Using Usage Data to Scale AI Adoption with Kelly CIO Sean Perry

0:01Sean Perry: Hi there, and welcome to Enterprise AI Innovators, a show where top technology executives share how AI is transforming the enterprise. In each episode, guests uncover the real-world applications of AI, from improving products and optimizing operations to redefining the customer experience. I'm Evan Reiser, the founder and CEO of Abnormal AI. And I'm Salmo Tamidi, a general partner at Greylock Partners. Today I'm talking with Sean Perry, Chief Information Officer at Kelly. Founded in 1946, Kelly helped create the modern staffing industry back when it sent workers to client offices with typewriters.

0:35Sean Perry: Today, it places everyone from substitute teachers to data center engineers, which gives Sean an unusual view of how AI is changing work itself. A few things stuck with me from this conversation. First, two years ago to cut down on seat licenses, Kelly built its own AI assistant, Grace. The side benefit was seeing how people actually use it. That visibility now drives what they build next, like a browser plugin that offers one-click actions such as reformatting a resume. Second, when Kelly switched from seat-based to usage-based licensing, AI usage dropped 30 % from Friday to Monday. Sean's view is that nobody misses that 30%.

1:08Sean Perry: The 70 % that stayed was the real work. And finally, Sean thinks we're building AI the same way we built mainframes. He sees the same thing coming as small models move onto laptops with only the heaviest work going to the cloud. His prediction? All this data center investment will turn out much smaller than we expect. Sean, thank you so much for joining. I was hoping you could give our audience a little bit of background about your role and maybe how you got to where you are today.

1:33Evan Reiser: Yeah. So I am the Chief Information Officer for Kelly Services. We're a very large professional services and staffing company. I think one of the things I like most about our company is that we literally invented the staffing industry at the end of World War II. When women were going to the office. The term Kelly girl was a real thing. And that I really enjoy about it is that we also provided them with their own technology, which in those days consisted of a typewriter. So we were literally issuing typewriters to receptionists and secretaries to take to the office and their clients so they can do work, which at that time was a whole new industry.

2:09Evan Reiser: Our 80th anniversary is coming up actually here in a couple of weeks. So it's a pretty good time for the company. And I love the work that we do. There's really something meaningful about helping put people to work and find their opportunity that speaks to me in a way that I didn't expect when I started in the staffing industry.

2:27Sean Perry: It's cool to have a company with such kind of interesting history. And there's kind of echoes maybe flashbacks to 2026 as all of us were trying to equip kind of the world to use AI and thrive in this AI world. Maybe before we talk about AI stuff, Do you mind just share a little bit about Kelly? I think probably most of the call have heard the name. I definitely probably fully appreciated the full scale of operations and what you guys do. Do you mind sharing a little bit?

2:54Evan Reiser: Sure. So we run across a number of different verticals. So we have a professional industrial business that staffs assembly line workers. We're doing a lot of work actually in the data center industry, helping stand up new data centers and commission them because we have a science and engineering division that focuses on that business entirely. And as you might imagine, because of AI, it's growing quite nicely. And then the other major area that we focus on is actually education. So we have one of our businesses is responsible for placing substitute teachers in classrooms all around the U.S. and making sure that when a teacher calls in sick in a school district, they have somebody in that classroom making sure that those kids are taken care of and business placing at the professional level in universities and colleges.

3:39Evan Reiser: So if you look at administrative and professional staff in universities and colleges, they're one of the top firms that are doing executive search in that industry.

3:46Sean Perry: You had this post on LinkedIn earlier where you talked about kind of, you got some power users that like can't live without AI. Some people are like barely use it. It feels like one of the unique things about AI outside of its capabilities, also like the speed, the speed, maybe the disparity in like diffusion, right both an understanding and application can you talk a little bit about that and like because that is a new i think frontier and challenge for a lot of cios and um leaders today what you got right and i know like uh we have the same thing if we um if we told if we said we're removing chat gbt half the company would quit and half the company be like whatever

4:20Evan Reiser: it's okay yeah and the question is which one you want to keep but so i have an opinion but i won't So what we saw was, or what we're still seeing, right, is that adoption is really varied. But I think the, and I'm a huge fan of analogies, right? It's the same way that like, you know, you had the homebrew computer club, like bringing their computers into the office. Those are the people that are really driving. Like I had a meeting last week, I think it was last week with one of our finance guys. He's just a regular finance guy, like super smart, but like had gotten this passion behind him to do stuff and had built like an entire agentic system using Claude that helped him manage clothes.

5:06Evan Reiser: Very cool. And if you looked at what he had done, like the agent would modify itself, like modify its own instructions. It was really, I mean, it was, wow, you're a finance guy. This is pretty cool. And so that's a great step. But if you if you then expand that and say, well, how many other people like that are in the finance organization? Zero. Like like he's the one that's pushing the frontier. But the other thing is that by him pushing the frontier, other people are seeing what's possible and doing things that, while certainly not that complicated, are definitely are moving them up. Right. And so I think that's the biggest thing that we've been focused on internally is looking at how do we communicate the new uses for AI?

5:51Evan Reiser: Right. How can people make use of those things day to day? So we have a whole like series that we edit and post about sort of different use cases that are happening in different departments with the goal of like getting that news out. because the other, depending upon how your organization is structured, like my old world growing up as kind of an old guy, always in the office, right? And so like a lot of things I learned were things that I learned with somebody just peering over my shoulder and showing me, or, hey, you could do this. Kelly's completely distributed, almost probably 80 % or more work from home or outside the office.

6:27Evan Reiser: So wow, that never happens anymore. And so there's this whole ability to infuse technology and stories and capabilities where, hey, this is how I did this thing. And did you know that you can do it too? It's a good problem to solve, but I think you need to have good data to figure out where to go after it. And I think that's where we've been a little bit lucky is getting some really good on the ground, sort of ground truth data about how we can, where our gaps are, who's using it more, who's using it less.

6:58Sean Perry: Sean, can you talk about it a little bit? Because it's funny, I'm allegedly a cybersecurity expert, but most of the conversations I have the last four months of people, outside of maybe hugging face open AI, it's actually about AI adoption inside the organization. I think one thing that people struggle with is some of that visibility. It's a combination of what AI stuff do I even have? How much is it costing? Who's using what? What are they typing into these things? And then is it worth it? Are there things that you guys have done that's given you better visibility or observability or kind of governance, right?

7:34Sean Perry: Maybe like ROI optimized, that's like stage four. Are there things you've done kind of beginning to just know what stuff you have, what's sanctioning, right, Sam?

7:43Evan Reiser: Yeah, we've got two, I think two things that have been really useful to us in looking at usage and adoption and they're sort of positioned differently. So the first thing in a tool called next thing that actually gives us, they have an AI agent that sits on, sits on the person's laptop or device, and we get to see what sort of tools they're using. We can see how often they use them. And so we can map that back to help us understand, like, from a user perspective, like plot them, you know, not only in the organization, but by job level and their task and their role and figure out like, okay, like we've literally built a matrix of like, okay, what's the, what's the depth of use in these different categories.

8:21Evan Reiser: And you can literally build out like okay like in this organization at this level i see you know ai heavily adopted at the senior level and middle management and then oh we get down to the task level workers and they're not using it at all okay cool like how do we go talk to those people is there is there a use case that makes sense for us to talk to them about right so we've been able to map that out and actually just recently we added to our quarterly engagement survey two additional questions about how people perceive AI and then how they feel they've been supported in their use of AI. And so now we've crossed that data with the actual usage and it's even more enlightening about like, okay, so, but we're also able to figure out like, okay, so this low usage correlation is also people who have a bad perception so we can tackle that.

9:11Evan Reiser: So, so that's been a like super helpful to us over the last almost two years now of trying to figure out how to roll things out. The second one is a little bit more tricky and had to do with we created an internal AI tool called Grace that allows people to use any of the AI engines as a backend. And the side effect is that we get to see what they're doing, right? So it's all about board. It's not like a surprise, but we can see the types of things that they're doing and those use cases then become prompts that we create and feed back into the system.

9:47Sean Perry: Grace, I was reading a little bit about this. This is like your internal AI assistant, right? And you've kind of like given this to like 5 ,000-ish users, right? And it's also kind of integrates some other platforms. Is that right?

9:58Evan Reiser: Yeah. So the first iteration was really just a website. Well, actually the first iteration was a Teams chat bot that you went to, ChatGPT in the background. And then we realized that we needed more capability and we could ship that way. So we created a full website that basically like cloned chat GPT's UI at the time, but still allowed us the telemetry and the usage analysis and things like that. And now the latest tweak is it's a browser plugin. So basically it sits in the browser and it looks at what pages you're on and it knows that there are certain actions that can perform on that page or using that content and volunteers to do them for you.

10:33Evan Reiser: So if you're a staffing professional and you are looking at a candidate profile, it knows that you are more than likely to want to either reformat the resume to a particular style. You might want to create a what we call MPC or most placeable candidate profile, right? A summary profile that you would position for a client. And so it gives you those options right there. And the best thing is it's one click. So where what we were seeing was that people's adoption was slowed down by the fact that like right now using like claw or using chat GPT starts with either copy paste or like drag and drop.

11:07Evan Reiser: Hey, I have these things in my email that I want to give to Claude so that I can ask it questions, right? These particular attributes or Excel spreadsheets or whatever, and a shared drive in SharePoint, I want to pull them in. And so we thought that was actually slowing people's usage down quite a bit. And so by injecting that right there on the page, we've given them like one click actions that have been really well received. And having visibility to what people are doing allows us to really do a much better job of positioning the features that we should be building.

11:36Sean Perry: Yeah, I mean, probably like there's something like maybe Gardner has like some of the like customer journey or hype cycle, but it's like first you got to decide if you want to be in the company, right? Then you got some tools. Then you got to have visibility. Then you got to track costs. Then you got to track usage, right? And then if you do all that stuff, then you can kind of figure out where to invest more, where to invest less. And tell you guys are pretty far ahead in that journey. Is there any kind of like pro tips, right? I think like you guys are almost, I think you're like probably six months ahead of a lot of your peers, right?

12:03Sean Perry: Talk to like a lot of Fortune 100 CIOs. Everyone's on that journey. Right. But like are the things that like have worked in hindsight where you're like, hey, six and six months ago, we did do this. We didn't do that. And that really kind of helped us in the journey. Right. Is there a kind of, you know, key piece, you know, key pieces or advice you'd have for other people that maybe, you know, a quarter or two behind you?

12:21Evan Reiser: I think the I think the numbers have been the biggest thing for us, like really being able to get the visibility about what's going well and what's or where people are adopting and where they're not. And so then you sort of learn like the we've literally been out and looking at people like, hey, you're you're using a lot of tokens. What are you doing? Like and then you get that story and you figure out, OK, that's something I can expand to other people like they're doing something smart. And so, well, the question, you got to be careful how you pose it. It's like, why are you spending so much money?

12:53Evan Reiser: It's like, well, if that's a valid goal, how do I share that with other people? And so I think having that data and those analytics and now, I mean, the good thing is that COG's usage reporting is good enough that we can literally like figure out like I get a weekly list of who my top consumers are not so that I can go audit them, but because I want to figure out, okay, who's got new use cases? Like we had a person pop up two, three weeks ago. We're doing a major system implementation and we're doing payroll and we're doing a lot of reconciliation. And the person was went from like zero to top of the charts.

13:27It's like, what's going on?

13:28Evan Reiser: And it turns out like the team had built a way to use Claude to help us reconcile all of the detailed line items on the payroll between the two payroll runs, between the new system and the old system. That's like super tedious work. And so, cool, noted. That's a cool thing we want to keep in mind and start thinking about. Because you already sort of have this idea that I would use AI in testing, but I would not have particularly chosen it to say, hey, I've got this detailed reconciliation that I need to do. And so I'm going to have Cloud help me with that. But that was a huge win.

14:03Sean Perry: Yeah, Sean, I think I can imagine your answer, but there's definitely someone who's listening to the show. It's like, Sean, what are you talking about? right? Like, why would I go build all this stuff myself, right? Like, why not just use cloud coworker? Isn't that kind of good enough, right? Can you kind of explain like, you know, and I think there's, I think there's a lot of people that are going in your direction, right? But like, why, like, what is it, what is the value of having kind of like this all integrated kind of own solution? Is it more security? Is it visibility? Is it just cost management?

14:33Sean Perry: What caused you to go down that route, right? For you to kind of build this way, first wait for one of the kind of AI labs to build it and like do you like do you think um is that still the right approach right like what is the kind of intrinsic value of kind of owning your own stuff on your own you know with that control and we're hoping someone else figures out for you so in in the case of grace for example

14:53Evan Reiser: um there there really isn't um anything that does what to do and a simple version of that is that at the time we launched grace two years ago the only way to get access to ai was a seat license yeah I can't afford to give them all seat licenses. Like that's a non-starter. Right. And so this way, like we like our real intent was get people access to AI as inexpensively as possible. Right. Cool. So like mission accomplished. Oh, side benefit. Right. Like side quest. I get to actually see sort of how they're using it and get those use cases. Yeah. So I think that in that case, it may have been more serendipity than a great plan.

15:36Evan Reiser: But the outcome was the thing that we needed. And I think more so the cloud coworking thing is interesting because I'm not too sure enterprises are ready yet to let that thing loose. But I'm not sure that I'd be ready to hand my credentials over to cloud coworking and say, hey, what can you do for me today?

15:56Sean Perry: So I think kind of like two years ago, right, or even a year ago, that wouldn't make total sense. Are there maybe like unexpected benefits you found along the way? right as example like there's eight different prompt vehicles and eight different models i got web and mobile and desktop right i can now see all the stuff one centralized way so i have visibility where's kind of the benefit like hey i can control all those like api integrations right so i can kind of i know kind of all my enterprise data is flowing through i got one kind of choke point for that there's some kind of benefit you've discovered along the way that like hindsight makes you feel really good about this path?

16:29Evan Reiser: I think at the moment, right, like the path is still, it's interesting because we had this debate a couple of months ago, right, when Cloud announced that we're going to token-based billing, we're like, hey, if the seed license goes away, then there's nothing that prevents us from giving Cloud access to everyone. And then we realized that we're not really sure that's the best case for us. We think that Grace offers us one thing that I don't get, which is the ability to figure out to program in model use, which is still a problem, right? So

16:58Sean Perry: How do we - Like model selection, should you use Fable versus Opus? Exactly.

17:03Evan Reiser: Right. With that tool -

17:05Sean Perry: Are you saying you don't want to train your entire workforce about the terms and teach them about which models to pick in the browser?

17:11Evan Reiser: No. Definitely. Because it shouldn't feel like that. But there's definite advantages to being able to switch models based on what people are doing. That's part of what Grace allows them to do is make that easy transition from one model to the next and also understand some of the economics behind it so that they know what they're doing. But again, it's almost like installing a billion meter on word, right? God, that just doesn't feel right, you know? And so we need to figure out, you know, maybe the economic models will change eventually, maybe not. But I think in the near term, it's more about how we can sort of steer things a little bit at a macro level where we have a chance to influence spend.

17:50Evan Reiser: From my perspective, we would say that if you're a heavy user, right, and you're really using AI a lot, like my friend on the, Steven on the finance team, I would say you should just go use Claude. Like you're going to use all the bits of it. You're going to maximize it. Here's Claude code, go to town, like have a good time. For the casual users, I'd rather have them use Grace. I'd rather have them use Grace because we've got more guardrails built into it. We have the ability to identify and suggest prompts to them. Like, Hey, it looks like you're, looks like you've asked me to reformat this resume.

18:21Evan Reiser: By the way, here's seven different ways I can do it for you. And a standardized prompt that makes it always look the same. There's things that we can sort of coach them into that they might not know about otherwise that makes that a better avenue for the novice user needs. Now, I wouldn't tell Steven to use Grace, right? Like totally different, but he's kind of like my frontier guy. Yeah.

18:42Sean Perry: There's a lot of, there's some big companies out there. They're spending massive amounts of money to push the frontier, right? And let's say, you know, the best models there are 200 IQ. So before like 2000 IQ and 10 ,000 IQ, right. Maybe you want Steven on a 10 ,000 IQ model. My guess is for like the average worker, you don't need 10 ,000 IQ, right. You'd like to an IQ is great, right. We just got to send some emails and schedules meetings. So do you see like a bifurcation and like, you know, maybe it could be economic driven or may just be unnecessary. You see a bifurcation. We have kind of like one AI kind of interface for like the normal like employee and then for your Nobel prize winners, you're like, you know, real strategic roles for corporate strategy, you're using kind of like the frontier.

19:25Sean Perry: Like, do you see, do you see there'd be kind of like a separation there?

19:28Evan Reiser: So I would focus more on like task level than worker categorization, right? I would say that, um, it may be true that certain, certain roles may use more frequently. I think I'd rather make sure that everybody has all the weapons available to them at any point in time. Let's help you know which one to use based on what you're going after and i think in that case um will will probably help them the most because like one of the things that we've seen it's actually been more of a struggle especially lately right is in order to and how much they're spending we put in place cost caps for people on their cloud accounts right and so i think it was week number one that that went into effect that i got the very first like seven o 'clock call on a friday i kid you not i have this due on Monday and I need more.

20:13Evan Reiser: Okay, great. We'll help you by Monday. Right? I don't think we want that to be the way we do work over the long term. When we made the transition from the seat-based licensing to the usage-based licensing, you can literally see the volume of usage dropped by 30 % from Friday to Monday. And I honestly think that nobody misses that 30%. percent like i think the 70 that was left was the true i need help to get this done and the 30 was probably the i wonder if this can be done in cloud or i wonder if right like okay cool like that's that's fun but maybe not necessary i think the model iqs are definitely definitely a thing but also the smaller models are faster so it's not just money right it's like i can i can spin more I can get things done faster with some of those small models.

21:02Evan Reiser: I also think that we're thinking about this totally wrong. So I'm just going to throw this pitch.

21:06Sean Perry: Yeah, tell me.

21:07Evan Reiser: We're building AI just like we built mainframes, right? Remember back in the day, you could get a mainframe service bureau and they had a mainframe somewhere in the cloud and you rented time on it, right? Like the whole model we're creating is exactly that all over again. And what killed the mainframe? more scale. And so the most exciting thing I've seen over the past couple of months is the announcements that the, I'll call them laptop vendors, right? But you know, you start with NVIDIA's DGX Spark, right? Which is actually should be classified as a personal space heater all the way down to some of the new like laptops that people are shipping with AI chips.

21:47Evan Reiser: And I think that is the ultimate direction that like we're going to get user level, like small AI on your device, on your laptop, it's going to be part of the build, and it's going to be what the equivalent of those smaller models are today, the flash models, things like that. Then you're going to see, I've got a whole different class of stuff that I need to send to the Cloud that is the Fable and Opus class models.

22:13Sean Perry: Yeah.

22:14Evan Reiser: I think the sooner we get to that endpoint, the better it'll be for all of those data centers that we're building out.

22:21Sean Perry: I was curious, do you think there was like a structural issue here, right? Because like the, you know, some of these big labs are very, they need massive amounts of money, compute, power, right? And so as they push the frontier, right, there's just some amount of money they have to charge these models, right, in order to like, you know, fund this. So you think there's like an emerge, like a new category of enterprise AI solution, right, that maybe is a combination of, you know, cheaper open source models, maybe on on machine models, right? With kind of set that has like more like centralized data security and visibility, right?

22:55Sean Perry: Where it's like all in like your cloud versus, you know, the, the labs cloud. Like, do you think that is that a, is that a market that's going to emerge? And like, I guess then the follow-up question would be like, what structurally prevents, you know, open air anthropic or some other lab from doing that. Right. Or do you think it has to be, but for some reason, it has to be a third party that doesn't train their own models. Right. To really make sure you're has, I don't know. Like, yeah. Do you, do you think that's something that's going to happen?

23:18Evan Reiser: I think there's a position here and maybe isn't there an opening here because you look at the deep seek models, you look at all of these like enterprise class models that you can go download from Hugging Face and you can run on your own infrastructure. And if you do that, then you literally like pennies on the dollar. Right. So I'm not spending enough yet for that to be a reasonable response for me to say, hey, I want to stand up my own infrastructure. I want to run either my own hardware, I want to rent it from Amazon, or I want to do something like that so that I can deliver that fundamental baseline service at a lower cost than I would get it directly from OpenAI.

23:59Evan Reiser: But I do think when does the open source model become sufficient? And if anything, like we are closer and closer to that every week. When you look at like Fable and Saul are so like beyond, like earlier you said, like 10 ,000 IQ, right? Like, yeah.

24:18Sean Perry: I need that to send emails, you know? Right.

24:21Evan Reiser: Right. And so now if the model is I can provide that service cheaper, that's what got everybody off of their mainframe in the old days, right? Was, well, I can provide that same compute capacity cheaper if I put it in my own quote data center, which was usually just a glorified office or closet space, right? I can do that myself and I'll do it for less money. Okay, cool. The problem there is how do you give somebody a fully rolled solution, right? So that their users can transition. And that was part of what we really focused on with Grace when we first launched it was to make that, we were like, look, make the UI look as much like the chat GPT.

25:00Evan Reiser: Cause that's what our users were using primarily at the time as much like that as possible. So the transition was really easy. Now, if you did that same thing and you said, by the way, I'll sell it for 10 cents a token or whatever. Cool. And I'll host your choice of AI model and I'll make sure that it gets upgraded to the latest version every time it comes out. There's probably a cool business there, but I don't, you're probably not like my size company if you're doing that, right? If we went and talked to the Walmarts, the Amazons, folks like that, I could see that being financially a smart move for them, right?

25:34Sean Perry: Well, I think also, I mean, it goes back to something you said in the beginning, like Like we're still in the kindergarten years of using AI, right? Again, like the people, I mean, if we look at our token costs by employee, it's a thousand to one, right? A 10 ,000 to one from the biggest to like the least. And so I think we're also kind of these, we're in uncharted territory where I think like the use cases will just increase. We'll use it more, but also the cost per intelligence will come down, right? It has come down. You compare Astra 6 a day to Chattanooga 5.4, the flagship model from three months ago.

26:04Sean Perry: It's more than 10X difference of cost per intelligence. So like it's, I guess the only thing we know is that the future is uncertain and we'll be

26:12Evan Reiser: Yeah, but the fun part is like, this is the best part of the curve, right? Like this is literally like we're back in the days when cloud was just starting. Although I think the innovation is probably even greater than back in the early days of AWS's S3 and EC2, right? Like back in those days, you still had the, I'll call them the unbeliever crowd that was like, oh, no, that cloud thing, who never put their data in the cloud, right? Like, who trusts somebody else, right? Now you're kind of like, of course you will, right?

26:41Sean Perry: And we're like, who would trust us? We'd love better than AWS.

26:45Evan Reiser: Exactly. But so you don't have that sort of pushback. Yeah, there's the environmental discussions and those types of focus areas, but I don't think you're seeing anyone say, this is nothing. It's a flash in the pan. It's going to go away. I haven't seen those pundits show up. It could be, I'm just not reading.

27:04Sean Perry: Yeah, I think they were there up until I think ChatGPT. There was this ChatGPT moment where everyone in their brain, they're like, oh, let me test out this chat bot. And they said, wait a sec, this is not a chat bot. They're like, oh my God, they've got that magical experience. And I think it went from 80 % non-believers to 80%. Like maybe no one knows what it's going to do, but like, okay, this is not normal. There's something special here. So at the end of the episode, we'd like to do a bit of like a lightning round, which try to get your kind of like quick, kind of like the one tweet version of some questions that probably are unfair to ask in the one, two response, but hopefully you'll humor me for a second.

27:40Evan Reiser: Give it a try.

27:41Sean Perry: Sean, how do you think companies should measure the success of a CIO in today's AI era?

27:46Evan Reiser: How are they helping the business move faster?

27:49Sean Perry: You know, maybe for some of your peers who are listening in that are, you know, maybe entering a new role as a CIO or CTO, if there's kind of like, if you had your like hit list of like, these are the must do or kind of like, easy win projects to put points on the board for AI transformation, what would be maybe like the first three areas to focus on?

28:05Evan Reiser: I think the first one would be get the data, figure out what people are doing and what they're using and how they're spending or not spending. Like where's your investment going in AI? And then the other one I think you can do in parallel is find the best use cases, right? So there's people out there that are already doing stuff. They might not be obvious to you, but start hunting for them, partner with them, figure out how to enable them so that they can keep doing the thing that they're doing. And then the last thing would be, and maybe this part is sequential, is when you find those, figure out how to replicate them.

28:38Sean Perry: Is there a book that you've read? It could be fiction or nonfiction or business or other, but like any book you've read that's had like a substantial influence on how you lead, right? Either culturally or anything about technology.

28:50Evan Reiser: So I'm going to give my current answer, which is The Reverse Centaur's Guide to AI, which is Corey Doctorow. If you haven't seen it, He writes a lot of fiction and writes a lot of facts. Really good book. It's sort of like, let's just say he calls AI at the beginning of the book, the word guessing machine. So you kind of know where it's coming from. But actually, I would say it's a very good argument. It's a very engaging book.

29:14Sean Perry: What do you believe will be true about the future of AI and the impact that AI will have on the world that most of your peers would consider science fiction?

29:23Evan Reiser: So my theory is we're going through a phase where all of this invest centers and horsepower is going to turn out much smaller than we expect.

29:33Sean Perry: Sean, really appreciate you having the show. Let's find a chance to meet up again soon.

29:36Evan Reiser: Sounds great. Thank you for having me.

29:38Sean Perry: That was Sean Perry, Chief Information Officer at Kelly. Thanks for listening to Enterprise AI Innovators. I'm Salmo Tamadi, a general partner at Greylock Partners. And I'm Evan Reiser, the founder and CEO of Abnormal AI. Please be sure to subscribe so you never miss an episode. Learn more about Enterprise AI transformation at enterprisesoftware.blog. This show is produced by Abnormal Studios. See you next time.