Magic School uses AI to help kids learn, not cheat | E2196

Hey, everybody. Welcome back to Twist. This is Alex. And I know you're probably just coming back from the OpenAI browser stream. Don't worry, we're going to get to that tomorrow. But today on the speak in startups, we are talking to founders. First up, Adeel Khan. He's the founder and CEO at a startup called Magic School that wants to bring AI into the classroom responsibly. Think about things like helping teachers with their lesson plans, creating more individualized setups for children that might need it, and, and on the student side, AI literacy and perhaps even getting early feedback on papers. It's a really cool idea. It's growing like a weed. And of course, yes, we do talk about privacy a little bit and the risks of increasingly digital classrooms and what that might do to our children's brains. But based on all the data I can see, AI is coming to school and Magic School wants to do it. Well, then after that, we're gonna take a founder question. This is a chat back from when we had Eric Gliman on the show, the CEO over at Ramp, you may have heard of it, and we thought it was a really fun one. Someone wrote in asking about how long it should be for an employee to vest their shares. Four years, six years. Should it be different than what a CEO sees in their own pay package? Well, we dig into that and more. And then to wrap up today, perhaps our favorite ever on screen pitch as part of our Gamma Pitch event. This is a company that I didn't see coming, that I hadn't even thought of as an idea, but I absolutely love and I got to learn quite a lot about material science and during the conversation. So sit back, enjoy. We have a three part show for you today and each part is absolute gold. Have fun. This week in Startups is brought to you by Gold Belly. Gold Belly ships America's most delicious iconic foods nationwide. Get 20% off your first order by using the promo code TWIST at checkout. Gusto. Check out the online payroll and benefits experts with software built specifically for small business and startups. Try Gusto today and get three months free@gusto.com Twist and Zeit. Zeit is the fastest way to build business software with AI. Build apps, forms, websites and portals that connect to the tools you already use. Go to zeit.com twist and get 50% off your first project. AI is absolutely everywhere. It's doing great in the realm of software engineering. It's taking a big chunk out of the legal market, is showing up everywhere in the world of fintech. And one other place where AI has found an enormous footprint is in the world of education. Now, when ChatGPT came out, everyone became worried. What will we do if the students just use AI to write their assignments? Well, the picture is a bit more nuanced than that. And some people, especially entrepreneurs, are taking AI tools and bringing them to the classroom in an official context, offering tools and services for educators and and students and administrators alike. And one of those companies, Magic School, seems to be doing quite well. It raised a 45 million Series B earlier this year, so I had to learn more. So please join me in welcoming to the program. It's Adeel Khan, the co founder and CEO of Magic School. Hey, Adeel, how you doing?

Alex, thanks so much for having me.

Ah, my pleasure, man. I love talking about bringing AI to places where people might not expect it to go and also to kind of flip the narrative because I feel like AI in education has gotten a pretty bum wrap so far. And so I like the idea of taking a tool that some people are worried about and turning it into good. So let's start there. What does Magic School offer to students and then teachers and then school districts? Let's kind of go up the chart here.

So Magic School ends up being this teacher copilot. So it is AI for educators where they can use it for their daily tasks, like maybe building a rubric for a project they're bringing to their students. Differentiating materials for students who are ahead of the curve and students who might need additional support. English language learners, students with special education plans who might be behind grade level in certain areas of academics. So it's their copilot for their daily work. And then they also can craft these AI experiences. These are all teacher led, so they have to actually build the experience themselves and then launch them to their students. These range from things like a ChatGPT for kids with guardrails and, you know, doesn't give them the answers, but helps guide them through, they're supportive, to support them with their academics, to more targeted, personalized learning tools like a writing feedback tool where students can submit essays that they've written on their own and then get AI aligned feedback, feedback that's aligned to a rubric that the teacher is going to grade them on eventually. Areas of strength, areas of growth instantly. Instead of waiting two weeks for their teacher to grade it and give them feedback, they get formative feedback before they get their final grade. It's incredible. As a former English teacher, this would have been a godsend for me. In my classroom. And then for districts, they end up picking up Magic School because they kind of want to make their own closed sandbox version of generative AI for their teachers in their district so they can infuse their own district knowledge with rag technology into the platform and make sure that it's aligned to their district's priorities and goals and not just a generic AI that their teachers might be picking up in different ways. So they get to kind of create their own ecosystem of gen AI tools for their educators.

So let's talk about the AI tools for teachers and helping them create rubrics and do some bits of grading or assessment and so forth. And my immediate concern is, oh Lord, the AI is going to be designing lesson plans. And you and I are both AI bulls, I believe. But I would say that the teachers that I grew up with were so experienced and so kind of in tune with students that they could certainly do a better job. So allay for me my concerns that feel very Luddite ish now that I say them out loud. And maybe give me an example of a rubric that a teacher could put together with Magic School that would be better or more in depth or more flexible for students than they could do on their own.

Yeah, so that's a great question. So one of the ways that we think about this is that, you know, the teach still the center of the classroom. Our slogan is teachers are magic. And it's a catchy slogan, but I know it's good. It is also kind of the ethos of the company. We think teachers need to be in control. AI should. Should not be the one teaching students. For the example that you shared a rubric, the tools are set up very similarly where the teacher is asked to give very specific input into the tool before the rubric is generated. They can do things like upload their standards that they're required to teach to their students from the state level. They can do things like, you know, give additional instructions or the, the text rubric would be associated toward, or, you know, multiple things that rubric would be associated toward. And then they could also add whatever additional instructions they.

They would.

They would like.

And they, they click generate, they get the rubric, they can edit it in the platform. And then more importantly, where this becomes really powerful is when you can differentiate this rubric for several different audiences.

In your I second time you said differentiate. My read of that is it means essentially creating variations of the same rubric for students that are at different levels of learning or preparedness. Is that correct?

That's exactly right. And that, that is. When I was a PR, our 9th grade classrooms, our English teachers would have students who were three grade levels behind in English. We had students in the classroom who were brand new to the country, who were just learning English for the first time in ninth grade. Then we had students who are well beyond their grade levels expectations, and the teacher as a single person is required to differentiate or meet all of their needs at once. This was kind of like an imagined thing that we asked our teachers to do, but really only the most veteran teachers who had versions of their materials that they, they had before, they could spend maybe the next year doing it. Now you can do this in like the snap of a finger. So you take that rubric and then you differentiate it in Spanish. You make a version of it that better serves students who are behind grade level, you make a version of it that serves students who are above grade level. And now you have this incredibly differentiated assignment for all of those groups. Reading like meeting a very practical need of educators today. One of those things that causes burnout is these, you know, incredible expectations we hurl upon teachers and don't give them support to, to actually execute on them.

That makes a lot of sense to me, and I think it explains why we need this. Because on one hand, if you imagine like the kind of ur classroom environment, there's like 12 kids in this fantasy land and teachers known them all throughout the, you know, I don't know, third through fifth grade or whatever. But in reality, classrooms are large, they're multilingual, they're spread out across the current capability curve. And so it makes a lot of sense to use tools to essentially take that one teacher and, and then accelerate them, expand their capabilities. I don't know what the correct word there is, but we do have a pretty large teacher shortage in this country. According to data that I found. This is from the Learning Policy Institute. There's at least 412,000 positions that are unfilled or filled by not fully certified teachers in the US So it's enormous. When teachers get their hands on Magic School, how long does it take for them to go from, okay, we have a new tool to being able to use it at a sufficient level of proficiency in the classroom to actually reduce their workload and then reduce the risk of burnout.

So, you know, I think that the core of the product is that it's so easy to use and it's so familiar for teachers. One of the compliments I still take a lot of pride in that we get today Is that this must have been created by somebody who worked in schools. Like, how is it possible that you really know my job this well? You know, there are AI tools for other professions. You mentioned when you started, like, there's a bridge for health care, there's Harvey for law. We just assume that these professions are, you know, worthy of like specific AI tools or there's a level of knowledge needed. And we don't necessarily, as a society think about how technical a teacher's job is. And it is extremely technical. There are so many facets of what a teacher needs to do every day outside of the classroom teaching, including in the classroom teaching. Things that go invisible to parents, they go invisible to the general society. But those were very intimate to me when I was a teacher and a principal and assistant principal. And we tried to build all of those things into the platform.

I love sending people little gifts, little gestures, you know, and if you're running a business, it's a great way to stay top of mind with your clients, your partners, your customers. So next time you're looking to impress people, consider Gold Belly. I think you've heard of Gold Belly or where you been. They deliver America's most delicious iconic foods nationwide. You've read about these mush dry dishes from places all over the country that maybe, you know, you haven't visited or maybe you grew up there. Here at the office we were at Stan's Donuts straight from Chicago. You know those old fashioned with the blueberries and the frosting, they got amazing reviews. People in the office went crazy. And just think, what are people going to think when you start sending Maryland crab cakes or my favorite, Terry Black's barbecue. So next time you're looking to impress, I want you to go to Gold Belly. Red farm has that Peking duck. They've got some of those great knishes from the lower east side that I'll send to people. Bagels from New York, you know, my hometown of Brooklyn. You can even get pizza too. We're gonna give you 20% off your first Gold Belly order. Oh, Magnolia Bakery. I forgot about that one. That's another one I love. Gold belly ships. America's most delicious iconic foods nationwide get 20% off your first order by using the promo code twist at checkout.

One of the things that's really interesting is that it's also solving that problem of like novice teachers in the classroom or under certified teachers in the classroom. We hear from schools and districts that sometimes the non teachers who are in the building the long term subs, the maybe the career and technical education teacher who's coming in to teach welding in a school to help students get living wage jobs. If college isn't a path for them, those folks didn't get trained as teachers. And Gen AI can be an incredible way for them to learn how to build a lesson plan and build content associated with these subjects that don't necessarily get big dollars for curriculum. So Magic School ends up being this incredible solve for the practical challenges of schools and districts right now. And yeah, it's been incredible to see the impact and the spread of the product. I think that the ease of use speaks to how fast it spread. There are over six and a half million users signed up for Magic School. There are three and a half million educators in the United States and about half those users in that number are in the United States. So it's like a staggering number for distribution.

I'll show the chart here. This is from. This is the chart that you shared on LinkedIn, but I pulled it up for everyone today. This is from a company called Burbio. This is a tracker of the percentage of school districts that have paid to an AI platform in the last 12 months. And what you can see here is that there's a company called School AI that has some market share. There's Chad, GPT, OpenAI, which is quite well. And then in first place there is Magic School. And I wanted to ask about this because when I think about AI tools that have reached the mainstream, I mean clearly ChatGPT with 800 million weekly active users has really crossed over into the, you know, normal conversation. How did you grow so quickly selling to a famously difficult industry? Because selling to governments, selling to schools has always been, at least when I was learning about venture capital and startups like a big, like red flag, don't even try. And yet you've generated enormous market share. So talk to me about distribution and sales.

Yeah, as like a non technology person, I learned down the line that I made a lot of mistakes that turned out to be the right decisions. I always think about like when I started the product there was, I had no monetization plan. You know, we were more or less a team of volunteers who launched a product in the world. I wouldn't even necessarily call it a company at the time, but I figured, you know, I've been a consumer of technology my whole life. I figured if I can get people to use this thing, we'll figure out the business stuff later. So I started seeding it to my teacher friends, folks I'd met through my career and it was almost purely organic. We didn't spend a marketing dollar until we got to, you know, nearly a million educators signed up for the platform.

That's crazy, man.

When you build something that people want, like they will share it with others.

Is that, is that just teacher word of mouth? Like teachers saying, look, this is really helping me organize my lesson plans, handle, you know, having lots and lots of kids and then it just kind of spreading across the nation?

Yeah, that's exactly. And it turns out that, you know, teachers who are, you know, wired to help.

Right.

That's in their DNA. When they find something that's genuinely helpful, they will tell the teacher next door. And we certainly caught moment of generative AI kind of coming to the world and finding utility. But I think there was the ease of use. It being really directed to their profession and knowing them really, really well. All of that in combination ended up in this extraordinary growth trajectory.

So speaking about the business side of this, I was curious who you sell to because on one hand I know this as, you know, free for teachers on the site. You also have stuff aimed at the district level. So are you selling individual schools? Can a teacher buy more magic school capacity themselves or is it district only teacher? Talk to me about the, the revenue side of things.

Yeah, so it's very, very generous. Free product for teachers. We have a plus version of the product that has a couple extra features, but it's not very heavily gated intentionally. We want teachers to get a lot of value out of the product even if their district hasn't purchased it yet. And then our primary customer is the district. We want to sell directly to the district. And again, the reason why a district wants to join is similar to, you know, the easiest comp is like a ChatGPT then ChatGPT plus than ChatGPT Enterprise is. You know, they want data security and protection, they want single sign on. They also want to be able to infuse their own knowledge into the platform. They want to create their own customized versions of our AI tools that are backed by their state standards, their district policies.

So it's going to be different based on not only if you're in Louisiana, but if you're in, you know, the New Orleans school district or a different part of that state, just to pick an example.

That's exactly right. So you know, education is incredibly regional.

Yes.

So like we have teams that are based more or less every state and region that understand like, hey, in this state the district's priority is typically this and they actually work directly with the customers to help customize the product to their needs. So and usually they're heavily contextualized to that area. So right now in Florida, we're working directly with districts to build customized tools for the Florida state standards. And there, there are employees are in the school districts working with their district technology leaders and academic leaders to make sure this tool is customized to their needs. In broader tech, this is considered like, you know, forward deployed engineers or you know, we call them solutions architects at Magic School and they are doing this work to really deeply customize the platform for the district. And you know, one of the things I share with our team, especially on our customer experience side, is the next greatest thing that Magic School builds as a feature is actually the application of the current technology. Like, if we can apply it really deeply and thoughtfully to the district's needs, then we can, you know, we can change the world. It's not about waiting for like the next shiny feature. It's about taking this incredible tech that we have and making it incredibly contextualized to the enterprise or the district.

And underneath Magic School, you guys actually use a variety of AI models. You mentioned a couple different providers on the site. Is the underlying technology still improving at a, at a material rate to help you guys do more or is there already maybe enough capability inside of, you know, I don't know, OpenAI's 4O and Gemini 2.5 to accomplish everything you're currently.

Working on at the current moment? I would say that, yeah, the models are pretty extraordinary for the tests that we have them do right now. They do most everything exceptionally well. We use mostly anthropics and OpenAI's models in the platform.

Okay.

Where I think they can still improve would be in for us. Like things like guardrails for students.

Yeah.

Like evaluations around text complexity, audio models. We partner with eleven Labs for some of our audio tooling in the platform. One thing that ends up being pretty challenging is when younger students are using like text to or speech to text, they'll often, you know, pause or stutter or you know, these things that normal language wouldn't adopt adapt to.

Right. They're not.

They're not. Students need to learn. Students need to learn how to type. I heard, I've heard from employers of recent graduates that they type like this. And we're roughly the same age.

Ish.

And like, I don't know, growing up, like we had to go to typing class, like, I sound like the world's oldest man, but bring back typing class.

But you know, typing class is kind of like, I mean, AI is the new typing class, right?

I know.

As I was saying that I'm like, but you use voice mode on chat GPT when you're tired. I'm like, okay, fair enough. Yeah. Yeah. What's your general philosophy about students using AI in the classroom? The College Board says that the percentage of high school students who report using Gen AI for schoolwork rose from 79 to 84% between January and May of this year. So it's gone from nearly everybody to nearly nearly everybody. And again, I haven't been in K through 12 for half my life because time passes. But I can't imagine, like, my school handling it well in terms of how they approached changing maybe curriculum or expectations or so forth. So what do you think is the right amount of AI for students to use and interact with in the classroom? Because I know we want literacy, but we also don't want a lack of creative thinking.

Yeah, well, I can, you know, quell your fears. Like, schools have done an extraordinary job of really thinking deeply about how they want to implement AI. We have been so inspired by our partners and how helpful they've been in their implementation documentation. So I'll give you my own opinion, but I think that a lot of districts will exist on a spectrum of how students should or shouldn't use AI. But for me, I think that there's real risk in cognitive offload to students early in their lives. And it should probably be pretty limited as kids are really early and maybe for very specific use cases. And then as kids get older, I think it ends up being. This still needs to be really thoughtfully guardrailed and like, you know, grade by grade, scaffolded in a way that ends up up ensuring that the students aren't getting too much support along the way. But I think it can be really, really powerful. The writing feedback tool I shared earlier ends up being this really powerful tool to grow students and their writing abilities. And they're still writing completely on their own, but just getting feedback right in time to revise and improve. So, you know, I'll capture this in a story of a student that I talked to recently who was in 10th grade. There's a friend of mine's daughter, and we're talking. She knows I used to be a principal, and she's asking me what I'm up to, and I say, you know, I work this company called Magic School. She's like, oh, should stop. Oh, we use this in school. I was like, oh, amazing. Like, tell me about how you use.

It, how Tall did you feel in that moment?

Oh, I felt 10, 12, 10ft. Nothing gets more exciting when students say it like they're excited about it. And so she tells me about it. Her history teacher uses the writing feedback tool with them. So she writes her essays and history gets feedback on it. She very articulately explained to me why it was valuable and how useful it was to get the feedback, revise her essay and why that was like a good use of AI. I wish I could, like, you know, she could have been the best advertisement for Magic School.

She should have recorded that moment and.

Just blasted Nobody gets better. So she ends up. Then I ask her, you know, tell me about your friends. You're in 10th grade. Like, you know, AI is kind of this big thing happening in the world. What else are they doing with it? Like, what's happening with AI? She's like, well, you know, some of my friends are using it to cheat. And I know that's like, you know, something I can do too. And it's like an easy shortcut sometimes. But, you know, I've chosen to use it, like, in a really limited way sometimes. I use it like to as a thought partner. I really like when Magic School happens in my classroom.

But Launch is a fast growing organization. We've got more than a dozen employees working with me here in Austin and another dozen spread out all over the world. But there's so many moving parts when it comes to hiring and managing employees. There's the onboarding, of course, payroll, you got to pay them. And listen, I've got all these podcasts. I don't have time for payroll, benefits, hr, taxes, answering questions, nor do I want to hire a full time person and then have them do five hours of work a week. No, I have the perfect partner, Gusto. They're the all in one payroll and benefit product that's built just for your small business. Easy to use, it's incredibly fast to get started, and it's designed specifically with remote offices in mind. And Gusto is not just giving you helpful tools. They're going that extra mile to keep your workers happy and keep everything running smoothly. And they're now offering level funded health plans to keep your insurance costs down and on demand pay to help workers get access to their cash faster without paying extra interest or hidden fees. So here's your call to action. We want you to try Gusto today. So we're giving you three months free when you run your first payroll. That's right, three months free. F R E. That's my favorite Price, folks, go to gusto.com/ that's Gusto. G-U-S T O.com/ what an amazing service and a partner. Great partner.

You know, I don't want to give away my thinking too much. Like my brain's in this important development stage and so I'm thinking about using it in a really limited way and trying to be creative. But again, like when I take my tests and I'm applying to college, like, I know my voice is going to matter and it's going to be proctored anyway. So like, you know, I'm not going to have this AI tool with me when I'm doing those things.

Proctor for folks who haven't been in this system for a long time is when you're essentially doing something by hand and people are watching you take the test. You can't, you can't cheat. You can't have something up on your screen.

That's right. So. And the first thought I had was like, what a thoughtful young lady. Like how incredible it is that you've been. You have this much agency to decide how you're going to use AI.

It gives me hope for the future, frankly. I hope, I hope everyone's like her.

That. Yeah, same thought. And then the second thought that I had was she must have an incredible teacher who like helped her understand, you know, the difference between good use and bad use.

And.

And again, shouldn't every child be able to have that articulated response, that thoughtful a way of experiencing technology in the world? So I think that captures a story of how I want the world to evolve and students to use AI is, have tons of agency and use it for the right reasons and be careful around the risk as well.

I bring it up because I was reading your guys post from August 5, 2025 called Magic School's Commitment to Responsible AI Use in Schools. And you have. I'll just pull it up so everyone can see it. You have a bit of a kind of like demo text. I think this is what a school district might say to children about how use AI and this 8020 rule stood out to me. It says, let AI handle the initial 80% of your draft and then add your final touch as the last 20%. Is that a standard rubric that schools use or is this just an example that I'm over indexing on because I'm paranoid about people not being able to.

Write in the future? Mind you, this is not the message for students. That is the message for teachers.

Lovely. Okay, good.

So yes, yes, you should Be fair to it. I would never say that to students. Like they need to be doing 100 of the cognitive lifting. And so that message is, when a teacher signs into the platform, we let them know that, hey, Magic School can be this incredible augmentation to your work and help you feel really efficient. But be, be careful about offloading too much of your work to Magic School. Make sure that you're looking at every piece of material produced by the platform before you get in front of children. Because of the things we know, like there can be bias hallucinations, sure. So still trust your expertise. But yeah, that, that rule would definitely not apply to students. And that's something that we're encouraging.

Makes perfect sense. Just wanted to bring it up while I had you. So just to summarize, what Magic School has built is a series of tools to help educators not only differentiate their lessons to better fit each student, but also just to do more and also provide better, faster feedback to students. With tools like the writing assessment, on the district wide side, you can ingest your own information standards and so forth to create essentially a centralized AI tool to offer to schools. And on the student side, they get better education via more tailored lesson plans and so forth. Okay, this all sounds fantastic to me. I'm glad things are going so well for the business. But I am curious what's next? Because you've really touched on a lot of the places where I can see AI fitting into K12 education in a positive sense. So I'm, I'm kind of curious, Adil, like, what do you want to build the next couple of years? Because one thing you said in your funding announcement was we're going to go build stuff. So I'm really kind of curious what the next year or two of product looks like for Magic School.

So one, like I mentioned before, I think one of the most important things we're going to do is forward deployment. Our solutions, architects, engineers, two districts and make this deeply, deeply integrated with their challenges that they're facing every day. So that's going to be. And that's not cheap, and that is not cheap. And then what does that actually look like in practice is kind of where the product will be going in the next couple of years. So one, we think that there is a huge opportunity to contextualize the use of both educators and students with the product. So if you think about like a unified knowledge graph for a student's experience with the product, that could include maybe their grade level, their categorization as a student who has, who needs a Certain level of text complexity, because maybe a special education designation or an ELL designation or a language learner. And then all of the tools in Magic School can very much directly support the student in their zone of proximal development. So they're not, you know, the. The material isn't below their grade level or so far above their grade level that's inaccessible, but it hits that zone where it's stretching them and helping them grow in a way that is incredibly valuable to, you know, to the student. And that also ends up being really valuable to the teacher, because no teacher can carry when they're looking at a student and building that relationship and having that connection. They also can't carry the student's last literacy test score, their last SAT score, all of the designations they have that can help them learn better. It's really hard to carry all that with you. So can AI be this partner for the educator that helps them understand the students in context of what they can learn, what they love, deeply contextualized to their world? And then on the educator side, Magic School needs to connect with their world more deeply too. If you had a real assistant assistant teacher, you would first train them on your students, their needs, the systems that they worked with, like maybe LMSs, and then you could be more effective as a teacher. So Magic School is going to kind of get even deeper into all those places to be this truly personalized education assistant for students, as well as a teacher's assistant for teachers.

If we're going to end up in a world in which there is an AI or an instance of an AI tool per student that knows them from, let's say, K through 12, we're going to need to have relatively robust memory and context tools. Because, I mean, I love my instance of GPT5, but it's dumber than rocks and can't forget me telling you I don't want emojis, please, roughly twice a day. And it just can't hold onto it. So I know we're not at a level yet when I can say, hi, I'm Alex. I'm in grade 11. And it knows that in grade eight I had a remedial English class or whatever, but I can see us getting that there. And then you would have essentially for each student, a buddy for really an assistant teacher for the student at that point, because the AI would know. Yeah, lots of out there. It have the context.

Yeah, it would have incredible context and still be driven through the teacher. Right. The teacher still have to ensure that the AI was accountable to the student. This knowledge also ends up being super powerful for the teacher. Like I. I told you the way that the teacher is differentiating the platform or the materials they're creating through the platform currently. But how incredible would it be if it just auto differentiated for you with the knowledge of each of your class periods? And it did it in a way that was even more precise to the students in your classroom. So it ends up being a really great assistant for the teacher as well because it has knowledge of all of the students in your class when it makes materials.

That's fantastic, huh? How far, how far from that do you think we are are in terms of bedrock AI capabilities and also Magic School's ability to take advantage of them and implement them and bring them to market?

Our technology team would tell you.

I.

Think we're pretty darn close. We're doing some really, really cool work. I expect that some of these bits of what we're talking about will be available as early as January. And I think that in the next year it can feel like a wholly different world. Especially for the districts that partner with us, they'll get kind of like the earliest, like testing and access to these things. We have really incredible partnerships with the foundation models who are really excited to support us. It ends up being this kind of.

Do they subsidize you guys at all? Do they offer a discounted access to their APIs? Because you're actually trying to help kids? Because if I was OpenAI and you guys were using my model, I would try to find some way to shave off a couple of basis points. Just because you're trying to help kids.

You hear that? OpenAI? We could use a discount.

Sam, we've met a whole one time for seven seconds. That makes best friends. Come on. There you go. Yeah.

No, we do not. But we have a very sustainable business model that allows us to do this work and great funding. So it's been. We don't feel like these things are outside of our capabilities. And while they don't give us free tokens, they do end up being incredible supporters of US. Anthropic has a beneficial deployments team that supports US.

US.

OpenAI has an education team that supports us. So we feel really close to the model providers and they've been incredible partners.

So.

All right, one last question, Adil, before I let you go. The US is one small slice of the world and I think we're all kind of seeing, you know, peak college student, and that's going to trickle backwards into K through 12. So we're not heading towards, you know, opening tons of new schools in the U.S. i don't think. Which makes me curious when you're going to take this abroad and which markets are going to be the most most fecund for the product. I mean, I immediately think of similar ish schools in perhaps Europe maybe, but what does the world look like to you?

So we will absolutely take the step to the world. We have a massive amount of usage internationally without any marketing or direct sales. We do get inbound customers and partners from international schools and places all over the world. One I mean it is also kind of the staggering number of US adoption. We'll find countries when we look at our data sometimes. And like the UA, we have reason to believe that like 80% of educators in the UAE are using Magic School.

The UAE is not the largest nation in the world, but 80% of any nation. That's insane.

With no direct marketing, no presence there whatsoever.

And that's why the $45 million Series B was led by Valor Equity Partners and Bain, Adobe, Atreides and Smash all kicked in cash. Because everyone loves low cac.

Yeah, it is one of this incredible things people ask us sometimes like how do you, how do you acquire so many folks? Well, we have this incredible product and legion of educators who want to support each other and that ends up being a really big superpower.

All right, deal. Well, I'm, I am I'm not only now less concerned about the state of K through 12, I'm actually more bullish about it. So gods be to you and the team and whatever you do put out in January, send me an email. We'll have you back on stock, but it sounds like you have quite a lot cooking. Thanks for coming on.

Lots of good stuff coming. Thanks so much for having me.

Alex, your startup needs custom software. By building your own secure production ready apps is hard, right? And not every founder has the expertise or time to get started. Well now you don't have to just use Zeit Z I T E. Make whatever you need from idea to a working app in just minutes, easily all on your own. Plus you can fill it in with forums, apps, databases and other automations that you need as part of your product. So many of the no code and vibe coding tools are just flashy demos or they work on legacy systems that take so much time, much time and effort to learn, let alone master. But Zeit is both easy to use and powerful enough to build out the apps you want and it quickly connects all the other tools you're already using. For example, earlier this month they added the ability to bulk create and update records. So it's easier and faster than ever to design data heavy apps that are constantly syncing between systems. That's the kind of attention to detail that sets Zeit apart. Start using the number one AI powered business software generator on the market and we're going to give you 50% off your first project. Go to zeit.com twist to get started. That's zeit.com twist for 50% off your first project. We like to do founder questions here. We have two good founder questions.

Let's do it.

Thanks for having me today tonight. I really appreciate it.

Okay, so number six, we got a founder question I think is pretty interesting.

Yeah. So one person was asking about how long employee shares should take to vest. They were looking at an offer from a company in the hardware realm and they were offered 0.023% of the company over a six year period. And so the question is, is that a standard term that employees should be willing to accept or is that something you might want to have only for founders that are raising venture capital money?

Your take six is definitely abnormal. I would say market standard is a four year vest is fairly typical. There are companies that diverge from it like I think probably the most the extremes on one end you'd see six when you could could see like even I think like the stripes of the world have every year you get an extra tranche of, of RSUs that vest over a 12 month period which is a one year kind of vest. And every year they kind of re underwrite where you are.

But they're kind of a public company. It's a little bit different. They're essentially offering the, the Google rsi.

No, but it's a good point. If you were going to come to a brand name company they can extract a longer vesting period because the shares are worth more. Right. I mean that makes, makes logical sense. It's not as speculative as a earlier stage company. To your point Alex, what I, what.

I would say ultimately is in some sense a six year can be good, right? If you believe the value of the shares are going to multiply. Presumably if that stock is going up 2, 4, 10 times what you're going to be vesting into in the six year is going to be worth a heck of a lot more. And so you may be willing to elongate it. But I think like any negotiation you need to look and say like do I believe in this and is this enough? You know, if you look at the total value of the compensation over the next year. Cash, stock in kind benefits, does it meet your bar or not? And so while I'd say it's unusual, sometimes it would make sense to, to take that kind of an offer. But if I were friends with this person, I would say like it should meet this level for you. And hopefully you believe, you know in this company enough that you're willing to wait for six years and want to be there for, for, for this period of time to see where this goes.

Six years is a long time. Jason, do you recommend the company is going through the launch accelerator, offer employees four or six year vesting schedules?

This is one of these like quirky things where founders who are a unique species in the world, I certainly was when I was starting companies. You want to reinvent everything. So you look at a vesting schedule, you look at shares, you look at the convertible note and you're like, oh, I can innovate on this. So. And I would do stuff like that and I would like over negotiate like totally meaningless things early in my career. And you know, at some point I think Ruloff said, you know, my best advice over at Sequoia, I said best advice, just keep everything standard. So then when people ask you to do weird things on the other side, you can say I prefer standard and everything I do is standard. So if you're not Elon or Carlson Brothers or somebody like really unique in the world, maybe keep it simple and focus on the product innovation in your product, not in the legal documents. A great example of this would be Bezos and Amazon. He went with this like 5, 15, 40, 40 I think is what it was where you first year you get 5%, next year you get 15% of your equity grant and it's backloaded. What this does is lets the company, if they really value their shares, say, you know what, we assume some number of people aren't going to make it to year three when really the magic happens and you tip over from being a drag on the company where we're investing in you to you're actually providing massive value for the company because, because you've learned how to be an executive here, you've got executive training, but you're Bezos, you can get away with that. And also it's a bit of a test because if somebody argues that on the other side, Eric, you would be like, okay, why does this person want it? The biggest tell I ever did was like one time in my career somebody told me it was standard to give people like six months severance or three months severance. And I did it and like I gave them three months severance and literally the person became a disaster six months in. And I was like, oh my God, they're working for their severance package. And so then I just told everybody, you know, in startups you don't get like a, there's no such thing as severance in a startup. If you do a riff or a layoff, sure you can give people severance, but I'm talking about like if you leave and you're or you're fired, you don't get severance. I mean that's like for Tommy Mottola at Sony or like some, you know, I don't know, Barry Diller at Columbia Pictures or something like that's something a major executive can extract. So anyway, it's a really interesting question. I would keep it standard just so that you can focus on other things. What I do, like, I don't know if you've done this Eric, with, with like high performers, but somebody's a high performer. You just layer on a parallel grant. So like you're like this person's killer. Like year two, you're like giving you another four year grant. So now you have them for six. Yeah, you know, you have just overlapping ones. And then it gets really costly to leave if you're that employee because you got double grants, you've done that.

My last job, Jason, they, they layered on extra merit comp on top of my salary. And I was like, well, I guess I'm never leaving, but sometimes people do leave. And a question that I wanted to ask today, Jason, is what happens when a co founder exists? It's a company.

Well, before we get to that one, I wanted to ask Eric if he's done the, the layered trance.

First of all, yes, and I agree with your advice. I think innovating on the business model, what makes you different, really good idea. In general, I think what you and Roloff and everyone is saying is, is, is right. Keeping it easier just reduces friction to people getting to say yes. And frankly too, there are some people in your business that like they are not twice as valuable or four times as valuable. They will be 50 times as valuable. And once you find it like hold on, invest and spend against it. Because frankly, one, you should do the right thing for people who are just care at the level and can have the impact. But second, it's just very clearly good business. And so that's where I think you should hold the equity to invest in those situations.

We have a friend from the Gamma Pitch competition. I'm really excited to bring up, Chris Cannetti. He's the founder and CEO of a company called on the Fly Energy. And Jason, I was initially skeptical until I realized what they're building and then I became an enormous fan. Please everyone, welcome to the show. It's Chris. Chris. Hey, how you doing?

Good, how are you?

Welcome to the show. Chris, why don't you go ahead and put on your presentation here. Powered by Gamma, great way to do AI presentations that look really sharp and crisp. We'll have you pitch us here and we'll give you a little feedback.

Hi guys. I'm Chris Kennedy, founder of on the Fly Energy. And I'm here to tackle one of today's toughest emerging challenges. We talk about. We take electricity for granted until the lights go out. Our grid is fragile and a split second fault can knock out critical infrastructure. And AI data centers are making this worse. GPUs swing from 0 to 100% in milliseconds, destabilizing the entire system. Today's solutions all fall short. Batteries are expensive, degrade and can burn. Diesel generators are slow and maintenance heavy. Take for example last summer in Virginia, a lightning arrester failed, forcing 60 data centers offline in an instant, nearly triggering a cascading blackout. Industry and I alike need resilient energy storage. And that's the gap that on the Fly Energy fills. Here's our solution. The rotor Seed is a 10 kilowatt hour modular flywheel that installs like a backup battery, charging, storing and releasing energy. Rotor seeds stack into rotor pods and scale into the rotor farm, capable of delivering megawatt hours of distributed protection for everything from a single server to the local electric grid. On the Fly Energy is built for America's backbone. Where uptime is non negotiable. The Rotor C delivers instant fire safe backup and conditioning for sectors that matter most with over two decades of life. And it's built in America to strengthen energy independence. Now we're targeting over a $100 billion energy market consisting of over 100,000 US facilities from AI data centers to manufacturing hubs, all where batteries fail. Diesel's lag and uptime is priceless. Needing to capture just 1/10 of a percent to get to 100 million in market share. On the Fly Energy wins where it counts. There's no fire risk, no degradation, no 30 second lag or upkeep like diesels and scalable beyond capacitors for real long duration protection. It's simply a better spin on Backup power. The rotary seed units are offered as a monthly subscription and collect revenue on demand shifting and energy delivery. Additionally, we plan on offering utility support and active energy optimization, all with a straightforward upgrade path that grows with the customer's business. Now, what makes this model powerful isn't just lease storage. It's the network functioning as a virtual power plant to help regulate the grid. We don't compete with the utilities, we help power them. Path to market. So far, we've built our proof of concept to validate the core physics, design and functionality. In early 26, we'll conduct MVP trials at beta sites. In 27, we plan a small commercial launch targeting telecom and small data centers. And in 28, we expand into large AI data centers and full grid scale storage capacity. I'm Chris Canetti, the hands on founder of on the Fly Energy Inc. And for nearly a decade I've built automated manufacturing solutions at companies like Tesla and other startups. I still run a small machine shop out of my garage and my teammate Justin Osborne brings deep electrical controls expertise. Together we're building on the fly Energy with AS9100 level standards and automation from day one. Because engineering quality and manufacturing scale on the fly energy, we're tackling $100 billion market with hardware that outlasts batteries, outpaces diesel, delivers protection where America needs it most. The grid's future doesn't burn, it spins. Thank you.

All right, I'm a little confused as to what the physical product is. I understand the problem.

Yep.

People need extra energy. I understand there's a flywheel involved and there's like a cutout of it. What is this physical product?

So the physical product. Yeah, so the physical product is essentially a battery energy storage system, but with a kinetic energy storage rather than chemical. So we're storing rather than lithium batteries we're storing in kinetic energy form.

Explain what that means to somebody who's not an expert in this. You're basically putting batteries into data centers, but they're different types of batteries essentially.

So rather than a lithium battery, which stores energy in chemical form, we're storing energy in kinetic form, which is a flywheel that spins. And so you take that excess energy and you put it into the form of inertia and you spin a disk around really fast and then you can pull power from that or buffer power into it. Whichever way you need to level out these spikes that the AI dentistry.

So Chris, a question about that, because there's always going to be some friction in stored momentum. So how much is. If I had this wheel, which appears to be spinning very quickly in my data center just waiting for me, how much energy do I have to put into it to keep it at max kinetic energy potential?

So depending on the size of the system that you're using, we see about a 95% return currently from the power that you put into the power that you get back out. And that's in our.

Sounds very good to me.

That is pretty Good.

Okay.

So 95% of the energy that I put in, I get to take out. And because it's a spinning. Is it iron? What's the actual thing that's spinning?

It's a carbon fiber composite.

That sounds light. Wouldn't you want something to be heavy.

And it is light. But so where you wind up storing, because the way that energy stores in kinetic energy is it's a to the power math equation, so that every RPM that you can gain is actually a significantly larger amount of energy that you can store. So if you can spin faster and not have the material weakness that say, steel or something heavy would have, you can actually store significantly more than you could in just a heavier mass.

And how big is this rotor seed, that 10 kilowatt hour module that you're describing, if I want to call it in my hands?

Yeah. So the current version that you're. That we have is basically an 18 inch diameter by an 18 inch long length. And that makes the 10 kilowatt hour base unit.

That's not very big, does it not very vibrate?

It does not vibrate. It is on magnetic bearings.

So you can just, you can stack.

These, you can stack these into something say like an ISO shipping container or another form factor like that that's easy to distribute if need.

So as long as they're not moving, there's no pushback from the wheels motion itself. So. Okay, that's. That's pretty awesome. What does it cost to build one.

Of these right now? The prototype that we're at right now is sub 10k and I'd like to be significantly lower than that. Looking at what the business model is going to wind up being for ROI, potentially being in the one to two years.

Okay. And then for a 10 kilowatt hour battery, what does that cost today?

So a 10 kilowatt hour battery can cost you anywhere from $10,000. If you get it really cheap and you're doing, you know, kind of getting a little scrappy and you're buying 18, 650 cel off of eBay, you can do it for probably a couple thousand dollars and if you do it the legitimate way and say you're buying like a Tesla Powerball, that's pretty close to a similar energy density, you're probably in the 10 to 15,000.

Okay.

So this is interesting. I asked producer Claude to explain to me like a 10 year old because, you know, I just. Still not getting it. I mean, I was getting it, but it's like spinning a top. You have this.

Yep.

Thing that you start spinning like a top. So you put some energy into it, it spins. You don't need to keep adding energy to it. It keeps the spinning going through magnets, etc.

Yep.

Then when you need it, it slows down and slowly lets that energy back out. So they use this in space and on roller coasters and buses. This is the other use case for this technology.

Yeah. So the original, the original bit of this technology, it did exist in buses, I believe, before the trolleys and SF existed, as my understanding that there were large mass spinning flywheels that existed.

Wow. So this is like you're bringing back an old energy source with new modern materials. So genius. So I think, like, as an idea, it's extraordinary because people need more energy and they need more backup energy. So data centers have a ton of money to spend, so that's the best place to go right now. And there's cheap energy available when you want to fill these up. Right. At certain times during the day. So you can do that arbitrage. You can fill them when energy's cheap, Correct?

Exactly. You can fill them when energy is cheap. And then you can also do this little bit where you're getting to condition these AI power loads that like to spike and then drop loads. Every time someone wants to generate their. Their image, the AI data center goes up. That usage on the grid goes up. And the grid is. It was really built for a pretty continuous load system. So we can kind of do this as the energy arbitrage with the virtual power plants and then you can also condition the power that's going into the building. So let's say the AI data center's load is actually all over the place, but you can basically make it so that this is the buffer zone between the AI data center and the grid, so that the grid sees the constant.

So you saw this Jason with base power in Zach Dell. Oh, yeah, sorry.

No, no, go ahead, finish it.

Yeah, you can.

Oh, I was going to say we just had base power on Chris a couple of. A couple of shows ago and he was discussing how they were putting batteries in front of houses in the Grid and how that would also dampen the spikiness.

Exactly.

Overall grid that we live with. Yeah.

And being I'm in California, I can't do kind of the same model that they were doing. California's energy grid is a little different than ercot.

So we're talking about being Texas's. Yes, being Texas's, which is completely deregulated and you can. Can add and take energy out in a much more fluid way.

Yes, they're much more free.

Much more free. So I guess. Would this also work for air conditioning in places where it's very hot? So you go Mexico City, Dallas, Houston, Austin, Phoenix, all those kind of places where you have really 100 days of 100 degrees or hotter. You have these things attached to buildings. At night they fill up. And then during the day, instead of hitting the grid, when I think it's a 5x price difference, you could power all the air conditioning.

You can do exactly that. And something that if you weren't what there's a model like that or you could do something that's following. What's the duck curve is basically we get a lot of solar cheap energy right when solar is peak and there's really not a whole lot of residential load at that point. And you get a really tail end. The tail end of the duck that curves up. That is a high demand speed peak. Right when solar and well, generally solar starts to taper off. You can kind of play that arbitrage game where you're going, where do you. Where's the. What makes the most sense to charge up? Do you charge up at night when the baseload's steady and the peak's really low? Do you charge up when solar is really cheap in the middle of the day and there's not a high demand? And you store that for a few hours and then you can dissipate that when everyone's coming home and turning on their ac. And then across the industry there's uses for this all over the place. One of the other sectors that I'm looking at, but we're not focusing on that because we're focusing on the AI data centers just to try to get ourselves into it. You look at all these EV fast charge systems that are being installed places, it takes 12 to 24 months to get these utility and permits set up to be able to do these fast chargers. And some of them don't have the support to be able to do 250, 350 or 500 kilowatt chargers. So you can do kind of this portion, where even a kinetic energy storage system, or just a battery energy storage system sits between the grid and this fast charger. And your, your technical site could only have 100 kilowatt hour hookup, but you could have multiple 150 or 250 kilowatt hour chargers because you can charge this and level out that peak. You're basically doing the same thing that you're doing with the data centers, trying to make this a level energy draw for a longer period. And then when you get a load, say a vehicle showing up and wanting to fast charge, that's your spike, but you're buffering it with the battery energy storage so that the grid doesn't see it.

And if you.

That's the duck curve.

And here's the duck curve, as you can see.

Yep.

You know, in the morning, 9am Then it goes 12, 3 and 6, the energy usage can go down. And you're like, wait, why is it going down? Well, people return home, solar stops working and you pee.

Right.

At the worst time, at the worst time. So the sun goes down, everybody returns home. Alex. They decide to do a load of laundry. Maybe they put the air conditioner up or in the, I guess they could also put their heat on. Sure. And it looks like a duck.

And that's, it's a brilliant analogy, but also it just goes to show how complicated the power question is between baseload power, variable power sources and storage, but between exawatt storing industrial heat, base power, doing batteries at the house and hydro stored, doing compressed air energy storage. And now we have on the fly energy using flywheels. Jason, I'm actually a little bit optimistic right now.

You know, here's the thing. We talked a little bit earlier about truth and the search for truth and how, you know, generative AI during this weekend protests or anything during the debate over the southern border, people can use data sources that aren't true to try to manipulate and, or fight for their position or their tribe, whatever it is. And with energy, one of the interesting things is the data's the data. Chris. Right. There's a price of energy that you cannot fake.

Yep. And it's well metered and it's reported at most places easily accessible.

Literally, with ERCOT, you can pull up ERCOT's data in real time. Alex. Because it's a real time exchange of energy. So when I'm interviewing Chris Wright and we're talking about the administration's love of clean, beautiful coal and solar is intermittent, and then Elon is saying, well, solar is the best thing in the world because you can just build it and it's cheap and clean and it's free. And then other people are like well what about geothermal? Or you haven't it used new technology or a new approach like base power or here what Chris is doing, you can just start to pull together reality by just looking at the data. Chris, it's become very political and very sharp elbowed. What's your take on overall? If our country needs to invest a massive amount of money in energy and you were in charge, I make you the energy secretary. How would you think long term? And when I say long term, 10, 20, 30 years for America and let's say you have a trillion dollars you're going to deploy over those 30 years just on a percentage basis, where would you, hey, get us the short term energy, get us the midterm, get us the long term. How would you advise doing it? Obviously you have your own book and, and you know that's a portion of it. But just broad strokes, what should we be doing with energy? I'm sure you think about this all the time.

I, I do think about it all the time. And kind of the broad strokes, the way I see it is kind of like you mentioned. So short term we need a solution right now, whatever that looks like, whether that be natural gas, peaker power plants, I don't believe that coal is the solution. Maybe we're looking at other, other forms. There's all these solar, wind, renewable things. And if you can have a way to sort that's kind of where on the fly energy is falling in, you get to, to get to play on that game. But then on the broad stroke as we're going, America needs energy independence and how do we get there? We really need to be self reliant. And if that means we can generate all of our energy as we need it and store it, I mean that's the biggest thing. If we can have renewables and be able to take out the downside of renewables, that is production, fluctuation and unpredictability depending on where it's at, that and being able to store that and being able to redistribute that across the grid to where you need it, that's kind of the solution that I see. So if you could do other renewable options being wind, solar and then kind of baseload with nuclear power.

Yeah.

You'd be able to cover that entire grid and be able to be essentially entirely energy independent as a country.

Okay, I have a question about scale here. Because in your deck that we went through, which by the way, very nice. Appreciate that, that it looks like a small device, 18 inches by 18 inches spins inside of it. Very, very cool. Why not make a really big one? Like, I mean, huge. Like, why can't we make one of these at a city scale, for example?

So the physics of it get really limiting. So kind of what has advanced in the last 50 years from the heavyweight flywheels is really a materials question. And where we've come is these composite high tech aerospace materials. And what you're getting is a higher strength option here so that you can spin it faster and faster. And one of the limiting stresses here is basically how fast the surface of that rotor is actually spinning. So the larger your diameter you make this, the faster the surface speed actually winds up going. And because that fact that higher surface speed is there, there's something called hoop stress. And I'm hoping Claude will pop in here and give you a little definition. That hoop stress is basically a function of the tensile strength material that you're making it out of on the outside diameter. And it's basically what's trying to burst this thing. It's basically why like a hot dog, when you cook a hot dog, why a hot dog splits down the seam and doesn't split in half the other way.

So what if we put a large iron rod, like an old wheel, around the edge of the. I'm kidding with you, but like, my question is, why do we have to make this in small units? Why can't we just make large ones?

Well, we can make large units. Yeah, you can make significantly larger ones that are. Power density actually works out kind of similar, but you'd get at other scale factors where you have lower. Lower number of components that go into it.

Okay, well, I'll just say I like the small ones, I like the big ones. And I think, you know, AI data centers is a great place to start because it'll attract investor attention. There's a lot of money there. But also, like, I want one for my house because this sounds awesome. Like, I do as well.

That's where the first prototype's at.

Chris, what's the story here with funding? I mean, I'm not an investor in this company. I feel like this is like going to be. If you figure this out and it actually works, you raised a lot of money so far. You're funding it yourself. You were at test.

I'm funding it myself right now.

Now, really?

Yeah.

That's my kind of founder. What's going on here. Maybe I can.

I.

You gotta. Can you save a slice for jcal? I might want to get in.

I will save a slice. I'll put a little sticky note right on the. The cap deck for you.

I'm not saying like I have to have the whole pie, like if you order a giant pie, I'm just. You can even cut me a half a slice is better than no slice. I love what you're doing. This is great. Let's stay in touch. And thanks to our friends at Gamma for making great pitch software and then helping us find great founders like Chris and great startups. Really well done. You can build amazing presentations in just minutes using Gamma. And they hosted this incredible pitch competition which, Chris, I don't know. You might be the leading.

You might.

You might wind up taking this down and winning.

Hey, I'm love to hear that.

Oh, okay. Maybe I'm intrigued. All right, Chris, great job and we'll talk soon. Let me know when you're in.

Thank you, guys. Yeah, we'll do.

Let's get some. I'm going to take you out for some barbecue and try and weasel my way onto that cat table. Great. Great job, Chris. I'll drop Chris off and if you're feeling inspired, head to Gamma. App. App G A M M A dot app so you can start working on your idea and you can change the world like Chris is doing. And maybe we'll see you on a future episode this week in startups. Well done.