Khan Academy CEO: How to Re-Skill for the AI Economy in the Next 3 Years | Sal Khan

0:01This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required, compatibility and availability varies 18+. And so they're asking and we're saying, hey, use more AI, use more AI. Software engineering, design, product management and all of these tech firms, including at Khan Academy, are changing fundamentally.

0:36This is Sal Khan. He runs Khan Academy. A hundred million people learn on it every year. He built one of the first AI tutors in the world. Excellent work. No one has spent more time on how people reskill for AI. If we could do three times more with the same resources, we will do three times more. So that would never be the catalyst for layoffs. If you can automate yourself out of a job, we're going to give you a better job with better pay. What are you thinking about the next five years? Don't run to where everyone else is running. Try to find the lanes that are most empty. And I would say... And we're in this interesting phase where AI is changing everything.

1:13And your previous book was called Brave New Words. And now you're writing a book called Job Shock. So what has changed in the past two years? Oh, wow. What has changed? So much has changed. You know, what I wrote in my last book, Brave New Words, about AI and education, when I was writing this in 2023, I was afraid that as soon as I write it, it would be obsolete because things are changing so fast. But I keep looking at it and it's still, I feel very relevant. I write about things like how AI will broaden the aperture of how we assess people, how, you know, people have talked a lot about AI tutors, but there's other forms of assessment, interviews, oral exams.

1:54And you actually are starting to see some of that. So we could talk a lot of, you know, some of those things that were just ideas three years ago, we're now seeing in reality. I just talked to the head of admissions at Caltech. And a lot of students who apply, they give these research papers they've done. And the admissions officer says, some of these are real and some of them not clear if the student did the work, et cetera. And so they're already using AI to essentially do oral exams with these students to evaluate if they understand. And this is pretty high stakes. This is for admissions to top universities.

2:23So you're already seeing AI in that part. We've launched Conmigo out into classrooms. We've learned a lot. There's, for a lot of students, it's been super useful, but we've also learned that AI 101 or the first 1.0 AI, it's waiting to be asked to do something. And some students don't know how to do it. A lot of them didn't. So what we're now, we're in the process of launching a new version. It's more proactive and it isn't AI for AI's sake. It's how do we get AI to motivate students, give them more practice, have the teacher in the loop, etc. In terms of why I'm now writing Job Shock, which is some people are surprised because most of not just my books, but most of the things I've ever said tend to be pretty optimistic.

3:06But I am increasingly worried, you know, not too far from this was about, I don't want to say about three or four months ago, maybe a little longer. I have a good friend from college who's now a prominent VC. And he said, hey, Sal, I want to talk about reskilling. And so I remember I drove over to his place not too far from here in Los Altos. And while I was driving, you see the Waymos like you see everywhere here. I'm like, oh, that's interesting. This is cool. Look, I live in the future. Then I go over to his office. And he was telling me that one of his portfolio companies has a call center that they've just automated 80 % of their workforce with a generative AI solution.

3:42I don't know the exact number, but it sounded like it was in the hundreds, if not thousands of people for that one call center based in the Philippines. And this is why he said, well, maybe I want to take 1 % of my firm's profits and put it into reskilling sometime. And we started brainstorming. And we appreciated that, especially in an economy like the Philippines, where depending on how you measure it, 5, 6, 7 % of their GDP is directly related to business process outsourcing, mostly call centers. Yeah. If this becomes a trend and there's a lot of reason to think it will over the next three to five years, what are all those people going to do?

4:19And then when we see a Waymo, it's awesome. I was at TED a couple of weeks ago and I interviewed Takedra Maokana, who's a co-CEO. I'm a big fan of Waymo's. I think it's going to save a lot of lives. It's super convenient. But my uncle's also an Uber driver in New Orleans. What's he going to do? That's not people's first choice job. It's usually their last choice job in a lot of cases. And if rideshare goes away, and that's just the tip of the iceberg. Out here in Silicon Valley, everyone's talking about what's happening to software engineering, product management, design. Are those roles going to merge?

4:51Are we going to hire more or less, et cetera? But it is crazy times. And I wrote a job shock genesis was I wrote a New York Times op-ed piece in December where I made this argument that we have to take this a lot more seriously. We can't just be like, oh, the industrial revolution turned out OK, more jobs. First of all, the Industrial Revolution also had many wars. The 20th century was the bloodiest. We experimented with communism. We experimented with everything. A lot of these things didn't go so well. And so we should have a plan. And that's what job shock is about. How will it possibly transform?

5:27What are the types of things people need to learn? My son, my oldest son is a junior in high school. What should he be caring about to be relevant in four or five years? But also, how do we as a society adapt? How much time do you think we have? for someone who's an Uber driver watching right now. Because I feel like with Wamos, it took them still, it took them 10 years to roll out the operation that we're seeing now when we're only seeing it in San Francisco. I think you're going to start seeing a real dent in this in five or 10 years. And that's slower than what a lot of the AI optimist evangelists are saying, but that's also pretty fast.

5:58That's our lifetime. That's before your kids go to high school. Yeah, exactly. Or when they go to high school or before they go to college. and driving a vehicle is the largest employer, one of the largest employers of men. I mean, women, obviously, my Uber driver just now was a woman, but like disproportionately men globally. If you have a lot of men who can't find a job, who are angry, and women, that's going to lead into politics, polarization. People are going to move towards extremes. So I suspect that you're going to see in the next four or five years, the dislocations start to happen for real.

6:38And we're going to struggle as a society on how do we reskill people. I'm not a fan of things like universal basic income. I think like universal basic work, people need to feel like a sense of contribution. But these are big questions that we have to be creative about. Yeah. And it's not just Uber drivers, right? You mentioned call centers. What are other jobs that are in danger? I think the writing is on the wall that software engineering, design, product management, these product jobs and all of these tech firms, including at Khan Academy, are changing fundamentally. But are they endangered?

7:11Like, are these people going to end up in the street? Because I don't think so. I think they're... I'm not sure. I'm genuinely not sure. You know, what I've told our Khan Academy team, because you can imagine the employees are very worried and they're saying... How many people do you have? We have 350 people on our team and we're two thirds product people. And so they're asking and we're saying, hey, use more AI, use more AI. And Anthropic just gave us a report that we're at the top of the stack, apparently, on organizations that are leaning into these tools. But you can imagine the team sometimes feels insecure about this.

7:39They say, hey, if we can be three times more productive, that means you're going to lay off two thirds of us. And what we've said very clearly to our team, and it's what we believe, is, look, the reason for us, we have so much to do. Khan Academy has the budget of a large high school. If we could do three times more with the same resources, we will do three times more. If we could do 30. So that would never be the catalyst for layoffs. Catalyst for layoffs is if we don't bring in enough philanthropy, if we don't bring in enough earned revenue, then just money will force us to it. But it's not going to be AI.

8:07AI will allow us to do more. But we have also said that people's jobs are evolving and people really need to learn how to keep up with the times. And we've put on our career rubric, like learning new tools and adapting, or we're actually putting it into our career rubric. So I suspect that many of our team members are already there. A lot of them are very willing and are going to get there. But I do sense that there's some people who are like, I don't know if I like this. I don't know if I want to do this. And I think that that might be a more a bigger struggle if folks don't want to aren't aren't in a position to adapt or change.

8:46And you think they're going to be at risk in the next five years? I think some. Yeah, I think, you know, we don't know exactly what these titles are going to be. But this designer, product manager, engineer division, it's really kind of blending. Yes, exactly. And the current framework came mostly out of the late 90s into the early 2000s of, OK, how do you build web apps? And it's that model. But the new world and not only are those roles, people who used to be designers or product managers are now not just prototyping anymore. You know, last year I met some heads of product. They're like, oh, I'm forcing all my product managers to create a prototype.

9:22And I'm like, wow, that's really bold. Well, now at Khan Academy, we're taking our designers and our product managers and we're giving them full dev environments so that they can actually do pull requests and do full, you know, deploying, deploying of code. Now it's still leaning on more of the prototype proof of concept, but it's blending. And similarly, because engineers can do so much more, I think there'll always be value of that engineer who could be super deep, but they're going to be much more productive. But I think you're also going to see a breed of engineer that can go more customer facing, et cetera.

9:50So yes, these things are all blending. And just the development cycle speeding up too. So people have to feel comfortable with that. Going back to the jobs that are endangered, drivers, call center, customer support, people who work in product slash designer, software engineer, and don't want to adapt. Anyone else? The people I think are safe are the people who really lean into the human element of it. So I've said, I think teachers are one of the safest jobs. But once again, even before AI, I don't view that the role of the teacher is to be the sage on the stage and to just disseminate information.

10:26We've had textbooks for now centuries. We have videos on demand. We have other places to get information. Yeah. I think the teachers and we've seen teachers do this. Good, great teachers have been doing this for decades, move up the value chain. And they're the planners, the architects, the coaches. They're able to form, you know, have eye contact with students and be their motivators. I can't speak for you, but for myself, when I think about the great teachers in my life, some of them were like, wow, they used to give a good performance. But most of them were the ones that ran a Socratic dialogue with all of us, really made us dream or do things we never thought about.

11:05Or say, hey, Sal, can you see me after class? Hey, Sal, I really liked what you did that time. You know, you have a knack for that. Or Sal, I didn't like what you did. Stop it. If you did that in the real world, someone's going to be really annoyed with you. That's what I remember. So teaching, I think nursing, hospitality, sales, anything that has a huge... Oh, absolutely. I don't think you're just, I mean, maybe telesales and things like that, AI. And I know there's been companies for a while that I can be the salesperson. Yes, some of these sales agents are, they're going to get better and better, but their cost of deploying is so low that we're all just going to get spammed unbelievably by it.

11:47And so they're probably all cancel each other out in terms of efficacy. But the ability to actually form a relationship with someone, be a convincing advisor. And, you know, the best salespeople aren't like salespeople. They're actually smart, thoughtful people where you actually respect their opinion. Yeah. And you want their advice. Yeah, we've seen that. And you know they're a straight shooter. And they'll say, you know, our product isn't best for that. You might want to use that. And you're like, oh, I trust you. Let me use you next time. And so I think that human connection, anything that can be purely done on a computer and you're not interfacing with other human beings, you're not having to build some of these durable skills of communication and persuasion and listening and collaboration.

12:25Those will change or they might be under threat. OK, let's talk about adopting those AI tools. You just mentioned you use Anthropic a lot. Can you talk about the processes in your company? that seem absolutely mind-blowing, that would have seemed absolutely mind-blowing two years ago, and now they're part of your process? Oh, well, I think engineering is where we're seeing it the most. And that is where we also, and I think you've heard many people say, some of the engineers are really leaning into this. They're running five, six, seven, eight, nine, 10 agents simultaneously, writing code, reviewing code.

13:03The Anthropic folks told us that we were one of the biggest users of actually the agents for reviewing code. Interesting. And, you know, we, we, I've gotten a little sticker shock for how expensive this stuff is. Our, our run rate on, on our, our Anthropic bill is about$1.2 million right now. Wow. And it's growing fast. That's a year or a month? A year, a year, a year, but it's still a lot for, you know, 200 or so engineers. It's mostly engineers. And we're looking on a daily basis and our CTO was saying, yeah, we just saw one of our engineers spent$3 ,000 in a day. I think that was the number of compute costs.

13:40And I'm like, oh my God, what is, I think. Are you trying to push them to use even more? Because you see some companies trying to do that. I was like, what are they doing? And then he's like, then we talked to them and they actually in a few hours were able to do something that would have normally taken them three or four months. It's like, okay, that was a great use of$3 ,000. Go spend more. But just today we had a company meeting where I said, you know, make sure you're not using the Opus model for just everyday tasks. use Gemini and use your chat GPT accounts if you're trying to figure out the local restaurant to go to.

14:11So we are trying to get there, but I had the same answer to our team. I was like, look, if we're using 1 million, 2 million, 3 million dollars of tokens every year, but if we look at what we're producing and that would have cost us 20, 30 million dollars in the past to produce, yeah, we should keep doing it. But anyway, the engineering is the most. I think the rapid prototyping. We just launched a feature a few months ago that in the old days, it would have been like, maybe we can fit it by next school year. And this is one of the engineers at a hackathon, vibe coded it in a day. And then on a weekend, he said, you know, he's an engineer.

14:47So he had the full development environment, but he was able to create working on our real data. And it's a very front end kind of game thing. So it didn't have a lot of implications to what might happen into the back end. He's like, I think we could ship this thing. And got some designers involved, made it a little bit nicer. And we were able to ship it in about a month, something that would have taken. And so that is the thing that's really, before people said engineering is hard. So do as much time as possible on design so that you don't build things that aren't going to be valuable. But now that model is changing, you should say, no, let's build stuff fast, put it out there.

15:25Not always. Certain things you have to really make sure work before they go out there, but certain things you can try out, test your thesis fast. So I think the velocity of the organization is much faster than I would have imagined a year or two ago. Do you think 2x, 3x? What's your estimation? If I'm honest, as we grew, 10 years ago, we were very fast. We started up, you know, 20, 30 people. and as we grew to 200, 300 people, from my point of view, it started to feel slower because you have coordination costs, et cetera. And I think relative to where we were two or three years ago, it does actually feel, and this is a feeling, we haven't nailed this down, but to me, it feels at least 50 % faster, maybe 100 % faster than we were.

16:11That's awesome for us, such a large organization. Yeah, yeah, and it's accelerating. Yeah. What about operational stuff? Have you tried automating that for yourself, maybe? What are you using? You know, we just approved all the connectors with the various AIs. So now our AIs can access our Slack and our Gmail and our docs and all of that kind of stuff. And we are learning and encouraging people how to use it, how not to use it. Don't let it do things without you being in the loop. Don't let it post on social media. Don't let it send the email. Maybe let it draft the email and then you check on it.

16:44But I am on a daily basis. You know, I talk to one or more of these of these AIs and I say, what what's falling through the cracks? And it's acting like a little bit of a chief of staff for me. And it's going to my Slack. It's going to my Gmail. It's going that it's like, oh, I've drafted something for you. Is this good? So I'm already using that and I'm finding that super useful for just there's always people who are asking me, hey, can you do this? Can you do that? And I typically probably 20 or 30 percent did fall through the cracks. And I would only when people keep bugging me, do I not bugging me, reminding me, does it does it does it does it actually happen?

17:18You know, there's just a lot of paperwork type stuff that we has dramatically accelerated. Totally. We have to write a lot of proposals to people for philanthropic or even school, you know, earned revenue type stuff. And now you can have a meeting, keep all the notes, have a first draft of a proposal. We won't just send that. But now within 10 minutes of the meeting, you have a first draft. And then that's a good catalyst for everyone to now start weighing in. That would have taken a week before, two weeks before. Content creation. When I make a video, I always try to like, what about this edge case?

17:52What about that edge case? I've been making financial literacy videos lately and financial literacy is full of edge cases. Like what if I'm 55 and a half, but I've done this and this and I'm in this tax bracket. and that has dramatically shortened the research time for me because I am literally, once again, I don't just go to the AI and accept its answer. I look at it, I'm like, oh, show me a source that validates that before I make a video on it. And I'm like, oh, it's from irs.gov. All right, I will now make that video. So that's dramatically shortened. As I said, our content creators are able to scale much faster.

18:24They're still reviewing all the questions, tweaking it, but they are definitely getting all the research, of course. Quick pause here. If you're enjoying this podcast, you will absolutely love my Inner Circle newsletter. So what I basically do is I take all the tips from these podcasts and apply them to my personal life, to my investment portfolio, and to my businesses, this media company, and my language teaching business. Sometimes we get amazing results and I share our real tactics. Sometimes we don't and I share that too. Think of it as an insider version of this podcast. The link is in the description.

18:54Join my free newsletter to stay ahead. Does it ever make you scared? Because when you're seeing this progress are you ever thinking like oh in three years ai is able to develop another kind of academy just by itself just because like why i wouldn't need humans yeah and i have two responses to that one is my um my emotional ego response to that and then one is my more higher-minded correct response to it thank you both my emotional ego response is yeah i do worry about that. And one could imagine, I think it's very hard right now to vibe code something as like the same fidelity and quality. You could make something that looks kind of like Khan Academy, but not have the rigor, et cetera.

19:38I think it's quite plausible that in three or four or five years, people could maybe start to develop these things. There, I think there still will be the burden to how do you have the efficacy data that it really works? How do you make sure that you're working with school systems in a way that meets all of their data privacy agreements. Talk to their agents. Yes, exactly. Figure it out. Yeah, yeah. But there will still be some of those frictions. It won't be as simple as vibe coding. So I think there will still be some advantages of already having something out there that's already being used by millions and that you have real data on, real efficacy on.

20:13You're able to work with the system. The correct answer, which I tell the team all the time, is if we or someone else can create a solution that's as good as us or better and actually show that they're doing it, it just doesn't feel that way. As a nonprofit, we should be happy about that. We should say, good, one of the many problems in the world has now been solved. What other problems are there to be solved? And for better or for worse, in the education space, there are many problems to solve here. And one of the things we're thinking about at Khan Academy is assessment. So standardized assessment, historically, it's multiple choice, etc.

20:51That's why a lot of people don't like it. It's very narrow. But we're working on real standardized assessment with real psychometrics that has some of the traditional, but also starts broadening the aperture, asking students to explain their reasoning, giving teachers not just what percentile their kids are in, but what giving a narrative. And those things like that are also very hard to just vibe code in a garage. Totally. Because you're doing all the psychometric data, you have to get it out there with millions of people to use it. And I think that's an area where very few VCs or frankly, even private equity firms are like, let's create a new standardized test leveraging AI because it's a lot of work.

21:27And it'll take years to get the right psychometric validation and to go out there into schools and other places. But that's where I think we as a nonprofit that's tech focused can be valuable. We're thinking a lot about credentials, high school credentials, high school diplomas, credit recovery, college degrees, pathways into jobs. And, you know, those are the places where I hope AI gets better. I hope it can create a lot of solutions. But I would be surprised if markets on their own are able to solve a lot of these problems. Yeah, but someone who's listening and thinking, OK, he's positive about being likely replaced in three years.

22:05I'm not. So for someone who has these thoughts, can you talk more about your new initiative? Because when I heard that at TED Vancouver that you're partnering with TED and ETS, for me, it was another full circle moment because I also took TOEFL. And we partner with ETS for my other channel where I talk about learning English. And you're doing this$10 ,000. Is it a bachelor's degree? How long does it last? And this is going to the whole job shock thing. I think we need new systems of skilling people. Yeah. And this started with a conversation with Chris Anderson, who is the outgoing steward at TED.

22:42And I've taken on this new role as vision steward there. And the reason why he wanted me to take on this role, he says, you know, we should do this, Sal. The future is all going to be about human purpose, human meaning and human connection. It's almost hard to focus on meaning and connection if you don't have a job. Totally. If you don't have a meaningful job, if you can't put food on the table. And so we were launching. We announced it at TED. CONTED Institute. It's intended to be accredited. We're pursuing accreditation. We're likely to do both four years degree and master's degree. And the idea is it would be essentially three components.

23:17There will be an academic component. We still think, especially for the bachelor's, you should learn your history, your civics, your accounting, your statistics. But most of the program that makes it really differentiated, it will be online, but we're going to be putting students of all ages. We've already gotten in the last two or three weeks, 3 ,000 people interested in being students. And it's amazing. Most of them have bachelor's degrees. Many of them have master's degrees. And you can sense the fear that they have. They need to requalify. Exactly, being relevant. And what we are going to be doing is putting people into group simulations, projects, likely leveraging another platform.