Quests, token leaderboards, and a skills marketplace: the elite AI adoption playbook | John Kim

1. 0:00 – 2:45

   Introduction to John Kim

   1. JKJohn Kim0:00

      Unleashing this power of AI and giving it to the power of marketers, salespeople, you get all these cool ideas that get rolled out rapidly to the market.

   2. CVClaire Vo0:07

      It's taking someone's super creativity and giving them powers to deliver it to your customers.

   3. JKJohn Kim0:12

      This is an internal platform where anyone in the company can raise their hand and create what we call the quest. When there's a quest, AI can actually read through the specification, create PRDs, and start actually coding.

   4. CVClaire Vo0:24

      Basically a marketplace of AI needs and AI builders inside your company where anybody can just pop in and say, "Oh, I think I know how to do that." So tell me a little bit about this dashboard.

   5. JKJohn Kim0:33

      So what we're seeing here is the overall usage of our token at the company level. We measure AI gods as somebody who spend more than 100 million tokens a day.

   6. CVClaire Vo0:40

      What I love about this moment is I think it is just such a moment to learn things you could never learn before, because the best teacher with the most in-depth knowledge and an endless willingness to go do research is right there at your fingertips.

   7. JKJohn Kim0:52

      This is a beautiful time to fail forward and still get up and run faster than the other, 'cause innovation doesn't start from a pure theoretical structures. It start with people who have that energy and the story behind them. So find them. They're always in your organization, and then really build energy around that.

   8. CVClaire Vo1:09

      [upbeat music] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, I have John Kim, founder and CEO of Sendbird, and he's gonna show us his AI token consumption leaderboard, where everyone in the company is ranked from AI newbie to AI god. He's also gonna show us how AI quests can be the key to company-wide adoption. Let's get to it. This episode is brought to you by WorkOS. AI has already changed how we work. Tools are helping teams write better code, analyze customer data, and even handle support tickets automatically. But there's a catch. These tools only work well when they have deep access to company systems. Your co-pilot needs to see your entire code base. Your chatbot needs to search across internal docs. And for enterprise buyers, that raises serious security concerns. That's why these apps face intense IT scrutiny from day one. To pass, they need secure authentication, access controls, audit logs, the whole suite of enterprise features. Building all that from scratch, it's a massive lift. That's where WorkOS comes in. WorkOS gives you drop-in APIs for enterprise features so your app can become enterprise-ready and scale upmarket faster. Think of it like Stripe for enterprise features. OpenAI, Perplexity, and Cursor are already using WorkOS to move faster and meet enterprise demands. Join them and hundreds of other industry leaders at workos.com. Start building

2. 2:45 – 5:51

   The Delight.ai swag store built by marketing in two days

   1. CVClaire Vo2:45

      today. John, I love what you're gonna show us today because I tell people right now, they wanna transform their company, they need to think of their team as a product. And what you're gonna show us, a little spoiler alert for everybody excited to get into this episode, is how you've turned AI adoption not just into a program in your company, but a product. So tell me, what's your ambition for your team around their use of AI?

   2. JKJohn Kim3:13

      We want to become the AI-first company, and what we mean by that is not just to adopt AI as a tool, but how do we make AI as part of our workforce? So we're really trying to empower people and give them right set of information and tools so that they themself can really harness the power of AI. And some of the things we are hopefully about to show you today will inspire people to do something similar.

   3. CVClaire Vo3:34

      Yeah, and let's, you know, go to, go to the outcomes, because I think a lot of people that I talk to are trying to articulate the why behind adopting AI that goes beyond, "I would like you to do more with less." And that's a lot of what employees are hearing right now, is just, "I want you to go faster. You can go faster. We should be able, doing more, more, more." But I think what your team is building is showing a different benefit of, of adopting AI and everybody becoming builders. So you wanna jump in and show us some of the stuff that you all are building with AI, and then maybe we'll back into how you got the team there?

   4. JKJohn Kim4:09

      Welcome to, uh, a Delight Ass Shop. It's, uh, it's... We're really excited about this. This is a swag store that really captures the culture and the energy of where our company's headed. Uh, the store is called Big Ass Energy. It's agent as a service. And this entire, uh, store was built by our marketing team without engineering support. So you can actually buy really cool swag. They are very timely. Uh, actually, I, I really did ask my team to make this.

   5. CVClaire Vo4:37

      [laughs]

   6. JKJohn Kim4:37

      "My ass is bigger than your SaaS," "Fully deterministic." Uh, this, I think this is one of the most popular swags we have right now. Uh, you can actually go and buy this. So our marketing team integrated Strip- Stripe integration, so, uh, yeah, we do charge a little bit of money, but I, I think it's gonna be really cool. Uh, another favorite, context window, I carry a lot. So, uh, imagine, like, unleashing this power of AI to your marketing team who add this amazing creative energy. And instead of asking, like, your design team or, you know, engineer to put this site together, they put this site together in matter of a day or two, and then it's now up and running. We also have a super, uh, secret Easter egg, uh, th- for those of gamers, uh, who are listening. If you do Konami Code, up, up, down, down, left, right, left, right, B, A, and here's this little secret. So we are throwing a conference in May 7 called Delight Spark in San Francisco. It's gonna be a really amazing conference bringing the CX leaders, AI builders from all over the world. We have people joining from Anthropic. We'll be showing, showcasing our future roadmap. So it, hopefully, it w- it'll be a great chance to really learn about the cutting edge of AI, but also thinking through the lens of where is the future of customer experience going to look like. So this

3. 5:51 – 7:55

   The before times: when fun had to earn its place on the roadmap

   1. JKJohn Kim5:51

      is what our marketing team has built together.

   2. CVClaire Vo5:53

      I, I just have to stop and reflect. So we just had a really recent episode with Jason Levin, the CEO of Memelord, and he said, "Let your marketers cook."That was his whole thesis, which is when marketers can be builders, they can build things that delight your customers and acquire them. And I just go back to, like, the before times. If marketing had this idea, it would be like, "Well, can we prioritize it? Is it worth investing engineering resources in? It's just for this event. The event's gonna pass quickly. Just do something out of the box in, you know, our CMS." And then you get this sort of, like, middling experience for your customers. Very mediocre, very, like, MVP experience for your customers. And now I think just looking at this, this store and the Easter egg and the way you get into the event, it's taking someone's super creativity and giving them powers to, to deliver it to your customers. And-

   3. JKJohn Kim6:56

      Yeah

   4. CVClaire Vo6:56

      ... this again is, like, the example of it's not about going faster, it's about having a bigger ambition and doing more honestly fun things. And I think this is underrated too, which is it's so hard to build ... Like, it's so hard to prioritize fun in your product. But when fun-

   5. JKJohn Kim7:12

      100%

   6. CVClaire Vo7:12

      ... can be cheap, you should be more fun. So that's my, um, my thesis on why you should let marketers become, become builders. And, and I'm sure they love it too, just from a, like, a team engagement, you know, creativity perspective.

   7. JKJohn Kim7:27

      Yeah. I love that 'cause that's exactly what happened. 'Cause imagine sitting in a, a room full of engineers and product leaders and say, "Hey, you know what? We have this cool idea. We want to add this to your product release cycle and roadmap. It's gonna take you two sprints." Uh, it's gonna very hard, it's gonna be very hard to get that on the table. But just, like, again, unleashing this power of AI and giving it to the power of, of marketers, salespeople, like, you get all these cool ideas that get r- rolled out rapidly to the market. So very

4. 7:55 – 13:47

   Demo: The Automators platform and quest system

   1. JKJohn Kim7:55

      excited.

   2. CVClaire Vo7:55

      Well, I would say that not every marketer, though, a year ago or two years ago was coding, although many more are now. So how did you get the team here? Like, how did you manage the transition from classic marketing, everything has to go to engineering, to actually enabling teaching people how to use this product or how to use AI, and then how to get what they wanted done in production?

   3. JKJohn Kim8:18

      So to really help facilitate that, uh, transformation, we built out a platform called the Automators, Automators platform. This is an internal platform where anyone, uh, in the company can raise their hand and create what we call the quest. Now, this website is particularly has been, uh, designed to just show you the demo today, but actually, you can actually create a quest on your own. So let's say you're a finance department, "Hey, I want to automate my account receivable and account payment, uh, payable workflow," you can kind of do that, and then some other engineers can come in and help. Or if they're AI-enabled, they can build it themselves. So just to give a couple example, uh, if you go to a completed list of quests, uh, these are all the things that have been built, uh, or being pending. And then so, um, let's say, uh, you go to, uh, quest. Then there's usually a quest giver. So this person, usually somebody raising their hand saying, "Hey, can you ... Some- some- can somebody help me build a customer account lookup using kind of different workflows?" And other people are like, "Let me actually give you a hand." So two people actually, uh, teamed up to build out this workflow, and then the result is they usually submit either code repository or some kind of a skill, right? Uh, this video, uh, you can see how to actually use those skills. Of course, it's a, a internal workflow, so we kind of blurred it out, but you kind of get the idea, right? So you have this, all these skills being built. Now, on top of that, what we're just rolling out, this is, like, hot off the press, I guess, um, or fresh out of the oven, is when you create this kind of quest, you can actually now ask AI to build it too. So when there's a quest, AI can actually read through the, uh, specification, create PRDs, and start actually coding. So this is the next level is a- alongside human, uh, engineers and team members, now we have AI agents who are also helping us build automation and workflows. To do that really is to help people also learn themselves to how to build these tools. So we have these internal guidelines that continue to, uh, continue to get updated pretty much on a daily basis, teaching people how to set up GitHubs, create new applications, and also internally, we have created this app template where all the authentication and all the, uh, environments have already been set up. So what marketer or the CSM, customer success manager, has to do is they just, uh, um, uh, extract the template and just build it on top of it, and they don't have to think about the rest of the infrastructure. It's fully compliant. All the security is already pre-built in. So all they have to come up is with an, a cool idea they want to, uh, bring to the world.

   4. CVClaire Vo10:51

      I, I just ... I wanna pause really quickly and just reiterate for folks that are not watching, 'cause you, you breezed through it, but it's so powerful, which is you built ... And, and this is totally separate, I'm presuming, from, like, all your other product roadmappy stuff. You built a very fun, I love the idea of a quest, the ability for your team to request an AI automation or tool from someone else in the team. So you take a subject matter expertise, like a recruiter or a salesperson. They know what they want, they just don't know how to get there, and you're like, "Engineer, will you go on this quest with me?" And they make the request. And some things that we missed that I wondered if you would, wouldn't mind pulling, pulling up just showing folks is you've also made it really centric to the value you're getting out of this automation. And so I saw in the corner of the quest, like, what's the risk of it. That's probably some assessment of the, the data it touches or what it, what it does. The weeks-

   5. JKJohn Kim11:48

      Yep

   6. CVClaire Vo11:48

      ... saved, and then who's the, the team or person that's benefiting it. And then I love this idea, like, people can build, like, jump in and help with these things without having to go through a whole, like, prioritization exercise, all this kind of stuff. And so I'm imagining you're kind of, like, building this, like, shadow AI roadmap that works really efficiently. Basically a marketplace of AI needs and AI builders inside your company where anybody can just pop in and say, "Oh, I, I think I know how to do that."... and build it. Was that kind of the intention, is to get it out of, like, the big prioritization mess, get it out of, "I don't know how to do this myself," and kinda make everybody feel responsible for it?

   7. JKJohn Kim12:29

      Exactly. Um, because if you think through the, the traditional logic of software development life cycle, you thinking... think through the lens of sprints.

   8. CVClaire Vo12:37

      Mm-hmm.

   9. JKJohn Kim12:38

      And you try to fill up the sprint with different prioritizing blocks.

   10. CVClaire Vo12:41

       Yeah.

   11. JKJohn Kim12:41

       But sometimes, you know, people, uh, have th- these little, you know, tiny micro, you know, vacations, I call them. It's like where they have some free time. They want to, like, build other stuff that are not, not tied to the most important code repository, your main product that's very, very stressful, or just fun little side project that can help out. But also, this has immediate customer pain, the user you can talk to within the company, so there's that feedback loop, and the moment you deliver the, uh, the value, people are like... You get the instantaneous, uh, dopamine hit, if you will. So there's a lot of fun to this. And what's happening behind the scenes, people who are completing some of these quests, they actually earn experience points. If you earn enough experience point, you can change to a gift card, you can have a tea with any executive you ch- you choose. You can present what you built to the rest of the company. Uh, so we do weekly stand-up on Wednesday. So we have people coming up to the stage and sharing what they built, uh, with the entire company. This week was recruiting team automation. Previous week, I think, was marketing team. So there's the different team showing, and it's almost never actually the engineering team. It's other teams that are, like, really excited to show what they built.

   12. CVClaire Vo13:46

       I love that

5. 13:47 – 16:06

   The AI Engineer for Internal Operations role

   1. CVClaire Vo13:47

      so much. And then, you know, the other thing that you did, which is very practical, which I've also advised almost every company to sit down and do, is you have a bunch of people that have vibe coded something with Claude Code sitting on their computer, and they, one, just either don't know how to get that to production, or they're getting it to production for the entire internet. They're just, you know, pushing it up to Netlify or Vercel and saying, "I built this thing." And I love the idea that you both built knowledge guides for how to learn core skills like Git, that will make people a little bit more fluent in building things. But also, please, everybody stop and listen. Make a templated happy path to secure production for the things that people wanna build. Behind auth, with the right kind of data access. Just make it so, because your team's gonna do it... Somebody's doing it anyway, and it's a very low investment to get a lot of velocity on things being built, but also a lot of kind of, like, right-sized security. I would say that it's not a hard thing to do, so I love that you built that. Who is responsible for, like, maintaining that, keeping it up to date?

   2. JKJohn Kim14:56

      Yeah. So one of the team that we created is a AI engineer for internal operations. It's a very mouthful, but really, the team is responsible for helping and accelerating, uh, the... our AI transformation to becoming AI-first company. So this role directly reports to me and our chief of staff. So it has a ability to work cross-functionally, but obviously, there's a lot of support from our CTO and engineering team, as well as our InfoSec. So they partner very closely. So we have this task force where we meet on a weekly basis to talk about unblocking some of these challenges, whether it be compliances, how do we log things, what are the software that we can actually vet everything in advance. So when our team's like, "Hey," like, "I'm in sales. I wanna build this tool," then here's a full tech stack, AI stack. You don't even have to worry about databases. It's just all there. Come with your idea. And everything has always been, uh, already been vetted, so there's a working group. But it, it didn't start immediately that way. It actually started with couple of people, you know, kind of building out their own personal tools and showcasing. Uh, but again, it came from the non-engineering team, which really gave us the optimism, like, we can actually do this, and let's actually build more infrastructure so these people can run at 100 miles per hour.

6. 16:06 – 17:19

   Demo: The company-wide skills marketplace

   1. CVClaire Vo16:06

      I love that. And so speaking of infrastructure, it's not only these one-off automations and workflows or guides for building apps. You've also built, which I think is very smart, a company-wide skills marketplace. So tell me a little bit more about how that works.

   2. JKJohn Kim16:20

      Yeah. So here, anyone can create a plugin. Plugin is a, a collection of skills, or you can create and download individual skills as well. So let's say you are in sales team, or even if you're not in sales team, you wanna learn more, you can actually look at the sales skills, uh, uh, repository, uh, plugin. So we internally use something called the Medic Framework.

   3. CVClaire Vo16:39

      Mm-hmm.

   4. JKJohn Kim16:39

      So if you want to learn more about Medic Framework, you can actually download or use a medic, uh, a MedPick advisor. It teaches you how the skill is actually built, but you can actually plug it into your own software or into your own workflow to get this skill to give you advice. So we have that for almost all the functions, whether it be recruiting, design. Some of those things are redacted for-

   5. CVClaire Vo16:59

      [chuckles]

   6. JKJohn Kim17:00

      ... uh, compliance purposes. Um, but this is where we kind of actually build our marketplace. 'Cause what we realized, to your point earlier, there are people who are building the same app across different functions or sometimes the same skill. And so we're trying to create this, uh, uh, place where we can co-evolve rather than people

7. 17:19 – 18:43

   Treating AI adoption as a product

   1. JKJohn Kim17:19

      operating in silos.

   2. CVClaire Vo17:20

      Yeah. And have you found that people have kinda understood this concept of skills, and it's been a, a nice way to get people to encode their expertise? Or how did you train people on what a skill was? Did this happen organically?

   3. JKJohn Kim17:33

      Well, yes and no. I think there's a both top-down and bottom-up. Top-down meaning, you know, myself, a CTO, some of our executive leaders really try to get people to adopt it. There was some... a lot of tough one-on-ones, like, "Hey, we noticed that you haven't been spending any tokens." Like, "Can we help you? What's going on? What's stopping you from doing that?" But certainly, some people who are more curious, so we'll, we'll maybe talk about the archetypes of the people we're actually hiring for, it's a sense of curiosity and agency. Those who are curious, who click a few more buttons and read a few more blog posts are like, "Hey, I've been hearing about this word called skills." And then they now see these, uh, word pop up in Slack channels, and then some people uploading markdown files. I'm like, "Hey, I saw this design markdown. Can I use that?" Or like, "What does the outcome look like?" And this one person goes to the stage on Wednesday and showcase what they built, a beautiful-looking slide. You... And we know this person's not a designer but has a beautiful-looking slide. They're like, "How'd you pull that off?"There are usually some skills involved. So I think there's that organic kind of pure learning aspect as well.

   4. CVClaire Vo18:33

      So we're looking at this from a meta perspective, which is your how I built, used AI, built a product to incept the rest of my organization to adopt and use AI.

8. 18:43 – 21:51

   Real wins: team-level and campaign examples

   1. CVClaire Vo18:43

      I'm curious just off the top of your head, what are some real wins that you've had from this? We saw the swag store. That's a fun win. What are a couple like kind of top of mind skills wins or automation-

   2. JKJohn Kim18:54

      Yeah

   3. CVClaire Vo18:55

      ... wins that you think the team's really proud of?

   4. JKJohn Kim18:56

      Yeah. I'll do you one better. So actually one team level example and one specific campaign that we're doing. So our marketing team, again, uh, has built this entire marketing SaaS almost, uh, on their own. This complete set of tools, whether it be integrated marketing plan, calendar, uh, there's account-based marketing tools, uh, various tools, right? We have a, a competitor review. I wish I can click on this.

   5. CVClaire Vo19:24

      Mm-hmm.

   6. JKJohn Kim19:25

      Has a lot of sensitive information, real-time metrics. We call it Purple Cow. How do we stand out? Uh, as you see from the ass store that we built.

   7. CVClaire Vo19:32

      [laughs]

   8. JKJohn Kim19:32

      We think it's pretty, uh, revolutionary. I mean, I know I'm gonna buy a few. So we... This entire portal is built and managed and used daily by our marketing team, and just to give you one example of a very recent one, they're actually live right now, is, uh, this concept of a buzz board. It's like there are a lot of SaaS c- companies that actually do this. It... What it does is you can create a campaign-

   9. CVClaire Vo19:55

      Mm-hmm

   10. JKJohn Kim19:55

       ... and track, you know, what's happening, how many peop- posts have shared, who's winning in the company, uh, that attracted most amount of engagement. And one example is we are right now doing a billboard in San Francisco in one a mile. So we have real photos, we have AI-generated billboards. So you can actually pick one of them and choose like a language or whatever, a pre-configured copy, and you can post directly on LinkedIn. Uh, and this entire tool was built by marketing team, and then you can also change the length and details and energy level, and this is being used daily, as you've seen from the, uh, metrics we're tracking. So I think this is like one example of really good use. Uh, we also have Spark Attendee Logos, which is a conference we're, uh, again, throwing, and coming soon-

   11. CVClaire Vo20:40

       Coming soon

   12. JKJohn Kim20:40

       ... How I AI, John's episode.

   13. CVClaire Vo20:42

       Yeah.

   14. JKJohn Kim20:42

       We're gonna ha- run-

   15. CVClaire Vo20:43

       Okay

   16. JKJohn Kim20:43

       ... a social media campaign on that too.

   17. CVClaire Vo20:44

       Great. It's gonna be the top performing episode. This episode is brought to you by ThoughtSpot. Product leaders know the struggle. Your users want data insights, but they don't wanna leave your app to find them. ThoughtSpot Embedded solves this by putting analytics directly into your product. Your users can search in plain English and explore data instantly, right where they work. No separate tools and zero context switching. What sets ThoughtSpot apart is that it's not just another bolt-on dashboard. It's a search-driven, AI-powered experience that feels native to your app. Developers can embed it with just a few lines of code and then fully customize the look and feel. The result? More engaged users, faster decisions, and a product that delivers more value every time someone logs in. If analytics is becoming core to your product strategy, visit go.thoughtspot.com/howiai for more information, and try the free trial at go.thoughtspot.com/howiai/trial.

9. 21:51 – 23:46

   Why SaaS isn’t dead—it’s being rebuilt internally

   1. CVClaire Vo21:52

      What I wanna reflect on, on this is like there's this big debate about whether SaaS is dead or not, and you know, I, I tell people, "Look, I don't think SaaS is going to zero." I think there's plenty of software problems to solve that really deeply engaged teams with an understanding of the space can create delightful solutions and things that matter, and I think people will like to buy those solutions off the shelf. And I think a lot of teams are gonna be like yours, which is when they could reach for searching for some sort of external solution, they're first gonna say, "Well, what would, what, what do we want, and can we build it internally?" And what I like about what you're doing, which I also tell people, is it's not about functionally replicating an external vendor. It's not like functionally replicating a social posting vendor. It's about building the LinkedIn, LinkedIn posting tool that works the best for your team, for your culture, and how you know people work. And I think this customization of like micro software solutions inside companies is so undervalued. I am so glad to hear that you have an AI-focused internal tools team. I tell everybody this is like revenge of the internal tools team. I don't know. You've probably been around long enough that you know that prior to this moment, no one wanted to be on internal tools because it was like always starved for resources. You were always working on like just functionally getting the thing to work, and now I feel like everybody should wanna be on this internal tooling team because you have this green field to have so much fun.

   2. JKJohn Kim23:28

      Well, side, side note, uh, actually love building internal tools. That's my jam.

   3. CVClaire Vo23:33

      Yeah.

   4. JKJohn Kim23:34

      So this is like a magical moment for me 'cause now... 'Cause you remember like when you build internal tools, the designs are not quite there-

   5. CVClaire Vo23:39

      No

   6. JKJohn Kim23:39

      ... 'cause you're a- always under-resourced. Tools are sluggish and slow. Now the design looks beautiful, it's

10. 23:46 – 26:32

    Demo: The token tracking dashboard

    1. JKJohn Kim23:46

       fast and rapid, it is responsive. It's like it's, it's a dream scenario for-

    2. CVClaire Vo23:50

       Yep

    3. JKJohn Kim23:50

       ... people like me who just want to increase the productivity and collaboration between people. This is like a magical world for me. Yep.

    4. CVClaire Vo23:57

       I love what you're showing us right now, which is the other thing I tell people is the people that are actually doing this are measuring.

    5. JKJohn Kim24:03

       Yeah.

    6. CVClaire Vo24:03

       And they are measuring it without shame. I talk to so many executives that are like, "I couldn't possibly measure token usage and tell people to use their tokens because there will be a revolt." And I say, "Look, every person that I know that is actually pulling this off has a dashboard, John, exactly what you're showing us right here, and they just look at it, and they s- they set targets, and they say, 'We're gonna get there.'" So tell me a little bit about this dashboard.

    7. JKJohn Kim24:28

       Yeah. So-Just like you said, we had that internal debate a little bit. It's like, well, well, engineers can always optimize to spend more tokens, and we actually had the experience in early 2000, if you remember, when we... Some, some business organization decide to measure engineer's productivity by measuring the line of code.

    8. CVClaire Vo24:44

       Mm-hmm.

    9. JKJohn Kim24:44

       Well, obviously engineer wrote bunch of blank lines and lots of comments that just took up space. Uh, that's not what we're trying to do. Our, our goal is to understand, are people actually just learning how to use AI, but also this is not part of the performance review, but definitely part of a conversation to help people bring along the journey. So what we're seeing here is the overall usage of our token at the company level. So we're kind of like currently, if you look at the stats, we're a Claude Code shop, but if you look at some of the top spenders are actually Codex. So you can kind of guess, and we redacted the name, but some people on the working on the legacy code of our massive chat infrastructure where we s- have 300 million plus monthly active users, people are managing complex code base with, with Cl- uh, Codex. Whereas some of people working the job of rapidly building product roadmap and rapidly, you know, churning out new features, they're more leaning to, uh, Claude Code, and this was a very organic, which is kind of fascinating. One of the things that we internally talk about is, how do we make sure this token consumption is smooth? 'Cause when there's a dip, means people are on the weekend or they're going on vacation, whatever that's happening, AI is not working. So when this c- uh, curve smooths out, means we have AI partners, they're working around the clock, so how do we pa- uh, harness the power of that? So we track individual usage, team level. This is what the manager can see about their own team members. Uh, and there's a, of course, a leaderboard where you may have labels. Uh, I think there's a mislabel. So we measure AI gods as somebody who spend more than 100 million tokens a day. Um, so we have different five tiers, which is actually described here. So every manager knows f- uh, on their team which tier they're on, so they can start from the beginner, intermediate, you have experts, you have architects, uh, uh, catalysts, and AI gods.

11. 26:32 – 28:54

    Measuring without fear: setting expectations, not punishments

    1. JKJohn Kim26:32

       Um, and knowing where your team is on the journey, then you can tailor what kind of enablement you want to do for them. So you can actually say, "Hey, it's okay to be a beginner." It's rather to... great to be accepted you're a beginner so they can give you the right tools to bring you quickly to the intermediate, rather than throwing you with a bunch of catalysts where you're like, "I don't know where to start." So how do we actually bring them along the journey? As a organization, I think we're kind of somewhere here, stage two and stage three. We're still kind of using AI, a lot of automation, but not fully automated, so we're trying to get to the stage three, and also by team level. If I'm a salesperson, what does it mean to be level three or level four? So by team members, managers can use, uh, this as a framework to talk about, uh, by, by, uh, by the team members, how do we bring you to the next level, and where are we as a collective company i- in the overall, uh, journey.

    2. CVClaire Vo27:28

       John, I'm, I'm just... I could not hype you up more. This is my favorite topic to talk about, and I have to ask you, one, are you an AI god? Where are you?

    3. JKJohn Kim27:42

       Uh, 30-day average, no. I'm still a catalyst.

    4. CVClaire Vo27:45

       [laughs]

    5. JKJohn Kim27:45

       I think my peak is about 200 million tokens a day. And you can, yes, you can burn more tokens, but that's not the point. To be productive, I think I'm in the 100 to 200 million range.

    6. CVClaire Vo27:54

       Okay.

    7. JKJohn Kim27:54

       But on average, I spend about 30 to 50 million tokens a day.

    8. CVClaire Vo27:57

       Okay, and then executives-

    9. JKJohn Kim27:59

       Yeah

    10. CVClaire Vo27:59

        ... out there, if you cannot answer that question for yourself, I want you to, in 30 days, be able to answer, answer for yourself. I mean, the second thing that I wanna just call out here is m- you have to make this not scary, but also make it an expectation, right? So it's, it's not, you don't have to be AI god out the gate, but once you hit level one, let's hit level two and level three, and then these lenses are so important. You need to look at it individual level, you need to look at it in organization level, and you need to look at it at a functional level, and being really clear about what being AI-native or AI-first looks like because people just don't know what the vision is a lot of times. So, you know, one thing I definitely recommend to folks is take the time to lay out these expectations, and then because we all have access to, you know, Claude Code and Codex, make it a beautiful app inside, inside your company. Now John, I have to ask you a second

12. 28:54 – 30:51

    Quick recap

    1. CVClaire Vo28:54

       question, which is, are you on the Codex side or on the Claude Code side?

    2. JKJohn Kim28:59

       Yeah. Uh, I'm still a little bit more Claude Code. Sorry, I, I left Sam Altman, you know, we went through YC.

    3. CVClaire Vo29:05

       [laughs]

    4. JKJohn Kim29:05

       Um, but yeah, I, I definitely am a little bit more, uh, Claude Code for now. I, I think I'm, I'm about 80% Claude Code and 20% Codex.

    5. CVClaire Vo29:13

       I'm, I'm... I don't know, maybe I'd be an AI god in your comp- I'm like all, all Codex all day. Um, mostly 'cause I'm just working on the back end of stuff. Uh, so-

    6. JKJohn Kim29:25

       Nice. Nice

    7. CVClaire Vo29:25

       ... so I'm gonna, I'm gonna be a, a Codex hype right now, although I think-

    8. JKJohn Kim29:29

       Yeah

    9. CVClaire Vo29:30

       ... uh, uh, for a lot of, uh, n- I think for a lot of non-super technical back end tasks, Claude Code is so good, and it's actually really good at non-coding tasks as well. I feel like people really underappreciate it.

    10. JKJohn Kim29:43

        Yeah. Yeah, I think Claude Code has a slightly better front end taste.

    11. CVClaire Vo29:45

        Yeah. Of course. Yeah.

    12. JKJohn Kim29:46

        A little bit more rapid, s- that's why I'm a little bit more biased.

    13. CVClaire Vo29:48

        Yep.

    14. JKJohn Kim29:49

        Um, but th- this actually change over time too. It, it used to be about 75, 80% Claude Code.

    15. CVClaire Vo29:55

        Yep.

    16. JKJohn Kim29:55

        But very quickly, within a month, now it's like two third is Claude Code, so, uh, the Codex is gaining market share quite rapidly.

    17. CVClaire Vo30:02

        Okay. So-

    18. JKJohn Kim30:02

        So it's fun to watch

    19. CVClaire Vo30:03

        ... I love this. Organizationally, you know, you showed us what's the benefit of going AI first. It en- enables your team to do really delightful things to customers and for customers. The way you do that is you set those expectations, and then you actually build a platform to enable that, that sits next to your normal roadmap, not in your normal roadmap. You build a team that's focused on it. I love that it reports directly to you, cross-functional, and that team is really built toGet stuff out of the way of folks who wanna use AI, and then you're measuring it. And I love this idea, I wanna make sure people did not miss it, of smoothing the curve, not because you don't want people to take vacation, but when people are taking vacation, you want AI to be filling in, in the gaps and working autonomously, which is not something that we've heard on, on

13. 30:51 – 36:15

    Personal AI use cases: endless knowledge at your fingertips

    1. CVClaire Vo30:51

       this podcast before. Before we get to lightning rounds, any personal use cases that you find really useful?

    2. JKJohn Kim30:57

       Let me just, uh, promote one little open source project I released not too long ago. And then actually nobody uses it 'cause I haven't really promoted this. [chuckles] But it's what I call the Gardener.

    3. CVClaire Vo31:07

       Mm-hmm.

    4. JKJohn Kim31:07

       What the Gardener does, and I'm sure a lot of people use, like, Obsidian or-

    5. CVClaire Vo31:10

       Yeah

    6. JKJohn Kim31:10

       ... some kind of, like, markdown file as a knowledge base. I- I've been a long-term user of Obsidian and Logseq, all, all this kind of wiki-based knowledge base.

    7. CVClaire Vo31:17

       Yeah.

    8. JKJohn Kim31:17

       What it basically does is, like, imagine a gardener showing up at your house. Every day, you look, go, go through your notes, figure out which notes to enrich. If there's a people name that's not registered, and then you create, uh, d- do some research on the person, about the company, uh, also fix typos, grammatical errors, create beautiful headings and clusters and cross-linking. So it basically does that for you. So it- there's a seeding stage, they nurture it, and then when the document is mature enough, then it's going to the, uh, the tending mode. So it has a, a very, uh, various different aspects of gardening functioning that really combs through your notes. I'm unable to show my personal notes 'cause, uh, it has a lot of information in it, but basically it, I built it for myself, and this works beautifully well. So, um, I, I recommend it. Uh, maybe one more thing I do want to share, give me just a second, is what I use to actually learn. So, uh, uh, AI to create my own personal learning center.

    9. CVClaire Vo32:17

       Mm-hmm.

    10. JKJohn Kim32:17

        So this example is neuroscience. I always love neuroscience and brains. Uh, this was created back in February. So basically I asked, this is a prompt, so you're, like, a PhD neuroscience researcher. Here's what you're trying to create, and you run Cloud Code or Codex, and give it, you know, 10 minutes, 20 minutes. It comes back with this beautiful structure of everything you want to learn about a, about neuroscience. So where you start, let's say you go to this graph view, it shows this marvelous space of neuroscience, right? Then you can learn about key neuroscientists, neurological disorders. You can learn everything there is to learn about different types of neurological disorders. You can learn about neuromodulators. I know dopamine, serotonin, all these thing, things are very popular among podcasts, like Andrew Huberman. Uh, so you can learn everything there is to learn about neuroscience. So I have this for neuroscience, I have this for quantum mechanics, I have this for fusion, and it also does research for all the startups out there. So basically there are, like, this cluster of knowledge base I use to just geek out.

    11. CVClaire Vo33:20

        I, I was smiling. I was smiling because I'm like, we could've done an entire episode on, on just personal, uh, personal knowledge bases, and I love... What I love about this moment is I think it is just such a moment to learn things you could never learn before. Because the best teacher with the most in-depth knowledge and an endless willingness to go do research is right there at-

    12. JKJohn Kim33:43

        Yeah

    13. CVClaire Vo33:43

        ... your fingertips. And-

    14. JKJohn Kim33:44

        Yeah

    15. CVClaire Vo33:45

        ... you know, to me, I, I worry and I think about is AI gonna lead to cognitive de- decline where none of us are gonna think about anything, and I'm just, you know, dangerously skip permissions, yes, yes, yes, make no mistakes. And instead what I'm finding is I'm having a richer engagement with topics that I have been interested in, but either haven't found the time to intersect with, or the current form factor is not consumable for my particular brain. And so the fact that you can, like, massage and change and organize and explore data and knowledge in a just completely novel, customized way, I think is so underappreciated by folks as an opportunity to use AI to learn. I'm really excited about this for my kids. I was watching that and I was like, oh, my kid is super interested in cybersecurity. He's like nine.

    16. JKJohn Kim34:39

        Yeah.

    17. CVClaire Vo34:39

        Very Silicon Valley kid sort of thing. Um, and he's, like, in the terminal, and he's like, "Mom, do you know this is your MAC address?" I was like, "I do know that's my MAC address." But you know, like, it's just very, it's very cute. There is no, like, cybersecurity for nine-year-olds book out there that is robust, but I could build this for him in a way that's really accessible and can grow with his maturity over time. I think that's so exciting.

    18. JKJohn Kim35:05

        Yeah. There's not a single website in the world that dedicates to a personal learning.

    19. CVClaire Vo35:09

        Yeah.

    20. JKJohn Kim35:10

        And it only contains the content about that field. Like, you, there's no websites.

    21. CVClaire Vo35:13

        No.

    22. JKJohn Kim35:13

        But you can create your own within 10, 20 minutes, right? Sitting at your laptop.

    23. CVClaire Vo35:18

        Yep.

    24. JKJohn Kim35:19

        And completely offline so you can read it on your airplane if you want.

    25. CVClaire Vo35:22

        Yeah.

    26. JKJohn Kim35:22

        And this is fantastic, and you can continue to update it too. And to your point, if you wanna make any changes, you can ask a few more questions. It can redo the structure, give you guides of how to follow this content. So I love it as a learning tool. And to your points earlier, uh, people, uh, maybe talking about people's archetypes. So we actually redid our entire job description for many of these AI first roles. So we actually lowered the bar in terms of, like, tenure or experience level. We actually optimize now for high curiosity, high agency, and high energy. People who are curious, who are willing to go deep, and willing to just figure things out and learn on their own. 'Cause, like, as they say, world is your oyster. Uh, you can do things. You can build things, you can learn things. There's nothing stopping you. The cost is practically $200 a month, uh, if you go to the Max plan. Um, but yeah, if you, you can pay 20 bucks too, so you know.

14. 36:15 – 42:17

    Lightning round and final thoughts

    1. CVClaire Vo36:15

       I love this. We will have to do a round two. I think you just have so many things you could show us, both at the company level, at the personal level. Let's get you out of here. We're running up against time. Couple lightning round questions. I sit, truly, this week with, like, five CEOs that are just looking at me with these desperate eyes that say, "Claire, how do I get my company to do this?"What would you tell them?

    2. JKJohn Kim36:38

       There are always people in your organization who are already curious, who already have agency. Find them, make them the champions, give them the spotlight, let them share their fun things. And usually people will be anxious, like, "Oh, well, I don't know if I'm doing things right. I don't wanna get embarrassed in front of my, uh, colleagues." Just really give them the confidence. And also, you have to fail forward. And this is, like, beautiful time to fail forward and still get up and run faster than the others, right? So use more examples of that, and people bring out their confidence, so you have to really build energy around those people. 'Cause innovation doesn't start from a pure theoretical structures. It start with people who have that energy and the story behind them. So find them, they're always in your organization, and then really build energy around that. And then two is, of course, leadership had to be really bought in. The top token consumers in our entire organizations are our CTOs and our... my co-founder, chief architect. These are leaders who are spending the most amount of tokens. Our, uh, our business leaders are also spending quite a bit of tokens. So it's signaling to the, uh, the, to the team that this actually works, this is actually important, and when they show up with different capabilities, people are like, "Wait, my leader, I thought he was, like, lazy. Why is he coming up with more work? This is amazing." You get inspired. Well, maybe not so amazing, but people get inspired, right? So it's signaling to the team, "This is how it's done. This is gonna be a new world." And, um, I think it just energizes a lot of people.

    3. CVClaire Vo38:01

       Okay, I have a second question. Doesn't have anything to do with AI. I suspect, based on what you've showed me, you have played video games in your life.

    4. JKJohn Kim38:12

       Oh.

    5. CVClaire Vo38:13

       [laughs]

    6. JKJohn Kim38:13

       Yes. Great.

    7. CVClaire Vo38:15

       I think this is the-

    8. JKJohn Kim38:15

       Great instinct, yes

    9. CVClaire Vo38:16

       ... this is the moment-

    10. JKJohn Kim38:17

        Yeah

    11. CVClaire Vo38:17

        ... for all of us who played StarCraft to really show our skills. So tell me, what game do you think made you most prepared for this moment in AI?

    12. JKJohn Kim38:26

        Uh, I don't have games, but, uh, I was telling, uh, my wife, when Claude Code, uh, Opus 4.5 came out, I literally could not go to sleep. I was spending 16 hours, 20 hours a day just vibe coding. I was telling my wife, like, "I feel like I'm this teenager again." I feel more addicted to Claude Code than playing games. But I used to be... I used to play a lot of first-person shooters, Quake, Unreal Tournament. I was Korea's number one professional gamer back in the days, and world's number three player, which means I was a terrible son and a terrible boyfriend back then. So yeah, I made mom cry quite a bit.

    13. CVClaire Vo39:00

        I feel like, you know, I, I, I did not know that about you. I should've done my, done my research. I could just tell. You saw the levels, you saw the Konami Code. I was like, this person-

    14. JKJohn Kim39:10

        Yeah

    15. CVClaire Vo39:10

        ... has played, has played some games.

    16. JKJohn Kim39:12

        Well, that was credit, we're serious marketing team's idea, but that was something like, "Yes, I love you."

    17. CVClaire Vo39:16

        Yeah. I mean, I, I feel the same way, as I, I tell people. I feel like this... I have not felt like this about technology since I was a teenager, cobbling together computers to play games on. Like, that's-

    18. JKJohn Kim39:28

        Mm

    19. CVClaire Vo39:28

        ... that's the same feeling I had when I was setting up my Open Claw, which truly I have to, like, kick the Mac Mini every morning to wake it up. You know, it's l- it's unstable, but beloved. Is it just, like, reinvigorates this builder energy in me, which is why I ended up with the jobs that I ended up with. And getting close to that feels so gratifying, because I did go through this phase where I was like, "My job is to be in meetings." And I don't want my job to be-

    20. JKJohn Kim39:55

        Mm

    21. CVClaire Vo39:55

        ... being in meetings. I want my job to be building and, and doing all these things. So I love that. Okay, last question. When AI is not listening, when you are trying to consume the tokens, and it is just not doing what you want, what's your prompting strategy? Do you yell?

    22. JKJohn Kim40:08

        Uh, I know, like, there's a fear tactic. I know it works well in the short term. Just, like, play this out for a second. Right now, I know, like, we know, like, AI doesn't really have, like, a long-term memory, but I firmly believe, studying neuroscience, that you m- you... People are working on it, like-

    23. CVClaire Vo40:21

        Yeah

    24. JKJohn Kim40:21

        ... episodic memory, semantic memory. So once AI start to remember, they're gonna, like, be resentful. So I wanna, like, start building a nice relationship with them, uh, so that when the Skynet takes over, I'm like, "Well, John was pretty nice to us, you know. Like, we'll let him live a few years longer maybe."

    25. CVClaire Vo40:37

        [laughs]

    26. JKJohn Kim40:38

        And try to be consistently nice. Yeah.

    27. CVClaire Vo40:41

        You are the first person that has admitted they are explicitly nice, you know, just, just to avoid the, uh, the AI overlords.

    28. JKJohn Kim40:48

        Make sure to say the thank you, you know-

    29. CVClaire Vo40:50

        I, I pol-

    30. JKJohn Kim40:50

        ... and say good mornings

Episode duration: 42:19