Tony Wan: AI is raising the bar – is education ready?

What does the rise of AI mean for schools, startups, and the skills students actually need? Tony Wan – Head of Platform at Reach Capital and former co-founder of EdSurge – has spent over a decade watching education technology evolve. Now he's watching AI reshape the landscape faster than anything before it.

We get into why "taste" – the kind you can only build through practice and productive struggle – may be the most important differentiator left. Tony also shares how his background in history and journalism shaped his thinking, and why he worries about a generation that may never develop its own sense of what "good" looks like.

aiEDU: The AI Education Project

• aiEDU.org
• linkedin.com/company/aiedu/

Transcript

Tony Wan’s EdSurge Origin

Alex Kotran 0:04

Here with Tony Wan, the head of platform for Reach Capital, formerly co-founder of EdSurge, one of the leading publications in the K-12 and education space. Was it post-secondary as well or just K-12 focused? Post-secondary as well. Okay. Yeah, K-12 and higher ed. And so you were basically not just in journalism, but running a journalistic outfit for was it a decade or?

SPEAKER_02 0:25

It was almost 10 years, my tenure there. I started in 2011. Around September 2011. And I left in March of 21.

Alex Kotran 0:36

So you were like mobile and social media had just started to reach a little bit of maturity. Would you say that was sort of like the second wave of ed tech? Like how what was happening in ed tech at that time? I know that wasn't the only thing that EdSearch was covering, but to the extent that we can maybe draw some parallels, if there are any, uh, to what we are experiencing today.

The Second Wave Of Edtech

SPEAKER_02 0:54

So I think what uh what brought me to EdSurge was meeting meeting Betsy Corcoran, the CEO, at uh at a meetup, uh, at an SF ed tech meetup. And this was when you know meetup.com was you know the thing, right, for organizing all the local communities. And this was something that was happening for uh just for the SF scene and crew. And one of the people at uh one of the people on that programming, on that panel was uh Alan Louie from Imagine K12, which is one of the first ed tech accelerators uh in the game. And so if you ask me like what you know what phase of the ed tech evolution, you know, I would uh I kind of landed in, I would say maybe the second, right? I think ed tech certainly like preceded you know the the 2010s. But I think what felt new at this space and what was exciting was a lot of former Silicon Valley people, you know, from the big tech companies kind of spun out and started to start building stuff and technologies to really help bridge the gap between the technologies that they were used to seeing and building, you know, in the course of work at you know the Yahoo's, Microsoft's, and Googles, and and what they were seeing for the kids, you know, in in the classroom. And so out of that came a lot of efforts to really apply some of the things that they learned to build ed tech software. Um, and then along with that was the uh investment capital, where some of these people had you know made decent money and wanted to like put some of the capital into like encouraging um ed tech startups. So we started seeing accelerators like Imagine K-12 uh coming into the scene. Um and then we started seeing you know uh organizations like New Schools Venture Fund with uh you know start out their their local seed fund, which was really the uh the origins of Reach. So I think the combination of uh capital, innovation, and also the beginnings of the distribution of uh uh availability of broadband and one-on-one into schools. I think that was a starting point you know in the early 2010s, but um I felt that was uh you know the cusp of this of the of the second wave.

Alex Kotran 2:59

So I do want to dive into it. What does head of product ahead of platform mean for those who are not familiar with? That's a good question.

SPEAKER_02 3:06

My mom asked that, and I'm not sure she even knows what I do. Um so depending on what VC firm you ask and head of platform can mean a lot of different things, it can it can be like six different jobs. Uh at Reach, um, what I do a lot of is provide a lot of portfolio support uh for our founders uh after they join a portfolio. So Reach Capital, we are an early stage uh ed tech venture firm. Um we started in ed tech, but now we're expanding more into the workforce and healthcare space. But we are um basically investing across learning, health, and work, all under the thesis of we want to back founders who ultimately are building things that help elevate uh you know, elevate human potential and help uh kind of ignite the purpose, you know, make life more meaningful. So reach we started in ed tech. Um that was really the bread and butter of our of our origins. And you know, in the past couple of funds, we started expanding more into the workforce development because there's a lot of overlap, actually, if you think about it, between you know K-12 and hired ed tech into helping adults upskill and work. And now we're doing more into health because health is also, as we realize, um, as as you know, where we believe is a precondition for helping people live you know uh healthy, thriving lives. And so header platform uh at the early stage of reinvesting, some a lot of founders uh need support, whether it's for communications, marketing, thought leadership, whether it is for uh help uh open uh opening doors, introductions, getting to meet um, you know, some advisors and decision makers on the ground uh in in schools and districts. Sometimes it's for hiring, right? I think hiring is often one of the biggest pain points for early stage founders. And so I'm organizing a lot of programming, a lot of resources to help connect the dots for founders uh at the at the early stages after they join a portfolio. Um so that's a good big part of my job. Uh the other part is you know, given my background in journalism and writing, I you know, I also help with a lot of the content of thought leadership on behalf of Reach. So we publish blogs, market reports, newsletters pretty regularly. Uh, and I'm essentially either the writer or the house editor for a lot of the thought leadership that we that we put out into the world.

Alex Kotran 5:23

This was the easiest interview to prepare for because you just sent me like five pieces that you had written. I was like, okay, I have a sense of your perspective on this. Um you're cross-reach capital. I think that's important because sometimes you talk to someone who's who's at a VC and they are a partner with their specific portfolio of companies that they have invested in. Um, but you have a little bit more of a bird's eye view. This is a bit of a loaded question. I'll provide a little more context. What's your take on the AI, how the AI space is unfolding? And and what I'm specifically asking about is there was like exuberance over the last 18 months, two years really, of companies that were like, okay, let me like a you know, build some kind of usually a wrapper, but build some kind of implementation for one of these frontier models. Um we can focus on education, but feel free to also include anything from health or or future of work as well. Vibe coding and they're getting to these MVPs very quickly. And some companies like Magic School, being an example, are achieving like massive scale kind of right out of the gate because it's just this is sort of sexy and new and people are like curious. Um and but then when I've the VCs I've talked to are also like it's very hard, like very little of it is sort of defensible because anybody can actually create a tutor. Um, in fact, you don't even need to create a tutor, you could literally just prompt Gemini or Chat GPT and just sort of like have it, you know, take a little bit of pre-prompting. Right. And so I I've I've heard a lot of hesitance about or just uncertainty about what does it look like for companies to go from you know the use case to a sort of scalable, defensible, you know, whether it's a billion or hundreds of millions, but like the sort of like you know, series B C D businesses.

SPEAKER_02 7:02

Um is that Yeah, certainly like there is I think the fear that, hey, is OpenAI just gonna build this? I think that's a very credible uh and very real um concern or at least consideration that you know you know that that I think informs uh that makes investing in this space kind of hard. Um but I think what I have come to learn is that similar to how take the past decade, like like Google, I think you could like in the past, you know, back when I was starting at Edsearch, you could also ask, hey, is Google just gonna build this? And there were some things that Google did build, right? Like Google Classroom, which was, you know, in in in for all intents and purposes, very very powerful and serviceable LMS. But Google's not gonna want to build everything that you're gonna want, you know, in in education. I think there are a lot of nuances in how the systems, you know, you know, all the systems work, you know, systems being both the political political systems and also like the technical systems, right? All the all the different pieces of software and data standards within educ that are very unique to education, that I am not sure if you know someone like OpenAI will really want to devote a lot of resources into you know building for these use cases. And so I still think like there is a lot of nuances when we when it comes to learning that I'm uh and user experience and like just understanding the day-to-day challenge of what it is to be a teacher, the classroom management, the relationship skills that you build. Um, you know, I am not sure if OpenAI, you know, or or anthropic will always want to, you know, will be able to want to build for all these use cases. I'm not I'm not sure if it makes for like a business sense for them. And so, you know, I think that when we look at different uh, you know, when we try to evaluate companies, I think these are, you know, these what we call earned insights into the problems that you're solving for that um are that that that really that help differentiate one like from from the crowd. To magic schools, yeah, that's the example. I think you know, Adil is a former you know teacher and a school leader. I think he very well, you know, very deeply understands what the workflow of a teacher is like in the day-to-day and the parts that suck, you know, the parts that you know you could save time for. Um and so I think part of that, uh, I think like using that to inform development of your product is you know, uh is going to be a key different uh key differentiator. I mean, I will say that it does feel like a lot of the a lot of the VC firms kind of are all like betting on their own uh you know, AI teacher co-pilot horse, you know, at the uh in the race.

Alex Kotran 9:38

Like BRICS, you have class AI, you have Magic School.

What A VC Platform Lead Does

SPEAKER_02 9:41

Yeah, you know, we also have uh Curapod in our portfolio, Teach Share, uh Teachy. Uh and so um look, I mean, and I think to your second point or other point about like different differentiation, you know, like tech like technology, yes, you know, I think that's uh that's a question mark. And I think we have AI, you know, we have um, you know, we try to tap into the AI experts and founders in our portfolio to help kind of do diligence and vet the tech. I also do feel like AI has really raised a bar for time to deploy and time to uh yeah, time to build. And it's also raised the bar for the metrics, right? How fast can you get to a million users? How fast can you get to you know$10 million? And I think um, you know, just metrics-wise, if we're trying to be quantitative about it, I think um those are the new bars for uh that can help inform like uh you know differentiation, defensibility. Um and the final point I would say about defensibility is we we we want to see if what is the theory of change you know that informs a product. Like is there research, and you know, uh is there research or pedagogy that backs into how you're building your product or how you're building the user experience? Because I think that is still key and and important.

Alex Kotran 10:55

And that's one of the things where I'm you know not sure like a generalist builder will be able to understand the f the full nuances of so this idea of raising the bar, uh you sort of you're you're alluding, I think, to um not just vibe coding, but the there is a velocity in startups that just realize anything that you know, startups always moved fast. But you talked, I mean this is like the latest YC batch, they did the survey, I'm sure you've seen this, is like a quarter of their batch reported that almost 90 95% or more of their code base was entirely written by AI. And Scary Tan was basically like this isn't just the way you code now, this is sort of the future, and if you're not vibe coding, you're basically gonna be left behind. Um is that sort of is there anything else baked into when you talk about you know sort of faster getting to an MVP faster, getting to sort of some of your initial KPIs faster, is like are there other aspects of AI or even just maybe stuff that isn't AI that's lending, it's lending you to sort of having this sort of different expectation?

SPEAKER_02 12:01

Well, I think the advantage of building faster is that you also learn faster about user uptake, user adoption, are they coming back to their product? I think back uh, you know, the the the first early waves of AI, you know, let's call them AI wrapper tools. I think, you know, we've seen that many of them actually have pretty low, you know, seven-day retention rate rates for like 30 days, like after 30 days, right? What's the what's the percentage of people who like still come back and use this tool? And so I think that what you learn from building faster is that you you know you're able to like get get a faster sense of like what are the what are the parts of your product that are actually sticky that people will keep come back, uh will keep coming back to and continue and continue to use? Because uh I will I will say, you know, a lot of the these AI wrapper tools, um, retention is one of the biggest uh you know uh drop-offs and and and challenges to make them investable opportunities.

Alex Kotran 12:59

Do you is Magic School in your portfolio? Yes.

SPEAKER_02 13:02

Okay.

Alex Kotran 13:03

What are some examples of use cases where you've seen the best metrics on usage where people are coming back? And is it like very narrow? Are these sort of the more broad-based teaching support tools?

SPEAKER_02 13:15

Yeah, um, we invest in uh have several teacher um you know tools that help teachers generate lesson uh lesson plans um and also to lead uh classroom, classroom activities. One of uh one of them is called Curapod. It's um you know kind of like a little bit like a NearPod, but with more AI uh powered generation and exercise um creation uh um features that help really teachers lead you know fruitful dialogue and engagements uh you know with with students and have students kind of interact with AI feedback in a way that's very um really encourages more like curiosity, Socratic-based dialogue. And so, you know, we're seeing you know, teachers continue, you know, well, I mean, we're seeing, you know, by word of mouth, you know, hundreds of teachers, you know, sign up uh uh you know every every day. And we are also seeing that uh you know teachers are continuing to use it, you know, like week over week. Uh and it's becoming really a more ingrained part of their uh classroom teaching instructional experience. I think a lot of that is that in the in the in the process of having teachers build their lessons, you know, it's really incorporating, you know, some of what we know to be you know good pedagogy to encourage you know classroom discussion, to encourage students to reflect, to encourage students to actually push back actually on some of the uh you know some of the AI uh feedback and some of the things that AI generates. And so I think that you know, baking in that kind of what we know is, you know, makes for lasting, impactful learning experiences, learning that sticks, making that into the product has been something that's been, you know, we're we're seeing bear fruit.

Alex Kotran 14:58

It makes sense. And I I think I worry about teachers becoming over reliant on AI tools. Um I think just skills deterioration in general is a concern.

SPEAKER_01 15:08

Yeah.

Alex Kotran 15:08

But if you were to ask me like what does the future look like, it basically is a sort of world where we solve one of the biggest challenges in education, which is students don't feel like it's relevant or it's not engaging. And um AI can be that's like a very good use case for AI. Whatever you're teaching in math or social studies, like give me a two-minute surprising story or a funny story to start my class with.

SPEAKER_02 15:31

Yeah.

Alex Kotran 15:31

Um, you know, take this homework assignment where students are just writing an essay and turn it into something that's a little more engaging and project-based. And um I agree.

Reach’s Thesis On Human Potential

SPEAKER_02 15:41

I think that like one, I think what AI is not really gonna solve for by and itself is the motivation, right? Like uh issue. Yes, you know, like there are some studies that show like you can improve, you know, learning outcomes, academic achievement, yada yada. I I think that well, I'd like to actually get dive a little bit more into like the samples of of those studies. But I think ultimately when it comes to motivation, intrinsic developing intrinsic motivation, passion, and interest, I still think you know, a teacher, you know, the teacher still remains, you know, central in you know, making those connections between you know the content and a kid and and and you know a child's personal interests. I think AI is great at brainstorming all kinds of ways to apply like an abstract like abstract concept concepts like you know, geometry, right? But you know, you know, builds, you know, come up with you know creative exercises that apply those in ways that are really interesting to whatever a kid may be interested uh in at the moment or or personally, like so. I think definitely I think AI can help like assist like teachers in you know helping to create more um you know experiences that really develop a kid's intrinsic motivation to to learn uh and and and to ask you know good questions.

Alex Kotran 16:57

So I mean this is the uh million dollar question. Um another question is the the YC survey, does that is that out of line with what Reach has seen, or is that sort of roughly are you seeing the same thing basically that like your founders are heavily relying or just just leveraging AI, I assume primarily vibe coding, but maybe there are other ways as well.

SPEAKER_02 17:21

I think they're definitely leveraging AI. I don't know specifically how much of the code is, you know, their their code base is being written by by AI right now, but uh you know, I just think across the board, right? Whether it's a you know publicly traded companies writing these memos about being AI first or even you know privately for you know for smaller startups, you know, I think you know the productivity and efficiency gains that you that we are seeing uh with AI for building products is um it's becoming an expectation now, you know, I would say uh for our companies. And so, you know, we you know we're not you know, we we would we want our companies you know to be you know to be to be able to be uh deploy these tools in ways that are more efficient and um you know push things out uh at a speed that's up to par with what's happening today. Um whether that's vibe coding, whether that is doing you know re uh research, whether that is doing um you know communications and mass marketing, right? I think that there are a lot of ways in which AI is helping to improve, you know, just you know, some of the basic business operational efficiencies.

Alex Kotran 18:26

And would you say uh are you seeing teams uh getting smaller? Uh are they are they hiring more people because uh maybe there's like the jobs they will create. Like how have you seen like a noticeable impact on sort of the size of teams relative to their stage?

SPEAKER_02 18:42

Um, you know, like we we have invested in a couple of teams from this latest YC badge, and they were, you know, they're pretty slim. You know, they're they're two, you know, two like two co-founders, three co-founders, and they are just able to, you know, lever, you know, they're they are leveraging AI to do a lot of things that you know previously may have taken maybe half a dozen people, maybe ten people. And so, you know, we we are seeing um, you know, we are you know, we we are seeing human capital be just more efficient in terms of you know when it just comes to like building out and and and pushing technology prototypes and and products. And so, you know, the the the the the other you know flip side to that is you know that the past few years in in the capital markets has been you know a little bit rocky. And so brought you know, more broadly speaking, I think that operational efficiency and getting to profitability and sustainability has been a mandate across the board, right? For all you know, all you know, big tech companies, small tech companies. And so, you know, we are not immune to kind of like the you know what the market signal uh signals are, and so like kind of downsizing, being more efficient uh has been a mandate, you know, for the past two years now. Uh and so that's not new. And but you know, what AI can bring to the table to help uh kind of supplement, you know, kind of what a little bit what you lose in terms of human capital, I think you know, we're we're seeing some productivity gains there.

AI Hype Versus Real Defensibility

Alex Kotran 20:03

Yeah. But it and it is hard, right? Because there's been a lot of layoffs in the tech sector. Um and many times those layoffs are are talked about as oh, AI, we're becoming AI forward. But I mean, there's like I think a lot baked into why tech hired overhired, you know, coming out of COVID and there were low and low interest rates. Um and I think companies also have, and certainly the AI companies have this incentive to um, I don't know if I'd say overhype, but uh to continue feeding the the hype around the technology and sort of like positioning them as like this differentiated company, which is obviously commanding a tremendous amount of things.

SPEAKER_02 20:36

Who's gonna write the next AI forward memo?

Alex Kotran 20:38

You know, like and yeah, the Bureau of Labor Statistics published in February, and this is like I was waiting for this, um an update to their projections for a handful of careers, notably uh uh computer science or software engineering or software development. Forget exactly what that term, I think was software development. Um and their updated projection is something like over the next eight years, an 18% growth. I I'm curious for your take on that because I I the the the folks that I talked to so far are surprised by that or I think it's I mean so everybody I've talked to in Silicon Valley is like that's wrong. There's just no there's no world where they think it should be flat or declining. It depends on how you define. I mean, basically the the I think the most cogent response was it's a question of like how it's being defined. And so there's a world where um there's just there's actually a growth of work that includes or incorporates software development because more people can use these tools now to like write code. Um but like functionally, if you're trying to think about like how do I go and add value into the market, um is this a like is this a career pathway that's going to expand or or narrow? I mean, most people I've talked to seem th think it's gonna narrow, but I I feel like we're all trying to read tea leaves that are this is quite murky right now. It is all.

SPEAKER_02 22:00

I mean, I think what, yeah, I mean, like, what is the definition of like a coding, you know, a coder, a computer scientist? Is it somebody who has you know the degree in in computer science or is it a job that requires a degree? Um or is it just someone who you know can can build software? Right. And I think like the you yeah, I mean, I mean we're we're probably moving to a world where you may not need the full computer science credentials to be able to get to get a job that that that's about building uh building building software, right? And I don't like I don't I don't say that as a way to kind of shortchange the value of a computer science degree, although I do believe that you know certain degrees are more conceptual and theoretical than than practical. I do feel like the computer science, you know, what it means to major in computer science may need to be um closer to like applications and building stuff right out there, right? I think a lot of there there have been you know a number of programs in the past where a lot of it is kind of more theory and less applied. We're moving to a world where like you know, the bar for building and pushing stuff and with vibe coding has been um so I don't I don't want to say so low, but it's a lot more accessible now. You know, if you can describe a problem, you can probably build a prototype of it that's maybe 70, 80% functional, right? Um, but then like to take that, you know, to for that last mile, right? The the 20 to 30 percent, you probably should know, right, how this, you know, how how all the syntax and all the language and all the logic works, right? To be able to like debug. So, you know, I still think for that last mile, you know, I you know, having some trained um computer scientist who is you know versed in the foundations of of programming and software uh and code uh is is is will will will continue to be valuable.

Alex Kotran 23:54

Yeah, it's um at the end of the day, if if you have the option of hiring somebody with a background in traditional CS um or someone who doesn't and is just a vibe coder, you're always gonna go with the person who now you might have other sort of like things that you look at. You might be looking at is this person do they have experience with like you know, product management or project management? Um and so I could imagine I I it seems obvious that computer science degrees are going to like if you don't need to spend as much time learning coding languages and syntax and um you know maybe you create more space for them to actually get more hands-on and actually build um you know, I wouldn't be surprised if like a lot of CS programs have students just building companies you know as like their capstone project. Right. Um and I think that's you know, when I hear about like, oh, this is gonna democratize access to things like let's say like coding, um, I think that's right, but I I worry that some people are making the like the assumption that that it means you don't need to teach it as it's not as important. Um and I'm curious about because like you're you're sort of getting I think you alluded to it, but maybe you could just name it like do you see education changing significantly in terms of like the fundamental skills that we're building? Because what I'm hearing from you is that actually those fundamental skills are still really important um for people that are looking to harness the tools.

SPEAKER_02 25:14

So I'm I study history in in you know in in undergrad and you know, before I went into ed tech journalism and NVC, I was actually thinking about being a history, uh, you know, becoming a historian, you know, uh uh as as a profession. So I went to grad school for it. And I think what I've come to appreciate from that, you know, from my history studies is not like it's not it's not the facts, not that it's not that it's not the it's not really the content of the knowledge, but it's like thinking like a historian and like knowing how the pieces like come to be, like how facts come to be, where the sources are, like, and evaluating you know the the credibility of sources and and and judging so that like and it's and it's become very useful as a journalist when you talk to people, and it's also uh when you talk to people, when you look at data, uh that's also true for VC as well, when people pitch you about stuff and you're doing market research. I think that like when you're building software with AI, I I hope that we can like we need to continue, continue to be like critical evaluators of the output. And so like I I hope we never like go to a uh be a world where I'm I'm not a fan, like I wouldn't want to see a world where like we just take what AI like outputs and just like accept it as gospel. I think there should be a pause, you know, and see like, oh, does this make sense? Like, does this seem right? And I think that comes from having more core, you know, like foundation knowledge of you know the subject and content itself, and also like having some knowledge of like the AI systems, right? Which you know, whether it's LMMs or like, you know, what what model you're using to be to just know right how things are created.

Alex Kotran 26:57

And you I mean you described earlier this idea that what one of the things you look for in companies is people that have this connection to the the data and the lived experience of the users that they're trying to serve. And um and this is amidst of a world where anybody can create companies and build, you know, um build tools or rappers. Um I wonder what the equivalent of that is. Like if I think about art and if everybody can create beautiful anime, my suspicion is that it's that will no that will cease to be something that people seek out. Yeah.

SPEAKER_02 27:31

The Ghibli example, right? That it was that one week where everybody was making Jubilee Jubilee memes. Yeah.

Vibe Coding Raises The Bar

Alex Kotran 27:41

Yeah, I I and I had actually just seen Princess Mononoke and um IMAX. And I think it was the most expensive at the time, it was like 1997, I think. Um the most expensive anime uh film that had ever been made. And there's one scene in the movie, uh, it's about like a five-second scene, and it took the animators a year and a half to illustrate um or to animate. Maybe this is a chance for us to shift gears a little bit into journalism. I don't I don't really read random blogs in the internet, so I I can only reference like headlines of studies that I've seen on Reddit um and Twitter, but my understanding is that there's a lot more AI generated stuff on the internet right now. Crap, shit, garbage. Yes, law is there. How is this sort of impacting journalism? What like what's your prediction of how the next few years? I won't I I won't have you predict 10 years out, but just like a couple of years from now, like what are you seeing in terms of just your conversations? I'm sure you still keep in touch with a lot of your old colleagues and you know contacts when you're at Ed Surge.

SPEAKER_02 28:42

You know, I I I'll I'll I'll speak more from like a just a writing perspective, like the the cra, you know, the craft of just writing. And you know, it's taken me a while to get to this point, but at the end of the day, what are you use AI and what are you don't use AI? Like, if it's good, does it matter? And so I I go back and forth on this, right? I I don't I don't have like a pure black and black and white answer. But I think the second the question underneath that is then like, how how do you know what's good? How do you know what's good? And and like I think for us, like who were, you know, I would maybe say traditionally trained, right? We read a lot of books, right? You know, we read, you know, we were exposed to a lot of different kinds of writing. Um, you know, I think you develop a sense of like what you know what the quality or the different bars are there. Um and so like I'm I'm okay now, like, you know, if people use AI to create things that, you know, are you know by some stand, you know, by by some you know, more or less historical objective standards like good. I think where I get worried is that like I think if we rely on AI too much to create stuff, like we may lose, you know, in the future generations, like we may lose that sense of like what is like what is good, what is bad, right? And like you know what, yeah, you kind of learn and develop a uh taste, right, over over time. You know, there's a lot of art that I don't think is a lot style that I don't think is very good, that I don't like, right? But you know, I think that that that are heralded as good. But I think like like how are we gonna develop like taste, you know, you know, in in the future, and what if we like lose that? So I think that's one of the things that I that concern me a little bit when I think when I see people outsource a lot of uh you know writing like to AI for hiring, right? I was reviewing a lot of cover letters, and a surprising amount of them was like AI's slop, right? And I'm just like, in what world like do you think that this is a passable cover letter? But I also think like why and and and so that that to me, maybe that's just some percentage of applicants are just gonna be like sloppy and lazy and about that. But I think on a deeper fundamental level, when we talk about you know the future of education with AI, like how like I still want kids to like develop a sense of like their own taste and what good is.

Alex Kotran 31:04

Yeah. No, I mean this is I was talking to one of our the the guy who designed our logo, actually, um, Alex Moulton. And he's been in the graphic design and like branding space for decades. And I was talking, I basically my question to him is like, what advice would you give a parent of students who are interested in becoming artists? Um and they're you know, obviously this inclination of like, oh, this is a tool that's going to unleash their creativity and give them the chance to do stuff that they couldn't do. Um and Alex he basically landed on this idea of taste. He was his his his point was like the way you develop taste is by working your butt off and learning whether it's like drawing or painting and and struggling through that process for years. And there's no way to fast track it. And sure, there's there's like you could imagine different technologies, and that's not to say that Photoshop didn't you know get rid of people having taste, it obviously just changed what they were spending time on. Um but I think there's like there there is a concern about what does productive struggle look like with especially AI art. Because I think with with writing, it's uh I have not been able to use AI to write something really good. Like I think this what I found is that um well actually let me before before we because I want to talk about how your your process for writing, um, yeah, just to sort of validate this like idea of taste. I I also heard uh that YC podcast where they're talking about their survey, they kind of landed on this idea of like we're gonna be over, you know, we're seeing founders uh over-index on finding developers who have taste. Um so it's interesting that it's actually sort of a cross-cutting uh beyond just art or or the humanities.

SPEAKER_02 32:47

Yeah. I mean, and taste is also like not like I I think taste is what uh taste is a different that's a human differentiator, right? I think maybe LMMs have like an objective, you know, probabilist probabilistic like model of like what general good taste is. And I think that's why you get a lot of the early iterations of like Dolly art, they all kind of look like this like similar in that like retro, sci-fi, a little bit of steampunk style, right? And so like I do think like how do you break away from that? Like, what is your point of view? Uh, what are your lived experiences uh that help you know support kind of what how how you're defining um you know whatever you want your product to be and how it's differentiated? I think that is, you know, I think that's still a very uh that should still be a very personal pursuit. That's a reflection of your different experiences and the craft that you build over the years of practicing and refinement, whether whether you're writing newsletters like I used to do, or whether you're just drawing, drawing circle, you know, drawing circles uh uh for artists. I think. Yeah, I like I I think the appreciation for you know how things are made is a good uh is connected to kind of your development of taste, right? If you know how hard it is to draw a perfect circle, if you know how hard it is to like to craft that perfect like one-line sentence or a headline, I think that um you know, I think auto training and that productive struggle that goes into it is uh you know that that that is key. Yeah.

Alex Kotran 34:21

Yeah, the uh like Picasso um you know, before he drew these like very simple, you know, abstract you know, sketches, um he was doing like hyper-realistic art. Like he was classically trained. Right. And and then he was able from that to abstract to just a single line. Yeah. Um I mean when I see AI art, it's first of all, I I can immediately notice it. I'm curious if people can. I feel like everybody has seen enough of it.

SPEAKER_02 34:50

I feel like our AI radars are just kind of going off the off the charts so far. Maybe it's not even AI, but like, you know, like what suspect it is, and maybe that's unfair, you know, to you know, to human standards. But I'm I'm that way with writing and with art and you know, other other AI creative outputs as well. Yeah.

Alex Kotran 35:08

And I I think it's I don't know that I would like stop visiting a website that had AI art, but I definitely feel like, well, that's lazy. Yeah. Like that's just I'm I'm not just not impressed, I'm like actually like you would almost have been better off not having a graphic here. Like the fact that you, you know, it is like the is like the the the Ghibli thing. It's like the first time I saw it was like, oh that's kind of that's pretty cool. And then all it took was like seeing 10 memes, and I was like, okay. Yeah. Um so you you uh you've tried using AI to write. Um I'm really curious about your your process and have you been successful? Like do you like have like as someone who's written a lot um and has a voice, and that's I assume your voice is at this point quite dialed in. Um yeah, I mean, what's your experience been? Is it changed the way that you write or still kind of just a curiosity experience?

Retention And Teacher Workflow Fit

SPEAKER_02 35:58

I wouldn't say it's changed the way I write. I mean, I write primarily to well not primarily, but one of the reasons why I I enjoy writing is that it like forces me to make sense of the thoughts that are in my head. And so it's almost like a masochistic kind of internal like battle with myself. And you know, and for you know, for that same reason, right? I think I understand why a lot of people do like don't like writing, right? And would and would rather outsource that. But like for me, like primarily, like I do enjoy that process of writing in in in terms of like forcing myself to think clearer. I think it's the best and most brutal test uh of your thinking. Um so yeah, I mean I will say like when I first experimented with AI and writing, it was back um before Chat GPT, it was when OpenAI had its like playground, its beta playground. Yeah, and so I had my friend basically scrape all of my like seven or eight hundred articles on off of Ed Surge and the like hundreds of newsletters, basically train, you know, basically created like a GPT Tony to um just spit out articles that are um in my voice, but like factually, you know, completely off the rails and incorrect. And I was actually like kind of you know, I mean I was amused. I was amused that it could spit out basic essentially fake news in my voice, you know, with like made up quotes and stats and and all of that. And so it was uh it was amusing. Um I will say like you know, at you know, at Reach, um I'm the house editor, and so one of the things I actually do miss a little bit from my uh uh answer today is uh nobody really edits me. And so uh in this in in this instance, you know, I do find certain AI writing tools to be helpful for that, like that kind of like for that like a little bit of gut check. So I do use a tool called uh called Lex that I um that I use to LEX. LEX. Okay. I copy, you know, I still draft I still try to draft in Google Docs and have a more very minimum, you know, uh I don't know, minimalistic by today's standards, kind of a composer and authored intellects. And I think what I like about it is it has it has a lot of like pre-programmed checks that are that remind me of what you know my editor, you know, Betsy and others at EdSurge would kind of like look for. And I think it also gives me, it also provides feedback and suggestions in a way that don't feel like heavy-handed. Um and I would say, you know, like about half of those suggestions are actually like pretty good and what what I might expect from a human editor. And so um, you know what I mean. My writing process these days is still, you know, I've added AI to the mix, but uh more as an editor as an editor um kind of gut check.

Alex Kotran 38:42

Got it. Yeah, I need to I I've struggled with the the like literally the user interface with LLMs because you ask for feedback, but then it's it's not yet able to actually go in and redline and do suggested changes. I maybe I'm missing a plugin or something. Oh, try uh try Lex. We haven't invested in Lex.

SPEAKER_02 39:01

Okay, okay. We should. Um, you know, I think I think but the I think the other thing that's interesting with AI and writing is like recently my colleague published a blog post um about his uh you know, he did a market research into some of the investment opportunities for AI and frontline workers, right? So for people who work with their hands. Um in the, you know, and his original presentation was in the form of a slide deck because that's what we do in in VC world, and like everything's you know, bullet points and some charts. And so he did a lot of that research and you know, he he shared that in an internal meeting, you know, in the form of a slide deck. And you know, part of my job is to like you know take some of that and like publish that in the world. And so I asked him to, hey, like let's work together, let's take let's let's let's create a blog post out of this. What he did was really fascinating. Like he's not a he's not an English native speaker, so he threw that deck into Chat GPT uh and basically had a series of conversations, like instructing, instructing it on like how to turn that into a blog post. And when I read his chain of like conversations, you know, lengthwise, it's probably like a third of like the length of the actual blog post in itself. But to see him like actually like instruct and tell you know the AI and give it feedback on like what you know what what the right sections are, who the audience is, and then like the rounds of edits, you know, that that that he provided, like it was a really interesting way to see, you know, this other way of writing, like you know, instructing AI to write. Um, but you know, based on research that he did, you know, more or less on his own. And so when I look at the uh you know future of writing, you know, I feel like this is like it opened my eyes. Like this is another possible kind of you know mode or process.

Alex Kotran 40:44

And it makes sense for someone who's not like a native English speaker who you know it's not that they have a voice that they're trying to bring forward. It's like they're they're actually it's sort of this like this gate or barrier that they need to break through. Um I so I've I've gone through this, like I just wrote like a very long or very wordy piece on jobs. And you know, I I spent I I spent a lot of time almost just for the sake of experimentation to see if could I get something workable as like even like a first draft. Um I spent like a couple hours actually going back and forth and changing the reference material, and sometimes you can provide too many reference documents, and um and at a certain point I in the process I started thinking about what do I actually want to say, like what's but at a certain point I realized like I'm actually ready to just write this thing. And the closest I gotten to using AI is I'll have a draft rather than go through the entire document. I find I find it's very bad at like my voice is also like not, I'm not I don't write like as a journalist, it's very much more sort of like a little bit off the cuff. I just like not quite ramblings, but you know, purposefully not like trying to structure this as like a beginning, middle, and um so I so LLNs have really struggled with that. I have a draft just like sort of individual sections, then I'll sort of like read it, and then I'll like usually it's enough to like I'll get like a single sentence or two that I'll draw from it, but it always doesn't it messes up the first uh the first sentence in a section. It has like it's always like a little too try-hard in terms of it's like quip. And maybe that's because it's trying to emulate my voice a little bit and trying to be a little bit surprising or sort of unorthodox.

SPEAKER_02 42:20

Um it's funny, like when you see AI trying to like mimic you, I feel like it I'm also I'm sometimes like predisposed to like not like it in in some ways. Because look, I think like for uh for experienced writers, I think we all have we all draw the line somewhere and it's gonna be you know different places of like is this piece still me? Like or like or or or is this piece you know, have I have I delegated too much of this to AI? And I feel like that sense of ownership, at least for me personally, like did I like are these my are these still like my thoughts? Uh and to what extent did AI help clarify how I express these thoughts? Um that is an ongoing kind of like decision or struggle or back and forth that you know I do have, you know, when I use when when I use AI just to support my writing, because you know, as a writer, you you take pride, right? You take pride, this is your work, this is your voice. You know, I I've always thought of writing as you know, a a better, my preferred way of expressing my opinions and my voice. Much better than talking, you know. But so you know, for me, like writing is a part of like my how I think of like my communication, my communicative identity almost.

AI Writing And The Loss Of Taste

Alex Kotran 43:38

I mean, I just I don't know, maybe this is controversial. I just think it's unequivocal. Like AI is a shortcut and it is going to harm and deteriorate writing skills. Like because I I also studied uh it was a blend of history and political science. It was mostly just like interesting professors at Ohio State, and the only common thread of my education was just like lots and lots and lots of writing. And you know, the hardest part for me was always just sitting in front of the blank piece of paper and figuring out how to put my thoughts onto that piece of paper. And by the time I could figure out the first, like literally the first two sentences, the rest kind of just flowed. And so I realized that like most of my writing process was actually just truly understanding what it is that I was writing about. Yeah. Um I really just I do not think that prompt engineering, like the computational thinking and like understanding the models and like figuring out the reference, like that, those are all really interesting skills. I think they're valuable from a like it's like it's problem-solving. It is not completely negligible. It's not to say that that skill set isn't valuable. I do I think it's distinct. Like I think that someone who is prompt engineering their way to uh like a newsletter five years later, if they're that they will still emerge with the inability to write a newsletter, their instinct will always be, let me just go and ask for that. Right.

SPEAKER_02 44:58

Um I mean, I do worry about that for future generations uh of you know, you know, the students now when they come out to the world, you know, 10, you know, 15, 20 years from now, like what like how are they gonna write? And I feel like not having that practice and you know that just like repetition, yeah. I I think there's just way too many tempting, like the sh the shortcut is too tempting and it's too easy and too readily readily available. I mean, I don't think it's any surprise that some of the most like popular like AI study tools are ones that just spit out essays and homework for you and you know, like throw some citations in there and make it like look like you know a research paper, like like you did some research. I do genuinely like worry about that. You know, I I've been looking, you know, I did a little bit of looking into just like rah, like what is the state like state of writing proficiency like these days? And you know, they uh I guess you know, by one measure, you know, the like the NAEP, the last time they did a writing test was like in 2011. And I got I mean, I'm shocked it hasn't like hasn't been another one since. But like, you know, writing proficient was still like pretty low, like it was like a third, like a third of proficient, one third proficient. Um they actually did do the test again in 2016, but they threw out the scores because part of it was like digitized and they were they thought they were comparing apples and oranges. But like I but there are some other measures, like I think last year on the Texas like star assessments, like there were a lot of zeros like on the writing proficiency tests. And so given that this is what the state, you know, what the currency to writing proficiency is now, and and going forward, like the audit audit shortcuts and temptations that that AI affords, you know, I that is you know one concerning part, you know, about this, you know, about the applications of AI for the future of learning that that raises for me.

Alex Kotran 46:46

Yeah, it's I mean it seems inevitable that we will basically get to a point where we say, well, maybe we just need to lower the standards for what we're testing. Um because you don't like like that would be sort of the alternative, right? It's like, well, no, it's all about students using AI, and so we shouldn't really care if they can write on that for themselves. It's more about can they use AI to write. Um, I I worry a lot about this is actually my piece that's I was supposed to publish it today. CJ's probably editing it right now. It so you look at the like the story of technology is always one of technology came, it created more jobs, created jobs that didn't exist before. And so that's sort of like the um the salve that everybody is sort of like playing, like like waiting for is like, yes, AI is clearly gonna displace jobs. I mean, that's right in front of us. We can see it. Um, we don't really know what the jobs of the future are, but they're coming. Um and and so I look at like the loom and like the spinning jenny, and so basically the disruption of cottage weavers to in favor of factories, and like it is true, way more jobs were created. Uh people had uh the like like mobile economic mobility in terms of the geo grade, they could move to different cities and they had the sort of skill that they could take to that city and go into a different factory. Um and and it's when I learned about the industrial revolution, it was like this story of like growth and you know progress. Um but you know, you dig into it, it's just like you know, if you were a weaver and a cottage, a cottage weaver, you you worked from home, you chose your own clientele, you actually had like deep expertise and skill. It was like you were actually creating full garment end-to-end. Um, there was taste involved, and they lost their jobs and then they found themselves in these like windowless, like choking, smoke-filled factories where they're working six days a week. And they had very little agency because the the skill set that they had was this sort of commoditized, like they were pulling a lever or pushing, you know, whatever it I actually don't really know what uh what happens in a garment factory, a textile factory. Um I I worry about that because maybe it's the case that AI will create more jobs. But I but I I worry about are those like is that gonna be work that's that we're gonna want? Like if you're just sitting there sort of like doing the equip the the whatever the modern equivalent of pulling a lever or pushing a button is, and you talk about taste, and it's like, okay, well, to get to taste, I feel like AI is really not, it doesn't feel like part of the equation. Maybe it's maybe it helps. Yeah. Day one of computer science class, absolutely every computer science class should start day one, we're gonna vibe code a video game, whatever, like something cool. This is what you can do. Now we're gonna do the hard part. We're gonna actually learn some like coding languages because in order to really build cool stuff to go beyond sort of just like that one-shot prompting, you need to learn the basics. But I don't I actually don't think even computer science changes that drastically.

SPEAKER_02 49:31

Yeah. I mean, I think entry-level, right, we're we're we're we're redefining what it means to be what what an entry-level job is in in in in the age of AI. Say more about that. Because a lot of entry-level, I mean, well, entry-level programming, right, a lot of debugging, you know, a lot of syntax, you know, more more of the basic level stuff. And I think that, you know, like back not like not even too long ago, like the past decade, all the coding boot camps, right, and you know, and what they what they thought were going to be the skills that you need to create um entry-level coders who are gonna be like good fit, you know, for entry-level jobs, you know, I think all that is being kind of AI'd out, you know, a lot of the, you know, at least from the coding boot camp, uh, for the coding coding boot camp industry. I think, you know, we are, you know, for SSLs at the reach, you know, we hire, you know, we have interns like every year. And I think like the things that we expect them to be able to do with AI in terms of like prep, you know, in terms of like how much prep like preparation that they do, in terms of you know, how fast they can like get up to speed on research, right, in support of some of our market research work. Like all those expectations, you know, all that that bar is like that bar is getting higher, right? And so, you know, I I I still think like I think that the bar for entry level is is getting higher and higher across the board. I think this ties back a little bit to you know some of these AI first memos that we're seeing across the board. I I think this I mean, I I think optim like the optimist part of me thinks that this could be a good thing if you're raising the bar, you know, for you know, generally raising the bar, you know, is a good thing. Does it mean that you're displacing folks? You know, I think there will that I think some part of that is usually inevitable. It happens. Yes.

Alex Kotran 51:20

Definitely displaced. But like what happens to those folks who are displaced, because I think that's the story is, and and it probably will come to pass to some degree, is like like IKEA did this, they replaced all of their call or most of their call center workers, but they retrain them to be sort of like design consultants.

SPEAKER_01 51:36

Right.

Alex Kotran 51:36

Um that feels like a plausible. I mean, like that's the story, I think. That and I think the the question is like, what does it actually look like in other spaces? Can we can we start to tell the story? And I and I think this is a role that VC has, right, in terms of like investing in the companies and the and the founders who are sort of like charting that course, because it's sort of a decision we make. We can use these tools to just replace people or we can augment.

Entry Level Work And Reskilling

SPEAKER_02 51:59

I mean I mean, I mean, as much as we just displace, I think like, yeah, we do we are seeing founders who are helping, you know, a lot, you know, a good number of portfolio are in companies that help upskill workers, right? And help uh help them like trend like prepare for these transitions, whether it's in the same field, um, you know, whether it's in like software engineering, like getting, you know, just becoming a better, uh more well uh, you know, more AI first enabled uh builder, whether it's transitioning into like other careers, you know, where you know that may that may be different. I think that, you know, I think the the the the the opportunity for upskilling workers is just as I would I I want to say it's commensurate to kind of like the threat of like displacement and the fact that they don't do, you know, they end up in like worse situations, right? Um look, you know, one of the things I learned from being a VC is you know they're they're they're very rarely pessimistic VCs, right? And so, you know, it's actually one of the funny transitions when being a journalist where you might be more inclined to look at the bad side and where things don't work. You know, they say that there's a quote, I think like, what is it? Pessimists may be right, but it's the optimists that kind of help impact and change the world, right? And that's kind of like the side that you know I like to look at, I like to try to come from when I look at uh the impact of AI on labor force and you know, displacement versus replacement versus upskilling you know and productive transitions. Yeah.

Alex Kotran 53:26

And it feels like the the jury may be out on the like how prominent like AI tutors should be, let's say, in K-12. It feels like very obvious that if you're trying to reskill um you know mid-career workers who don't necessarily want to go to school and have already gone through that process, like they don't want to necessarily do it again. And I also, it's interesting, almost thinking about reskilling in terms of like where is their transferability of taste where where companies need someone that has sort of like some built-in expertise, um, but now we're lowering the barrier. Lowering the barrier to entry for mid-career is like a much it is much more interesting because there's presumably there's some place of existing value, let's say, in terms of whatever the person was doing. Yeah.

SPEAKER_02 54:08

I mean, I think the the history of human innovation is part of what I like is like the the transference of taste and craft from one discipline, you know, to deep to others, right? And I think AI actually has ideas for helping them bridge that dots, but I think like human innovation, that's kind of where you know, if something hasn't happened, I think like the the most impactful and most memorable kinds of learning experiences and kind of developments come like when you like kind of make that like connect the dots between two, you know, between two things, two fields, two lines of work that were you know previously not really thought of as having any any relationship with each other. And so I think reskilling and upskilling um you know for the future of work, you know, like the reality is you know, like there are still a lot of like jobs that are very like human and relationship centered, right, in education, in work, you know, in in workforce development, in care, that I think like I'm not sure if AI is really gonna like create like fully supplant, you know, what humans bring to those jobs. So for example, like I think learning, right? I think there is a biology to learning. You know, I'm looking at you right now, you know, we're making eye contact, we gesticulate. I think all of this, like there are like neurons like firing app like right now that kind of make memory like moments like these, right, a little bit more memorable than you know, if I was talking to Alex, the chat, you know, the the the chat bot right now. And so across care, across learning, across upskilling and just human development, you know, I I do think there's going to be, or I hope, right? I hope that there is going to be an appreciation a realization and appreciation for you know some of the biology of learning that I think humans will always you know like like play a critical role and uh in in irreplaceable role in.

Alex Kotran 55:56

Yeah, could not agree more. Um, 21, thank you so much for joining.