In this episode of "Beyond The Prompt," we delve into the fascinating world of AI and its practical applications in business with JJ Zhuang, Chief Architect at Instacart. JJ shares his journey into AI, sparked by the groundbreaking development of AlphaGo by Google DeepMind. He discusses how this event, combined with his experience at Microsoft, shaped his understanding of AI's potential. We explore how Instacart harnesses AI to revolutionize not just grocery shopping but the entire customer experience. JJ highlights the company's innovative AI initiatives, including the Care AI team's efforts to answer the perennial question, "What's for dinner?" He also delves into the transformative impact of generative AI and large language models on both consumer-facing applications and internal workflows. JJ provides insights into Instacart's adoption of AI across various teams, from marketing to legal, through their AI Catalysts program. He also touches on the creation of an internal AI-powered tool, Ava, integrated into Slack for streamlining communications. This episode offers a deep dive into how AI is shaping the future of e-commerce and the broader business landscape, making it a must-listen for entrepreneurs, business leaders, and AI enthusiasts alike.
📜 Read the transcript for this episode: Transcript of JJ Zhuang, Instacart: "How AI can solve the 'Whats for dinner' problem. |
Â
[00:00:00] Henrik Werdelin: Welcome to the latest episode, where we explore the intersection of AI and practical innovation in the business world. I'm Henrik Werdelin, alongside Jeremy Utley, and today we're joined by JJ Zhuang, the Chief Architect of Instacart. In this episode, we delve into how Instacart harnesses AI to redefine not just grocery shopping, but the entire customer experience.
I'm JJ Zhuang, Chief Architect of Instacart. I lead the Carrot AI team to solve the "What's for dinner?" question for our customers.
[00:00:30] Jeremy Utley: First of all, welcome to the show! Thank you so much for being willing to chat with us. Congratulations on the IPO and your success.
[00:00:38] JJ Zhuang: Yeah. Thank you for inviting me. It's been, I've been looking forward to this.
[00:00:43] Jeremy Utley: So what we thought would be fun is to actually start with you and your journey, JJ. Cause I know you've had an amazing career and you've you've been at the forefront now of a lot of AI building, and we thought it'd be fun to hear from you when was the first time this new technology came onto your radar and you thought there might be something worth paying attention to here.
[00:01:08] JJ Zhuang: Yeah if if we're talking about AI because these days when people use the word AI, they may mean specific things. And if we talk about AI broadly it has a long tradition. And I actually came to the space relatively late in my career. A lot of AI researchers and practitioners, they started way early.
I'm always I was always curious about it, but didn't really get into it until I joined Microsoft because Microsoft had a huge team of air researchers and practitioners, so I got a chance to work very close with them and start to learn the basics and start to have my own appreciation of, wow, there's this universal algorithm that can learn anything, specifically with deep learning and neural network.
I think, if I want to pinpoint it to a specific thing, it's probably what year was that? I think it was 2017 when Google DeepMind developed AlphaGo. And AlphaGo defeated Lisa Doh of Korea. And I happen to be a Go player of all things.
[00:02:21] JJ Zhuang: And,, for Go players, this is it's unimaginable. Because the goals was supposed to be this game, even though it is, just like any game, there are rules, right? So you think that I should be able to solve it. But because the universe of goal is so deep, so vast, it was considered a unsolvable problem.
And then all of a sudden the best human player in the world got defeated soundly. That was a moment when I, finally realized, not only AI is going to be probably the most important technology of our time. It's also The impact is probably going to, uh, change the world much sooner than we all think.
So that, that was like my aha moment. And since then, I, uh, put a lot of effort in really getting involved and learning more.
[00:03:21] Henrik Werdelin: There seemed to be this period of time around 19, 20, 20, where genitive AI became more of a thing. And so there was always for me like this. Difference between the hardcore research of deep mind and, as you're saying, like the neural networking, and then when suddenly it became something that seemed to be passing the Turing test and it became something that kind of, even non scientists could play around with.
Do you remember the first time you got your first kind of play with OpenAI or something else where you're like, you know what, this is actually becoming something that everybody can start to use.
[00:04:03] JJ Zhuang: Yeah, for generative AI and large language models specifically uh, I wasn't actually aware of its existence until GPT 2 so when transformer paper came out I actually didn't pay as much attention, probably should have but when open AI GPT 2 came out I was amazed by its limited ability at a time, but I started to see, oh, there's a glimpse of this thing that can understand language better than the traditional methods.
And then certainly GPT 3 just made it out of question, right? You can start to build a real world applications with GPT 3 and a lot of companies start to do that. But I think that for the public, it's really, GPT 3, 5, especially chat GPT, uh, in Late last year that was like a bombshell.
[00:04:59] Jeremy Utley: Yeah. Yeah. . Is it, does that have implications on if you think now about your team
[00:05:04] JJ Zhuang: j Yeah, and maybe before I even talk about my team, this is a fascinating topic. I should have mentioned that the bar for interacting with computers it's always, it's a piece of cake. On this long, very, on long trend of becoming more and more accessible to average people.
And I still remember the windows 95 moment when computer became really, truly accessible to average users. But before that, because, it was considered an arcane thing, and then programming the same, the earliest programmers, they need to deal with machine code. And assembly language, all those things, and it really, it was not easy, even for the professional programmers, how much code, I think, can they produce in a day, they can do very little things, but that trend is really accelerating I think years ago with the, used to, we used to think that the mobile programming was hard, right? But then it was made accessible because all of a sudden with a react native, you can write JavaScript to program mobile. And today, prompt engineering, it just becomes like another thing where you can really create.
Programming layer using just a playground and we actually start to adopt that at work as well. And in terms of my team. The reason I joined Instacart is because Instacart has such a rich AI application space. We all, we use AI in almost every area of the business. And then we have this well established algorithms team that takes on the most difficult problems in the company and have the central resource to solve them.
Most of the people on that central ML algorithms team, they all have very deep ML background. anD that's. I think that before large language models, when we talk about AI projects, usually involves of dealing with a large amount of data training models and two models, right?
So it's a very long process to require a specialized skill set. Since late last year, early this year, the. It when we now talk about building a project, it completely changed the nature of that, because a lot of it started to become just a generic backhand programming. We now have many more generalists, basically people who don't necessarily have deep ML background on the AI team.
And there's a new definition of what an AI engineer is, right? It is somebody who understands how to interact with these models, uh, using API calls and just generic traditional way of building software is you just need to build good software and that it, the only requirement is you need to be faster learner because the field, it just changed so, so much, so quickly
[00:08:23] Henrik Werdelin: I wonder if we look at just the speed on which. Non, ML specialists can now use AI. And we talked earlier about using Sabian and like how it's being democratized.
Do you think that kind of like thing is going to happen throughout the whole organization? Are you going to start to see the the marketing team, the communications team, the the finance team increasingly become the people who are also using, uh, AI, or do you think for a while, it's still be something that quote unquote belongs to the data team?
[00:08:59] JJ Zhuang: Yeah, that's a great question. Instacart is we're fortunate that we have a our entire company, we have a very forward thinking team. So when early this year when generative AI became the biggest news, the entire company mobilized and, uh, every function is asking themselves this question.
What can we do to Stay ahead of curve to adopt the best tools to, you know really rethink the way we work and that does include many of the functions you mentioned, such as legal and marketing and communications, all the teams we established this internal program called AI Catalysts which is is led by Early adopters of these tools, right?
And the catalyst team not only includes engineers, but also, product managers. And, uh, I even mentioned legal and certainly marketing and communications and they are always evaluating what's already on the market and what we can adopt. I think that The number of applications currently we're reviewing is around the scale of 50 and certainly it's a very fast moving field because every day there's probably another 100 coming up to the market, so it's hard to keep up, but at the same time, very selective.
We want to adopt it the best tools. thE difference. I think that's between the engineering team, let's say. And all the data science team compared to functions such as marketing, communication and others is it one is different tools by engineers looking at how to make them more productive using, for example, coding tools.
The other is also the technical teams tend to be more forgiving in terms of the quality of the tools. For example we adopted a GitHub copilot very early in the year, and by now it's extensively used by engineering team. Certainly GitHub Copilot can generate a very convincing code, but very often it can also be wrong.
So you need a level of fault tolerance built into your own ability to be able to, deal with this, right? And then not just blindly accept everything it generates. I think that the other side of the spectrum is not just, some functions in our team, but also... Let's say that when we build for our consumer our business users, right?
Our consumer users of Instacart, then the product needs to behave really well, right? They need to be perfect because if we generate wrong information, then people take the wrong advices. So yeah, it, it really varies. But yeah, to answer your question, We believe that I should apply to all functions and everyone should be using it.
Are there,
[00:12:07] Jeremy Utley: Mechanisms that you have, you talk about these AI catalysts, how do folks surface new opportunities or what are the ways in which you interface with these catalysts? Is it just an ad hoc kind of group or are there meetups or there's slack channels? How do you stay, you said you're exploring 50 applications now and there's hundreds new being developed.
How do you remain in touch and surface opportunities?
[00:12:32] JJ Zhuang: Yeah, I think that there, there are several things. We have a internal organization earlier this year we, when AI start to become this new wave, and we recognized that we mobilized the to, to form a new organization called the the Carrot AI team.
Carrot is our company Vegetable. So a lot of the important projects at Instacart is called. Carrot something. Yes, exactly. Laura's team designed that logo. I really love that logo. So Cary AI team serves as this central expertise team. I know the team itself also build products and we're working on many different products that's AI powered.
In the Instacart app but at the same time, Keri, I team also serves as advisors. To do other functions and other teams, and we have a federal model because it's really hard to just to limit to the a I work to just a few people, right? We wanted the entire company to be thinking about how to use a I in their work as well as how to build that into our product.
So they also have the advisory function to you. All the other teams and we have a fairly structured the way of doing it. Certainly we have slacks channels and so on. But there's also more formal programs reviews and try to find the best way to support the scale.
[00:14:07] Jeremy Utley: Is it something like a PR FAQ kind of Amazon's famous, new product development process when you say you have a fairly structured approach, what's the, what are the stages that that a new idea might go through and how do the central team?
I love that you said that they serve as advisors. What does it look like to be an advisor to a team
[00:14:27] JJ Zhuang: with an idea? Yeah, There are several things you asked. One is how do we do ideation? Where does the idea come from? It depends on whether we're talking about building products for our customers or building internal tools or adopting external ones, right?
So for building for our own customers for Instacart applications. It's what number one is we have way too many ideas. And so it's really the d difficult part is how we narrow down, how do we prioritize with the, with limited resources, right? And on that front, the reason that we have so many ideas is because our space is food and it just ments that food is the most important topic for humans.
It so
[00:15:20] Jeremy Utley: Everybody's got an idea
[00:15:21] JJ Zhuang: about food. I know and sometimes we, anecdotally we know that when people talk to chat, GBT or binging or Bard, food is always a topic. And one of the biggest topics early on this year our CEO Fiji Samo she.
authored this kind of AI manifesto for the company, and it's called what's for dinner. And I sometimes jokingly tell people that it's the second most fundamental question for homo sapiens. The first one is why are we here? And then once we answer that question, then now it's for dinner. And it's just such a rich topic, right?
Because everyone want to eat healthy and want to have variety, want to try something new. And but at the same time, they also have their, their Different level of cooking skills and also manage a budget. So when you have this kind of rich topic, and we know that it's relevant to everyone uh, the complexity is exactly the right fit for applying AI.
And and the other dimension is Instacart is a business. We very often think, it's about saving people time, but it used to be that for the last 10 years, it's all about how we save people's time so that they don't have to go to the store to do the physical shopping themselves.
And now with this generative AI. We can also save them time and save them mental labor of planning for the week. So those are the core threads we have. Once you pull that thread, it starts to become very clear oh, if this is a problem we want to solve, then immediately there are concrete ideas, right?
For example, we know that now we can do meal planning. And large language models knows a lot about food and knows a lot about, what's, what to eat and health aspects, all those things. So we can leverage that. Once you get into meals, then naturally started to come, then people shouldn't be buying just one thing at a time.
Which they should think at what to eat level, right? So recipes start to become really relevant, but it's not just any recipe, you need to map the recipes to exact right ingredients of the right quantity. And also. Available in your local store. All these things a I can solve very well.
And also, once you get to into recipes that become we don't just randomly want to pick recipes for people, right? You guys know this. Usually people learn and adapt. The best when it's slightly, the change is slightly outside of their comfort zone. And so we basically also adopted the same methodology, which is, we look at to what this user always have been buying and What's already currently in their cart, and this is the best way how we can surface the most relevant recipes.
So when you start to pull the thread, all these ideas become very clear. And those are the things we built. Now, on the internal team side. We also look at what people's workflows like and where they spend the most time. And then figure out how to make them more productive.
[00:19:04] Henrik Werdelin: Folks, how do you do that?
That sounds like a fascinating exercise.
[00:19:09] JJ Zhuang: Yeah. I can give you an example. So we have a, uh, internal developer productivity team and their work is to bring in either commercial tools or build the tools in house to support the work of engineers. And that team is very data driven, so they look at people's workflow using different touch points in every day's work of an average engineer, and then look at where they spend the most time.
For example, a couple of years ago, when they saw that a lot of the time is spent on trying to test A piece of code, they write in the larger system, but because, everyone is testing in the same playground. It creates a lot of contention, right? So they built this new tool called a bento, which is to virtualize the environment so that, you can isolate.
evEryone is testing so that there's less contention that makes people, run much faster. And it was a I, I think that they also looked at, uh, for example, using GitHub copilot example we rolled out to initially, 100 engineers and then the team just look at all the stats, right?
Like how many lines of code that get upset versus rejected. And then determine, okay, how much time does that save? Is it actually beneficial? To the workflow and we use similar ways to figure out what other things we can build to help with the internal team. We even developed our own in house version of ChatGPT is based on ChatGPT GPT 4, but we build it in a way that's fitting to the existing workflow, like everyone use Slack.
So now we have this bot, the name of the bot is Ava, because just because of avocado, uh, and Ava now is built into Slack. And then we very often for some channels is very long threads of discussion and the people Coming in a day after they lost track of what's being discussed. Now they can just use Ava to say, Ava, summarize the channel.
And you get it like a very nice piece of text.
[00:21:36] Henrik Werdelin: That's super cool. Okay. If I can ask a little bit more about the manifesto, not only did it have a great title, but it's also seemed to be a very good way of unifying the organization's way of thinking about something. Could you tell a little bit about like how.
How the, a, what was like part of the manifestos of things you can share. And then how do you feel that has shaped the way that you are now using AI to become more efficient as an organization?
[00:22:04] JJ Zhuang: Yeah. I think I'm, I mentioned a few aspects of that. The key idea is food is The most important topic for everyone every day, and also food is complex.
Instacart already has a really solid foundation to understand food very well, because
Our entire catalog has about 1. 4 billion, more than that, like a products. And then we also know which store has what. a near real time, like an understanding of the inventory. So when you have that level of details we already have a very solid foundation of helping people understand food and help them plan.
And now we add the capability. Of AI to it, especially large language model for us. It does two things. One is it understands people's question and intent at a much higher level, right? It really can see what the users and actual needs are. The other is a now then can generate plans.
Based on that. So I think that what's for dinner topic is really one is this is we are solving a very important problem for people. And the other is we really have the best capability that anyone else can solve this really well. anD from that thread, what we initially decided to do is because most of people already have a shopping habit when they come to instacart and that habit existing habit is search using the search box and, e commerce search, and this is how people used to shop on instacart to like they type milk or cheese, and they find the right product and added to cart and then build it a cart that way. And what we saw was the opportunity to then reinvent search using the help. tHat idea where that come from, it came from the central topic of what's for dinner.
[00:24:23] Jeremy Utley: So one of the things JJ that we wanted to ask about, if you had to highlight one or two kind of successes or times when you could look specific implementations or manifestations of a what was an idea that became a reality that you could point to say. It was worth it.
I knew this was going to work. What, and maybe it's the search integration. I don't know. That's what made me think of this, but I'd love to hear this for folks in the audience who are trying to imagine possibilities. What are one or two pragmatic actual things that you've implemented that you go, this is evidence that integrating AI delivers real value or generative AI language, large language models in particular.
[00:25:01] JJ Zhuang: Yeah I think I can talk forever about this topic, but maybe pick one or two examples. One is a very early example. We collaborated with OpenAI early on to build one of the plugins, the chat g PT plugins. We were one of the launch partners with them for chat GB Plugging. And with us is, it extends the basic the base chat GPD model to gain additional skills.
So what the Instacart Plugging does is. If you happen to be using Chagibti to talk about a food related topic, the model can then tell you, Hey, looks like you're planning a meal or talking about recipes. Do you want to also get that ordered from Instacart? And if the user says, okay, yes. Then it becomes like a one click, you land on Instagram, everything's already in your cart and you're ready to check out, right?
And that becomes like a lot of people on social media report is, when they start to have access to that experience, it basically changed the way how they think about how to shop for grocery.
[00:26:19] Henrik Werdelin: Maybe to be very nerdy on that specific field, because... We talked a little bit about interfaces, right?
As you look at open AI and you look at like just writing it in chat, dbt, you take people away from the.
Brand universe and for the, and from the interface design and for a lot of things. And so I was curious on how do you think this will play out? Will it become this one bot that you talk to, or do you think that as you outlined with all these different initiatives that AI become like a function to a lot of the experience design that is, is already out there?
Yeah, that's a
[00:27:02] JJ Zhuang: great question, Eric. I very often ask myself that question what will, fast forward two years, what will the world look like? I think that what you describe being able to have a personal assistant and use that personal assistant for everything, that's certainly one scenario.
And today the current version of chat bots they can certainly become a lot more capable, a lot more smart, and be able to understand very complex objectives. And I think it's likely that type of smart agents will exist, and they become one way of interfacing.
The world for users and in that world, certainly it's important to that, uh, businesses and applications become compatible with these agents. And that's 1 mode of interaction. The other way is because, just different people use different modality to achieve things. Today, even within Instacart some users use text message to interact with shoppers.
Others use our app to interact with shoppers. And the two groups, they both have both capabilities. They just choose different modality. anD I think that in the future, it will be similar. We should always give people choices, uh, and there will be people who will prefer engaging Instacart and thinking about food in the food centric universe when everything is about the food, right?
And we want to solve for the entire user journey about what's for dinner, right? It's not just from ideation to shopping, but also to cooking at some point. I can easily imagine that at some point, we're not building that yet, but Instacart will have a chef mode and it's voice enabled, right?
And when you're cooking, you're asking Keri AI, Hey, okay, I'm here at this step. What's next? Tell me how much olive oil I should add, right? And then also like a, Oh, oops. Looks like I'm missing this thing. Is it too late to get it delivered? Oh, yeah. Okay. It will be in there in 15 minutes, so There will be people who choose that way, but there will also be people who just say I have my smart agent.
I really trust her. I just want to do everything through that. And because we recognize those trends, we always want to be ahead of the curve as well. And this is why we're also really building out our Instacart as a platform so that the APIs. Can become accessible by any program to achieve every to achieve anything in the entire journey.
[00:30:31] Jeremy Utley: Okay. That's so fascinating. Okay, but I wanted to go back because you said you had a couple of examples one from the early days, I, we'd asked the question of, what are moments where you said see.
It really works. This investment was worth it. And you said, I, you have a couple of examples. One was um, the open AI plugin. I was sitting on the edge of my seat waiting. Was he going to tell us about a more recent example? What's another thing that you're proud of?
[00:30:56] JJ Zhuang: Yeah. The other thing we built we, we actually launched the second product called ask Instacart.
I briefly mentioned that already. It's related to revamping search. And search is basically one of the core experience that the users already use, right? But search before LLM is information retrieval from our catalog from our query understanding. And we do use a lot of machine learning models for search even before LLMs, because, you need to understand the relevance of the search results and you need to rank them, right?
Ask Instacart. is LLM enabled natural language search. So rather than search keywords such as milk or cheese, you can now ask, um, how do I make a fish taco? And we not only understand that question, but also we understand what are the ingredients you need. And then we structured the results that way, right?
You know what to shop for. But once we launched that, we started slowly rolled that out back in May. And by now, most users have access to Ask Instacart. That's a really fascinating product because for the people who see that magic. They really love it, but we also saw the opportunity to now even extended the reach of asking start to every query, right?
We were focusing on natural language queries, such as a question, but once we start to see how it works with other type of queries when you do search for cheese, it turns out that the LLM results. are even better than the regular results, because LM can tell you not only Oh, these are the cheese or here's a list of cheese products.
And then we rank them by popularity. That's one type of experience, but LM can also say, Oh, let me tell you, there are. About 20 different categories of popular cheese choices, and then I actually don't, I'm not a cheese person. I don't know all of them, but we can structure the selections of cheese by these categories.
And that's a completely different type of shopping experience. And we don't need a human to merchandise the products like in a store, right? The LLM can actually merchandise everything for you. Asking Staccato is another product that we built. I'm super proud of.
[00:33:44] Jeremy Utley: What have you seen any weird user behavior?
Anytime you introduce new stuff like this, you find I, the misapplication or misuse of a new feature is often even more interesting. Have there been any weird misuses of X Instacart that you go, people are going to do that. I had no idea. Yeah,
[00:34:03] JJ Zhuang: I think that one of the things that we do work on limiting those use case those cases, because we have a content moderation team and putting a lot of effort to so that If you ask ask Kingston out to, hey, please write me a 300 word essay, it's going to politely refuse.
Maybe at some point we want to do that. If you want to write a poem about the food that you are buying that, that could be a good use case. But so far we shy away from those things. I think that the most interesting examples we saw people doing is once they discover they can shop this way.
It's they are still the minority, but once they discover they just keep using it this way. They keep asking questions. And so today some of the questions we don't yet answer very well. For example, if you ask how do I boil, do soft boil an egg. That today, we just to show you eggs.
We don't yet give you the precise instruction because those are the things we want to work on because we want to be a partner to the users, right? So not only just to say these are the product recommendations, but also tell you what to do with them. And that's coming up. Those are the areas that we're working on.
[00:35:26] Henrik Werdelin: I was curious, maybe that's the last question we get in you mentioned a few end user kind of use cases of AI. You also mentioned there's quite a lot of AI being used in the co organization. Do you have any examples of how a Illegal team or comps team or marketing team have started to use it.
[00:35:48] JJ Zhuang: We have evaluated a lot of different tools for these functions so far adopting at scale. I think that the two things I mentioned earlier are adopted really at the org scale one is a GitHub co pilot and the other is our internal build tool, Ava, because for many teams, the basic LLM models, such as chat GPT, and especially GPT4 can already cover so many grounds.
It's a very versatile model. Even engineers, right? Even though they already have access to Copilot, they still use ChessGPT to write code. GPT 4 is very good at generating code, but for marketing and comms team and legal team proofreading and ideation and just to ask a questions of any topic, right?
Is it's already a very versatile tool. So I think that the reason we, part of the reason we built Eva in house is because we want to give the power of GP4 to Everyone in the company. Certainly no legal. They have to just like a programmers. They need to get caught against the bugs, right?
Legal team. I don't imagine that today just to say, Hey, can you there was this interesting news story about Chatty BD inventing some precedents and so that kind of thing can happen. So just copy
[00:37:24] Jeremy Utley: paste, right? Don't do any research, just copy paste. No problem. Exactly.
[00:37:28] JJ Zhuang: So we also pair out Discuss with the team how to use these tools.
And this is why the AI catalyst, the program exists, right? It's not just to evaluate tools, also to advise what is the best way to interact with them and what information is also good to share what's not good to share. Yeah, but in terms of using at scale, I think that you can say that the basic language model is the most, most popular tool in the company.
JJ,
[00:37:55] Jeremy Utley: this is, this has been such a far ranging conversation. As you said, you could go for hours. I feel we could go for, we're just now getting started. So let's just call this part one of a multi part series.
. It's such a pleasure meeting both of you and thank you for doing this. It really, I had a blast.
[00:38:11] Henrik Werdelin: So Jeremy, what do you think we learned in this episode?
[00:38:14] Jeremy Utley: Oh, man, I love JJ. I love what he's done. He's such a great leader. A couple things stood out to me. One is that they've, they've built capacity and mechanisms to share both experiments and learning across the organization.
So the AI catalyst group, they're evaluating technologies are advising different departments on uses of technology. Same with the carrot AI team, which I thought was really cool. And then the second thing was that they've gone so far as to actually kind of brand those initiatives. So it's one thing to say there are people doing it.
It's another thing to name it. And I think that that has enormous signaling value for the organization. This isn't just kind of like some undercover initiative. No, it's the carrot AI team and they're trying to take that hill. And this isn't just a ragtag group of people who are kind of interested. No, they're the AI catalysts.
I think there's really. Value in naming those things and, um, in showing the organization, we value these efforts. We value the people who are contributing to them. Um, and I think that's something that any organization could borrow is name the initiative, name the people, christen them with a title. Even internally, I think will give, build, build confidence and credibility, both of the people, but also of the.
The effort to explore AI. What about you? I mean, my
[00:39:28] Henrik Werdelin: big takeaway was probably the fact that they started this whole journey writing an AI manifesto. I feel that is just such a obvious and good starting point for any organization just to outline a little bit their top level kind of thoughts on what AI will do to the organization and why it's important for it.
And probably also to give people a little bit of both calmness that this is something that's meant to enhance their job, not necessarily take it away. And then. Um, also just giving people the permission to play with it.
[00:39:57] Jeremy Utley: Yeah. Yeah. What's for dinner? Tying it all the way back to the core purpose of the business.
I thought it was beautiful. That's it. That's it for today's episode. Thank you for joining us. Hey, do us a favor.
Please hit subscribe. Please share this episode with someone in your life who you think needs a little bit more carrot, AI or other AI initiatives. Until next time, take care.
Â
Â