Ep 53: The Future of AI-Powered Supply Chains with Prologis CTO Sineesh Keshav

Evan Reiser: 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.

Saam Motamedi: And I'm Saam Motamedi, a general partner at Greylock Partners.

Evan: Today on the show, we’re bringing you a conversation with Sineesh Keshav, Chief Technology Officer at Prologis.

With over 1.3 billion square feet under management and over 8 billion dollars in revenue, Prologis is the world’s largest logistics real estate company.

There were three interesting things that stood out to me in my conversation with Sineesh

1. The progress he’s made to achieve 95% company-wide adoption of GenAI through personal productivity use cases
2. His hypothesis about the future of SaaS as enterprises take increased ownership of their data and AI models, 
3. His vision about how AI enables a shift from static infrastructure to dynamic, self-optimizing supply chains.

Sineesh, thank you so much for taking time to join us today. Maybe to start, do you mind giving our audience a brief overview of your career and maybe your current role as CTO of Prologis?

‍Sineesh Keshav: Sure, great to be here, Evan and Saam.

So, my name is Sineesh Keshav. I've been with Prologis, my current company, for seven years now. But I've had a long career in IT and started off as a C++ programmer, and have been through a few different gigs in a few different industries.

American Express with financial services, Experian with credit profiling and information, Safeway Albertsons with retail, and now here at Prologis - Real Estate. So yeah, it's been an interesting ride across all those companies. 

And so what I do here at Prologis, I’m the CTO, and so all aspects of tech, whether it is our internal facing back office systems, your typical ERP systems, billing systems, HR systems and so on, to the customer facing technologies, which in the warehouse would mean your equivalent of a digital twin, perhaps any levels of automation that we sponsor at our customers' locations. That would be all under my purview. So my team and I work hard to make our customers' lives easier as they go through the warehouse experience.

Evan: Do you mind sharing a little bit for people that aren't familiar, kind of an overview of Prologis?

Sineesh: We are the world's largest logistics real estate company. And by that, what we mean is warehouses. But I'll talk about some other aspects of real estate that we're also getting into. 

And so when you think of logistics warehouses, we operate on the consumption end of the supply chain. So we are closer to the consumer and not production. So less towards manufacturing, more towards consumption. As a result, our facilities are in some of the biggest consumption centers around the world. So virtually every single metropolis worth its name around the globe. We are in the US, the Americas, Europe, and Asia has our facilities. We have about 5,850 buildings. 6,500 customers spread over 20 countries and the square footage that's under our management now adds up to about 1.3 billion square feet. So it is a very large operation and a virtual who's-who of the retail, e-commerce, 3PL world is a customer of ours.

In fact, we're sometimes known as Amazon's landlord because Amazon's our largest customer. But then we have 6,500 customers. So you can think about if Amazon's number one, number 6,500, it's probably a company that no one's heard of, at least most people haven't heard of. So yeah, there's a wide diversity in our customer base and that's who we are.

I will be remiss if I didn't mention that we are also now expanding into other areas. So we have a thriving energy business, data center business, that's all adding on to this scale we talked about. So that's a little bit about us.

Evan: What kind of inspires you, what excites you, what gets you motivated to go to work in the morning? What do you feel like your source of motivation and passion are today?

Sineesh: Well, when you think about the world we live in today, it is a once in a multiple generations sort of transformation happening on the tech sphere. And we didn't see that coming perhaps, even though we were all dabbling in AI, we were too, we didn't see the big chatGPT moment coming for even three years ago. And so that's what keeps it interesting. I think it's a constant learning experience and an ability to make a real impact. 

When you talk about the savings or the benefits that we get or the revenue generation that's possible based on tech today that was impossible 10 years ago, that's what gets me up at night. gets me up in the morning and keeps me up at night too. But that's the challenge.

Saam: So Sineesh, I think everybody by now is using chatGPT. I'm sure their kids are using it on their homework. Yet, there's still like a bit of like, okay, I get it, but like, how's it changing the world? Right. And I think you sit at such an interesting place, you know, given the portfolio that Prologis has and the different aspects of the economy and sort of country and world that you guys touch, like what are some of the ways over the last 18 months that you've begun implementing AI? And I don't know if there are a couple of examples where you've seen an impact of AI on your business that you can talk about?

Sineesh: For sure. So a couple of things. I think good AI begins with good data and good data foundation. And we have been on this journey to harnessing the data that we have access to, even before the latest AI trend came on. And I think that sets us up well to take advantage of what's happening. We have a really rich data foundation. We are a hundred percent cloud-based business, but the data infrastructure of bringing all this data together, modeling it, curating it, and ensuring it has high quality are all those foundational things we've done in the past that is now paying back as we jump on the, on the AI train. That's one point. 

The second is that for the last five years plus, we have been using traditional AI in a big way. So we started rolling out revenue management models in 2018, circa 2018. So when you think about that, the ability for basic pattern recognition and prediction is already there. What has amped up with GenAI is A, and this is, think, very important. It's brought our entire company a chance to participate in this revolution. This is very accessible. The chat experience, as you mentioned, makes it super accessible. 

And, so I'll give you an example. We have a PLD GPT, which is our version of chat GPT with prologist data behind it, that we've opened up to all of our company and we have over 95% adoption. I will be, it'll be hard for me to think of another instance where we've rolled out technology at anywhere I worked at, not just Prologis, where it's had 95 % adoption with very little fuss.

And so that right there tells you about the traction there.

Saam: That's an amazing data point. Maybe there's like two questions. One is like, what have you done to convince your teams and your employee base, some of whom I'm sure are newer to AI that like they can actually use and harness value from this. 

And then the second is like, what are some of the ways your teams are using it?

Sineesh: So, at the very basic level, it is a productivity driver. And that's what we hammered in where it's now resonating with our employees. And in that sense, it's very much a concierge, a personal concierge. And so there is no real pattern, but there is a pattern. And it's about how can this helped me today with whatever I need to do. So one day it might be you just proofreading an email and giving suggestions. The next day it might be uploading a 40 page memo and trying to seek the highlights of it or drafting that memo on the third day. So the same employee uses this in very different ways, day to day. That's what's gotten the adoption going. So that's the personal concierge. 

Then the next level is now how can it help the team? We have encouraged our employees to build their own GPTs. We've given them short little video lessons where anyone in the company can watch that and kind of play and within 10 minutes at least get some working feel for what does it mean to create your own GPT. So that's the next level. And we have a thousand plus GPTs that have been created by our 2,500 employees. So half of our folks are creating GPTs on an average. So that's been the next level. 

The third level is when it comes to enterprise level productivity. And I'm just sticking on the easy part, the productivity part, because there is a life beyond that which we can touch upon. And that third one is where we have rolled out enterprise GPTs. And this covers everything from search functions to some of our common processes being now GPT enabled. 

I'll give you an example. If you are talking to a customer and you're out in the field and you're a leasing manager and the customer says, well, I'm looking for a property that has at least 60 parking spots. I definitely would love roofs, rooftop solar. I'd like it to be in the LA area. I need at least 250,000 square feet and they're just freewheeling conversation. They're throwing out requirements. If in the traditional way of doing things and now that you have all these requirements, by the way, I need 12 doctors and I need it to be closed no more than two miles from the nearest freeway. And you put all this together, it is a data expedition to go find the property that matches all that. How cool is it that today we can basically write that up as a sentence in one prompt and a single shot prompt gets you the answer that there are five properties that match your needs. That's the empowerment we are providing our employees out in the field. 

The level of adoption is choppy, but it needs a journey for everyone to go through themselves before they can trust and completely depend on these tools. But that 95% gets us right up there. Hopefully that gives you a flavor of how we are approaching the adoption piece of this.

Evan: You know, imagine like one of your peers is a CIO and a big company calls you up, right? And says, Hey, like I heard your podcast. I can't figure out how to get my team to use like chatGPT. It's so obvious, but like, I'm not quite sure how to activate that. We'll be like, do you have any, like, you know, imagine they said, Hey, bye bye, bye. The next town hall, I get like three quick wins in place just to show some adoption. You know, what would be your advice? And like, do you any other like pro tips of like, or quick wins that someone can do to kind of get more adoption of some of the, you even some of the more, you know, simple AI tools.

Sineesh: Well, first make it easy and accessible. So do not skimp out on licenses and you know. It's easy to think of these things as, well, let's see how this turns out. You can't afford to be on the sidelines. So there's a certain level of minimum investment that every company needs to do. So my advice to the CIO would be jump in and invest. So make it accessible to everyone.

Then find champions. And it is amazing how you, if you look for them, you will find champions. I mean, we had, we had reports of the early days of ChatGPT as soon as it came out, where people were experimenting with it on their own and reporting back what's going on. Now we had to shut some of that down because they're uploading sensitive data, but what it provided us was these are the people that eventually are gonna be our champions. So identify those champions. You can't do it all alone. It can't all be top down. This is best works when there's a good mix of bottoms up and top down because there are certain things that have to be top down. You don't want everyone out there going and figuring out how to put your enterprise data behind a model, an LLM. So, this one's a case where you have to do both. So you have to invest yourself centrally in building out some of those key...

And there's not much to do, and especially now. I to do, I've been a different story two years ago when it first came out. Today it's largely turnkey. If you're going with one of the foundational models and you just want a chat experience, that's pretty easy. And a short jump from there, provided you have the right data foundation, is to put your data behind it. And if you do those two things, I talked about the magic happening that I don't quite have a pulse on. It's amazing. I think it will happen in every single company. People gravitate to it. 

Now, some of them, I would be remiss if I didn't mention that some of it needs a push because it's not all going to be a bottom sub-adoption. As an example, code generation. It's one of the areas that we could be doing a lot more, but it's a little slow to get adopted because there's always subject matter expertise kicks in and the notion of using AI to code. That's a journey in of itself. 

I was talking to my team that's focused on it the other day. And we were talking about things like if you're writing a 700 line prompt to generate 700 lines of code, is that truly saving anything? Well, but we all agreed as we discussed through it, yes, because what you're writing provides at a minimum clarity of design, because you're having to think about what the code you want to generate. And second, it's reusable now. You never again, when you need to tweak it, you're not tweaking code, you're tweaking the logic behind it. That's an easy combination to manage. And so making people see that is the trick, I feel, to adoption. 

But if a CIO wants to do it within three months and present it at the next town hall, good luck to you, because it's a journey. It doesn't come that quick.

Evan: If Prologis was built today in the age of AI, presumably the way you guys work and how you operate the business and maybe even some of the product and services would be a lot different. Like you can talk about like, what does that future look like? Right? If it's five years, you know, we've fully achieved everything we want to achieve with AI. We've fully transformed the enterprise. Like, how's that going to affect how people work? How the value provide your customers, you know, a day in the life of the average person at Prologis. Like, I’d love to hear your, see into your crystal ball there.

Sineesh: Oh, wow, it'll be a I wish this crystal ball comes to fruition. I think our employees get a lot more creative and less follow a playbook. So we are rolling out a new product almost every week. So that the product development side of it is where our people are focused. Leasing, property management, all the functions that we normally think of what's needed to run our warehouse today are, take care of themselves with human supervision as needed, but for the most part, completely decoupled. 

Our our investment underwriting selection of sites is automated. And, who knows, we may not be talking about anymore the warehouse in its current construct either, because what is happening within the four walls changes enough where we are a company that I can't even begin to imagine what that warehouse of the future is. But I talked about a few things that are more immediate term, but 10 years out. That may be also out of vogue, but we are there. Whether it's in the air, on water, or on land, we are that milestone on the supply chain. 

And you have the ultimate pizza tracker version of the supply chain. Everything is visible, end-to-end, full visibility. You can run millions of simulations of what happens if that Suez Canal thing happens again, and within five minutes, we are back online. The world is back online because one of those million different simulations hit. That would be where I see it going.

Evan: So obviously AI will have some big impact on most businesses, right? Even if you don't think you're a digital business say you're gonna become a lot more digital in the age of AI. If you think about kind of maybe like just the biggest companies in the global 2000, 2000 biggest companies, what is kind of one way you think AI is going to improve or transform these companies that maybe the average CIO or CTO might disagree with, right? What do you think people underestimate?

Sineesh: I think some people have said it and it's a controversial topic, but the future of SaaS is bleak. And that may be something a lot of CIOs may disagree with me on. But I see it. I see it happening where we, for instance, shy away from investing in vertical stack solutions where an existing SaaS product has an add-on, an AI add-on, and they're charging us a pretty penny for it, if you're interested.

And the reason that I feel that that's not a good play for us is it is hard pressed to find an enterprise, and we are no exception, that has all of its context buried in one single system. We're a connected ecosystem. We are 100 % cloud-based, but I can reel off 20. If you talk to some folks within Prologis, they probably will say that we have way more than 20. But the too many systems conundrum. But we have very specialized systems that we take advantage of, and they're spread all over. So the context is also split.

The only way that we can then make sense of this data and have a unified model is to bring the data together. Goes back to that data foundation I talked about. And once you have that, then the layering of an LLM or if you want multiple LLMs and LLM mesh, the orchestration layer, all of that becomes much easier.

So as good as some of these tools may be and so far it's been a mixed bag. I like some, I don't like some. But when I think in totality, our IT strategy here is to always compare a vertical stack solution with what we can do ourselves. And this is with just good prompt engineering. I'm not yet at a point even where this is before we are even distilling or fine tuning. That will come. But this is the crawl walk phase of the crawl walk run. And I'm already saying this. So imagine what it would be if I made that leap into the next evolution of how we use AI. And this is why I think SaaS is in trouble.

Evan: I'd love to hear your take on, you know, what are kind of maybe properties or attributes of some software solutions that you think are more likely to kind of be built, you know, internally and which ones you actually think become more valuable to have a third party built?

Sineesh: I think anything around proprietary data needs to be built internally. Whether it is for internal use or for, especially if it is external facing. Because you'd never want to share that data to begin with. I see all of this becoming a level playing field very quickly where, if everyone has access to the latest, greatest models, which should be our assumption going forward, what is our differentiator? And it comes to hard physical assets like we do, which is a plus, or data if it is in the virtual world. And so nourishing, nurturing, growing that proprietary data footprint is a must.  And protecting it is a must. 

By the way, yet another reason why vertical stack solutions are a little bit of a problem. And so that's what I would focus on. That would be internal. Pretty much everything else renders itself to a buy.

Saam: In addition to CIOs and CTOs, we have a lot of founders listen to this podcast who are thinking about new companies to go build with AI. And so, what advice would you have for them and like what problems would you tell them to focus on in this era, where you would want to partner with you new startup vendors versus just default to building it on your own, you know, it's a house of data fabric.

Sineesh: The end-to-end automation of business processes is still painful. It needs a lot of hand-holding. And I don't see that in spite of agent-tk-i here, and we're all embracing it now. I don't see that changing in a hurry. It's a big problem statement and I think to be very specific that would be an area where if I was starting up something I would look at investing in because it's just not there. The ability to orchestrate and mediate a business process end to end. So that's one. 

The second is that and this is a, this is both AI driven, by the way, no surprise there. The second one is, I go back to that comment I made in Jest a little while ago about the common complaint is in an enterprise, and even for a consumer, is that there's too many systems, too many apps, too many. So where is the universal super app, both in the enterprise and outside. And I know we've talked about it for a while. Now we may have the tools to begin to build one. So you don't go, you don't have to have 20 different systems to understand who your customer is. You just have this one virtual platform that everything hangs off of. That would be nice.

Saam: Yeah, you know, today, if I have a process that's done by human team, like a year from now, if not sooner, that process will be done by a combination of humans and agents. And then there needs to be an orchestration layer on top that understands what humans are capable of, what agents are capable of and how to even manage the process across those stakeholders. So I agree. That's a, that's a real pain point.

Sineesh: And by the way, the benefit we have from AI is that that process can change much more frequently than before. And so it's an auto correcting, auto more efficient, auto more effective layer than just, you know, doing what the human told it to do.

Evan: Yeah, because also AI has some subjective or it can apply some subjectivity, right? The problem with that procedural code is does exactly what you said, whether you like it or not, right? But there is some kind of subjective interpretability. So like the dream for the future of AI software is that there's some judgment being applied to make sure, is he actually being taken in line with the spirit of the intent or the deployment or the installation?

Sineesh: Absolutely. And not making mistakes with that level of independence comes great responsibility and that responsibility part is held up too.

Evan: Alright, I think we got like about 10 minutes left. So the end of the episode, Sineesh, what we like to do is to have a quick segment, which is a bit more of like a lightning round. We're trying to get your like one tweet answers to questions that are like impossibly difficult to answer in one tweet. So just forgive Evan and Saam and directly blame our producer if these seem unfair. 

So Saam, you want to kick it off first?

Saam: Yes. So to start, how do think companies should measure the success of a CTO?

Sineesh: Change. How much change has come in that has stuck.

Evan: What's one piece of advice you wish someone told you when you first became a CTO?

Sineesh: IT is not always about being a service provider.

Saam: So maybe to switch gears to the personal side, what's a book you've read that's had a big impact on you?

Sineesh: Oh yeah, it's nothing to do with tech, but it's a book called When Breath Becomes Air. I don't know if you guys have seen it. It kind of shook me, so I remember it. It kind of shows you the, in all of the talk we do about AI, there's a bigger goal.

Evan: What's an upcoming technology that you're personally most excited about?

Sineesh: Quantum computing.

Evan: Give us one more tweet version of why.

Sineesh: I feel like the growth that we are seeing in AI is limited as in this is a you can see it in the news everyday by our ability to scale GPUs. And GPUs are hard to come by. I think we need an exponential increase in compute to keep up with what's ahead in terms of our next-gen AI applications, and quantum computing can possibly fit that bill if we all put our lot behind it, which is not the case today.

Saam: What do you believe will be true about AI in the future that most people would consider science fiction today?

Sineesh: That people don't need to work anymore. You only engage in your passions, whatever they are. It seems kind of far-fetched, but I think it'll happen.

Evan: Sineesh, thanks so much for joining us today and excited to chat with you again soon.

Saam: Thanks a lot, Sineesh.

Sineesh: This was a great chat, gentlemen. Thank you.

Evan: That was Sineesh Keshav, Chief Technology Officer at Prologis. 

Saam: Thanks for listening to Enterprise AI Innovators. I’m Saam Motamedi, a general partner at Greylock Partners.

Evan: 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

Saam: This show is produced by Josh Meer. See you next time!