about? You probably need to skim a lot of pages for doing that. And maybe you missed some context overall, but you can just ask now, right? You can be like, so this is what I thought, and this is what I wanted to learn. So let's learn, right? It's very adaptive that way.
09:59.13
Eric Stano
Do you think there's any, just a follow-up question to that: do you think there's any danger in having these conversations, potentially because you've suggested that maybe that structure and sequence doesn't really play a role any longer? Do you think there's a danger that students might not achieve certain learning objectives that are important for, say, a certification or a degree or mastering a particular concept or skill, because they've had these conversations, but because of their interests might have moved somewhere else rather than towards achieving a particular objective. I'd just be curious as to your thoughts on that.
10:45.85
Dakshay Mehta
Yeah, I mean, so how AI systems work is they have, or at least our systems are moved in a way that, hey, they have intent.
10:53.57
Eric Stano
Sure.
10:55.53
Dakshay Mehta
Like, there is an end goal here, right? Until this end goal has been achieved.
10:59.57
Eric Stano
Right.
11:00.56
Dakshay Mehta
We're going to continue talking, or we're going to lead to something else, or we're going to ask questions that complete our objective here. That's how these AI systems are, kind of, built. So, I don't see that it's not going to end up like missing something.
11:11.96
Eric Stano
Right.
11:17.07
Dakshay Mehta
It's more on the sense that you might actually miss something when you're reading sequentially, because you might be missing, you don't have the context of the whole book unless you've read this book previously.
11:30.04
Eric Stano
So there's still a movement forward toward any particular objective, but the student is given greater agency to.
11:38.58
Dakshay Mehta
Exactly.
11:38.94
Eric Stano
It takes as much time, and it goes down as many tributaries as they feel is productive for them. Or, at least, is, you know, it's something that's going to engage them. So they're kept in the material and eventually will move their way to the objective at their own pace.
11:57.31
Dakshay Mehta
Yeah.
11:57.93
Eric Stano
I think I'm hearing. So, one other just- and it, you know, that I think a great deal of, you know, your product and what your company is doing, so I don't mean to sound like I'm challenging it all. But I wonder if there's a risk in conversational systems like this. While they're making learners feel supported, you know, is there any risk of them, the systems or the products, quietly doing too much thinking on behalf of the student? How do AI-powered systems keep the learner cognitively alive?
12:33.37
Dakshay Mehta
Yeah, I mean, that's a good question. I think there's a lot of people asking that question, which is the right thing to do right now. It's like there's stuff like AI psychosis. Will this kid just be too dependent on the AI to think for him or her?
12:50.24
Eric Stano
Yeah.
12:50.55
Dakshay Mehta
Um, those are the right questions, right? And that's why the guardrails come in for a very specific edtech tool like this would be like, hey, let's not just, you know, hand the cake and feed it. Let's also be able to explain what this cake is made of and then have an engaging conversation so that we make sure they're able to understand, okay, these are the ingredients, and this is how the cake is made, rather than just, hey, here's the cake and here, open your mouth and let's eat this, right?
13:24.18
Eric Stano
Right, right.
13:24.54
Dakshay Mehta
So those guardrails are definitely something we should be talking about with that tech tools as AI psychosis is a huge, huge problem.
13:33.77
Eric Stano
Right, right. And yeah, I'd like to get your insights too on, sort of, shifting topics ever so slightly. I'd like to get your thoughts on moving from from content design to to to knowledge system design. Because, you know, I think the the old way for for publishers and product companies who are, you know, focused on learning. I think that the old ways of doing things are are are evolving. And I'd love to hear your advice to an education company or an institution that they were asking right now: what capability should they build internally so they don't accidentally outsource their own future to generic AI layers? how How should they be rethinking what they do right now?
14:27.06
Dakshay Mehta
Yeah, I mean, I think the, I guess you're asking pretty much what is the moat that they need to highlight in order to stay relevant?
14:34.30
Eric Stano
yeah and and Yeah, and a lot of these these companies, these institutions have focused on content design. um And, you know, they're they're aware of AI and its power, and they they they look to harness it.
14:42.65
Dakshay Mehta
Yeah.
14:49.30
Eric Stano
But there you're you're doing something that's entirely organic to your own company. How would you how would you advise others to begin thinking about,
14:55.32
Dakshay Mehta
Correct.
14:59.29
Eric Stano
you know, their own services, their own role, their own products. If they've been accustomed to simply designing content for consumption in perhaps a more traditional or static way, what should they be doing right now?
15:13.14
Dakshay Mehta
Yeah, I think data I think data is the biggest moat in any kind of industry, be it AI, stock trading, or education as well.
15:20.26
Eric Stano
Sure.
15:24.03
Dakshay Mehta
So data would play a huge role for any kind of institution or application or a company, right? It's all about data now. Now, what do you do with this data? Yeah. that's That's like the biggest question, or the billion-dollar question here is what I've seen other companies do is and and we're we're also doing internally, is recursively learning through this data. You can't just pick a lemon and be like, it doesn't have access to this new data that you have created after the users had a conversation in this chatbot that you have. And that is a new organic data. It's like a goldmine, pretty much. Now you run evals and see what what did the guy say wrong? What did it do right? And then kind of feed that into your product or service and make it better. So I think that would be something everyone should be doing.
16:19.99
Eric Stano
No, I think that's ah's that's really wise. And taking that a step further, again, thinking about these, essentially your competitors or other occupants of the ecosystem that you've entered. You know, there are new roles that are beginning to appear in the marketplace. Things like knowledge architects and AI trainers and even conversation designers, I understand. Which of these, or maybe one that I haven't mentioned, what roles are going to become strategically more important in the immediate or the long-term future?
16:58.65
Dakshay Mehta
I think, ironically, like everyone's going to end up being a manager, right?
17:07.04
Eric Stano
Sure.
17:07.26
Dakshay Mehta
I think everyone's just- any field, let it be biology, let it be engineering. I think every person is just going to have a bunch of agents that they're going to be overlooked and they have to manage and QA, quality assurance in terms of, hey, are they doing the right thing? Have they drifted? Do I need to bring them back? Or what is the output they've put in terms of content or code? Let it be if it's engineering. So, I have, in the past year and a half, personally, just become this manager of agents. I call myself a manager of agents, and all it is is like there's a content agent and there's a coding agent and there's 10 of those. So, you just like, look over these people or agents in this matter. So it's kind of like you're just mimicking being a manager. So you're this mega product ah product manager or whatever you want to call it.
18:04.00
Eric Stano
right.
18:08.76
Dakshay Mehta
Whatever your role is, you're just becoming this orchestrator.
18:12.98
Eric Stano
You know, that's funny; you bring to mind a friend of mine who has always insisted that he never wants to be a manager. He always considers himself an individual contributor in his company. And I have to have to break the news to him that his future does lie in management.
18:26.52
Dakshay Mehta
Yeah, I was about to say.
18:29.39
Eric Stano
And also, I'm convinced there is a Karen-style joke in here about asking for the manager, or demanding to see the manager. But I like, but I can't quite define the joke, so we'll just move on to the next topic here. So I do want to shift to, you know, there's a lot of hype around AI, just broadly speaking, and its various applications. You know, as more learning products start using the same underlying AI models, what will distinguish AI native learning experiences going forward?
19:08.70
Dakshay Mehta
Yeah, I mean, I think we'll go back to the point that data is very important. AI native being you learn from what is going on rather than just making an LLM call, which is very generic, what we call are GPT wrappers.
19:13.56
Eric Stano
Yeah.
19:25.02
Dakshay Mehta
And that's what anyone can do, right? Anyone can pop in and create this project, whatever you're building.
19:33.36
Eric Stano
Sure.
19:33.40
Dakshay Mehta
But then, what you can bring to the table is data.
19:34.84
Eric Stano
Right, right.
19:37.30
Dakshay Mehta
So if you have a million users, a thousand users that are using your product, they are essentially, even if the product is not as good as you want it to be. They're essentially helping you improve the product. So I think that's like a huge advantage here.
19:53.37
Eric Stano
right
19:59.00
Eric Stano
Right. So the, it did, not to put words in your mouth- but just to try and even rephrase that, it almost sounds as though that the product, the best products are going to be the ones that learn from the learners themselves and the learners' own experiences.
20:12.31
Dakshay Mehta
Exactly.
20:14.80
Eric Stano
So there's, sort of a meta style to that in and of itself, which is that.
20:18.17
Dakshay Mehta
Yeah. And ironically, we call it recursive learning. So, this is an EdTech podcast.
20:22.01
Eric Stano
Right. Right. Of course. And, you know, having been also a writer on the side, just the word recursive alone sings to me. So appreciate the terminology there. And, last question, on this front, you know, AI products can often feel pretty magical in the first five minutes. Yours certainly does, to have, you know. I think you told me that you have Santa Claus talking to somebody and Jesus talking to somebody, and you've got obviously people like Aristotle and, you know, various luminaries talking at you.
20:51.30
Dakshay Mehta
Mm-hmm.
20:57.77
Eric Stano
That, by definition, almost feels magical. How can, just relative to AI products overall, how can an institutional buyer tell if a conversational knowledge product is going to have real staying power? Is it in that that that harnessing of the data that you've been talking about and its ability to learn what works and what doesn't, or is there something else that's gonna communicate to ah to a buyer? Yeah, this is going to this is going to last, this is going to persist, and I'm going to succeed using this product?
21:34.65
Dakshay Mehta
Yeah, I mean, you're essentially asking if it's going to be a sticky ah medium going forward.
21:39.72
Eric Stano
Yeah. Yes. Yes.
21:42.92
Dakshay Mehta
Honestly, honest answer, we don't know, right? As AI is moving at light speed, there's different modalities that people are getting used to, specifically being voice, voice also being conversational. But then recently, what trends we've seen is people are moving towards shorter conversations, like, hey, back and forth, how you text a text your friends, you just have smaller back and forths ah with attention span dropping and blah and blah, like those things happening. I think we really don't know. It just comes down to who we're talking to in terms of ICP. Are we talking to kids? Would that generation really doesn't want to spend time reading verbose responses or listening to longer um content. And then there's older adults, like 30-plus, that love podcasts. Two-hour podcasts, this guy will just listen to it like it's um like it's It's like a prayer, honestly. Like you can sit there and the people are just listening to these two-hour podcasts, no problem.
22:51.52
Eric Stano
Right.
22:55.64
Dakshay Mehta
So it just comes down to the person um of what they prefer. It's generational, honestly.
23:03.42
Eric Stano
Right, right, right. And yeah, hopefully there is, you know, some future for that long-form podcast style. You know, as you were talking, I'm thinking, you know, is it purely the modality that people are gravitating toward, and, you know, if it's not a conversation in which they could imagine themselves participating or just in, you know, being near that dialectic and that dialogue. um does if it's going to be just them consuming content straight, does it need to, you know, be more bite-sized or more modular by its very nature? Just because it's not as engaging as the drama of a that's inherent in a podcast. um questions we won't answer here but uh but you've got me uh you've got me down my own rabbit hole. All right, now we have, the folks who set up these podcasts have institutionalized in them what's odiously titled the lightning round.
23:48.50
Dakshay Mehta
yeah
24:02.65
Eric Stano
So what I'm going to do is fire a couple of questions or quasi-questions at you, and we'll just go through and if you could just answer them quickly. We'll just move through these, but I think folks will learn a lot from your answers because the questions themselves or the statements are pretty interesting. So you’re ready?
24:23.58
Eric Stano
Okay.
24:24.63
Dakshay Mehta
Yeah, let's go.
24:25.76
Eric Stano
All right. Here we go. That was a little slow, so you've got to be faster than that. All right. What assumption that publishers should retire right now?
24:38.58
Dakshay Mehta
I think IP rights management um is going to get tricky.
24:45.46
Eric Stano
Okay. Um, as an old publisher that, uh, that, that just made my, my heart skip a beat. Um, one side, one sign that a product is building a real knowledge infrastructure.
24:59.93
Dakshay Mehta
um Again, data mode, very important. if it' If it's not there, then you're you're doing something wrong.
25:03.13
Eric Stano
Yeah.
25:07.00
Eric Stano
Yeah. Yeah. No, I anticipated that one. One thing about the future of learning that you are genuinely optimistic about.
25:16.17
Dakshay Mehta
I think the adaptability and personalization, we're still not there yet. I think we're seeing the the early days, and you're going to be very surprised on how these AI systems know and how much they know about you, basically.
25:31.96
Eric Stano
Yeah, yeah, which is both, I think. ah you know, something good to be optimistic about, something also to be, you know, to furrow your brow about as well.
25:38.42
Dakshay Mehta
Yeah. Yeah, meta question. Yeah.
25:42.88
Eric Stano
But yeah, let’s hope the future is rosy on that score. All right, you know, moving out of the lightning round, five years from now, what part of today's learning product model do you think is going to look obviously outdated in hindsight? Is there something that is on the score that's popular now or hot now that you think, nope, doesn't have staying power? That's going to look like an eight-track tape um five years from from this point.
26:17.57
Dakshay Mehta
Yep. I think that's a great question. So I think you brought it up, the sequential learning part. Like, yes, we have textbooks.
26:23.70
Eric Stano
Right Yeah.
26:25.07
Dakshay Mehta
Yes, we have um like, hey, we have this agenda. You need to follow this curriculum. I think that's going to get retired. I think what kids and everyone are going to go towards is this goal structure that, hey, I need to learn theory of relativity. Now, I don't know how I'm going to get there. Eric's going to have a different path getting to that. Maybe he goes and has conversations with Albert Einstein or, you know, watches these different videos, reads these different books that are like a second degree to explaining the theory of relativity. Just people will have their own journey to getting to this goal. So if there's a course, let's say AP Physics, and the exam could just be on relativity now, maybe the kid learned theory of relativity through just doing math problems. Like that's, that's like, I would have done that. And then there's some other person on the other side who just, like, watched these different videos to get here. So it's like, it's just, everyone's going to be on their own journey is what I mean. And the sequential part is going to be retired for sure.
27:37.24
Eric Stano
Yeah.
27:40.73
Eric Stano
Yeah.
27:41.05
Dakshay Mehta
Like there's no A, B, C, and then you get D. It's just going to be you figure out how to get to D. You have you go B, C, A, whatever you want to do in the middle. Yeah.
27:52.44
Eric Stano
Right, right. No, i'm I'm smiling because I'm just thinking, you know, as, as you know, Dakshay is ah choosing to learn a theory of relativity and has his choices of methodologies or modes to do that. I'm just imagining you thinking about your own product, and you're like, I'll just do math problems because Albert Einstein just drones on and on and on. I can't say this. So, yeah, yeah.
28:15.74
Dakshay Mehta
I mean, why not, right? Like, whatever works.
28:18.46
Eric Stano
Now we're going to be imagining all these people at dinner parties and choosing who to sit next to. um So, um, yeah, if an edtech builder, publisher, or institution wants to prepare for the shift that you articulated. You know, you've already talked about data, but what else? What's the smartest move they can make this year right away to prepare for this move of, Dakshay learns theory of relativity by doing math problems, and Eric does it by watching videos or listening to Einstein drone on. You know, what's the smartest shift in the edtech builder public organization?
28:58.58
Dakshay Mehta
Yeah, I think it's just like not sticking to one modality. It's just all about like expanding your horizon on interactive and interactive experiences. Yes, you have texts, but then you should also have these artifacts of different ways to doing things. So it's just about what you provide your end user - more options. That's very important right now because it's possible with AI because you have AI VideoGen, you have AI Music, then you have text. There's different modalities now you can just give the user in the same amount of time and great quality. So it's just all about this, you know, go more horizontal rather than just going vertical aspect of things.
29:45.28
Eric Stano
All right. Well, listen, you know, we've been talking a lot about conversational knowledge systems, and I have to say that this conversation alone has been, for me, really, really fascinating. Now, if others, our listeners want to follow your thinking on conversational knowledge systems and, so to speak, continue the conversation, where should they find you?
30:10.23
Dakshay Mehta
They should find me on Twitter, or we call it X now. That's my first name is my username: Dakshay.
30:16.20
Eric Stano
Okay.
30:18.42
Dakshay Mehta
I'm very active there. other Otherwise, um yeah, you can't find me anywhere else, pretty much.
30:25.10
Eric Stano
Okay. All right. Well listen well listen Dakshay. thank you for not keeping yourself all that well hidden. I appreciate you spending time talking with us today and um sharing your thoughts again. I think it's a it's a fascinating product. I think you've got some really uh interesting innovative insights and, you know, with folks like you leading the charge, I think that the future of learning is certainly going to be interesting. But I suspect it's going to be a whole lot more effective as well. So I really appreciate you spending the time with us. Anything, any last words that you'd want to leave with the listeners?
31:02.38
Dakshay Mehta
No, think we're in very exciting times, and I appreciate you having me here. And thank you so much.
31:09.27
Eric Stano
All right. And thank you for invoking the Chinese curse of May you live in interesting times. So let's hope it's a blessing in this particular context. All right, Dakshay, thank you very much for our listeners. Thanks for tuning in. This has been Tech in EdTech.