#293 Greg Shewmaker: How Enterprises Can Implement and Scale with Agentic AI

0:00The mission is to strengthen and secure the full potential of humans with AI. So it's very human-centric. We don't believe that AI, at least in the near term, is going to automate everything as some claim. We think that it's probably much more around sort of maximizing human efficiency. And I think the companies that are going to win are going to be those companies that sort of take that gained efficiency and open up new opportunities. AI gives us immediate productivity gains. I mean, it's just a linear line that goes straight up. And every time there's new models or new tools that come out, we see efficiency gains almost overnight.

0:30But that's not happening in the enterprise because the enterprise is, you know, by its very nature, super complex. And what I'm seeing is the companies that say, hey, let's have an agentic strategy to automate as much as possible. And most of the time it's driven by the IT department or the innovation team, which that's not necessarily wrong with that. But it's probably the wrong approach because AI is truly having an impact across the entire organization. So you need to think about it from the C-suite, from the CEO on down. Build the future of multi-agent software with Agency. That's A-G-N-T-C-Y.

1:04Now an open source Linux foundation project, Agency is building the Internet of Agents, a collaborative layer where AI agents can discover, connect, and work across any framework. All the pieces engineers need to deploy multi-agent systems now belong to everyone who builds on agency, including robust identity and access management that ensures every agent is authenticated and trusted before interacting. Agency also provides open standardized tools for agent discovery, seamless protocols for agent-to-agent communication, and modular components for scalable workflows. Collaborate with developers from Cisco, Dell Technologies, Google Cloud, Oracle, Red Hat, and more than 75 other supporting companies to build next-generation AI infrastructure together.

2:07Agency is dropping code, specs, and services, no strings attached. Visit agency.org to contribute. That's A-G-N-T-C-Y dot O-R-G. Hey, Craig. I'm Greg Shoemaker. I'm the CEO of Our Potential. Our Potential was a company that was recently spun out of the ADECO Group. We're here in San Francisco, California. It's where our headquarters are. Prior to Our Potential, I was the head of global operations for the ADECO Group for several years. what that really meant was we were building talent supply chains for some of the biggest companies in the world. A Deco puts about 2 million people to work every single day across every industry, across every country.

2:52Before that I was doing primarily supply chain and merchandising work for large retailers and e-commerce companies. And many, many moons ago I was in the military from Montana, born and raised in Montana. And since then I've lived all over the world. Yeah. Our potential, it's helping companies implement AI into their workflows. Is that right? Can you talk about the mission? And how did it happen that it got spun out of ADECO? Yeah, the mission is to strengthen and secure the full potential of humans with AI. So it's very human-centric. We don't believe that AI, at least in the near term, is going to automate everything as some claim.

3:33We think that it's probably much more around sort of maximizing human efficiency. And I think the companies that are going to win are going to be those companies that sort of take that gained efficiency and open up new opportunities. So we're trying to figure out with companies, hey, what's the right combination of, let's call them digital workers in quotation marks and human workers to accomplish whatever it is you want to accomplish. And it's, you know, there's not one simple answer. In fact, there's not even one complicated answer. There are many. And the reason why we did this is because, as I said, the ADECO group puts a lot of people to work every single day across every imaginable sort of job there could be on the planet.

4:16And several years ago, we saw, you know, this obviously shift towards AI coming. And we saw that that was going to impact our business on two fronts. One, it was going to change most likely how we do work. Coming out of the recruiting and staffing industry, we knew that it was going to impact us internally. But more importantly, it was going to change our primary business model of putting 2 million people to work every single day. Obviously, the workforce was going to change based on everything we were seeing and hearing. So we went out and we talked to hundreds of customers and executives around the world and kind of gauged sort of their sense of what was happening.

4:51and kind of two things. One was we realized quickly they didn't have a clue what was coming and what was about to happen. And the second thing is that we needed to think about the future and call it next generation workforces, whatever it might be. But there was going to be some combination of humans and AI that was going to make up the workforce sometime in the near future. And just the last point I'll make on that is we knew that we couldn't drive this amount of change in a$25 billion company and run the business at the same time. And so we said, hey, we need a dual strategy. We're going to run the business as is, and we're going to go change the business.

5:27And to do that, we can't do that with the same people and the same methodologies and the same sort of risk tolerance. So we spun out a new company, and that's our potential. That's interesting because that's part of the challenge that enterprises are facing, how to integrate, figure out the role that AI should be playing in their operations, how to integrate it while operating business as usual. And you talk about AI agents as labor, but at the same time, there has been a real failure of agentic workflows to succeed in the enterprise to date. but you talk about that as a management failure, not a software failure.

6:25So can you talk about that? Yeah, I think it's just a failure of sort of approaches. So it's not necessarily individual failure. We're all sort of testing and learning and trying to figure this out on our own, all companies. But if you look at AI today at an individual level or, hey, our potential as a startup, AI gives us immediate productivity gains. I mean, it's just a linear line that goes straight up. And every time there's new models or new tools that come out, we see efficiency gains almost overnight. But that's not happening in the enterprise because the enterprise is, by its very nature, super complex.

7:01And I think, at least in my observation, what I'm seeing is the companies that say, hey, let's have an agentic strategy to automate as much as possible. And most the time it's driven by the IT department or the innovation team, which that's not necessarily wrong with that. But it's probably the wrong approach because AI is truly having an impact across the entire organization. So you need to think about it from the C-suite, from the CEO on down. Hey, how is this going to change our business model? How is this going to change our, you know, capital expense policies, our procedures, our, you know, everything.

7:34And so you need to think about it. How does it get incorporated into our existing business strategy, not how do we have a separate AI strategy. And so, as I said before, I think the companies that are thinking about how do I use AI to increase the efficiency of my people and then take that efficiency and open up new opportunities, I think those are the ones who are going to get it right. Everyone else who's thinking about it as a technology strategy are going to continue to get it wrong or at least spend a lot of money to get it right. Yeah. And the reason they're going to spend a lot of money and maybe not get it right is what?

8:11And just on the productivity thing across the enterprise, you know, a lot of those productivity gains are incremental on the personal level. You know, I use an agentic product to answer emails and sort emails and that sort of thing. And it certainly saves me time. Maybe it saves me a couple hours a day. But in an enterprise, those couple hours a day don't flow to the bottom line. I mean, they aren't aggregated into this massive productivity game. It means that the employee maybe takes a longer lunch or just works at a more relaxed pace. So yeah, those two questions. How do you capture that productivity gain across the enterprise?

9:09And why are they failing so often currently? Again, lots of different reasons. I think one of the main reasons, though, is you're trying to take something that's new, AI agents, and you're trying to shove it into something old, which are existing systems and existing processes, looking at, you know, sort of existing data sets, which oftentimes, you know, are fragmented and are incoherent. And so I think the agents are struggling with that. It's as if you brought in an intern into the company and you said, hey, you're going to go run finance. Now you've got to run paywall on Friday. Good luck. You know, and chances are people aren't going to get paid on Friday because the intern, you know, as smart as they may be, is just not going to be able to navigate, you know, this environment and make it all happen.

9:53And I think the agents are struggling with the same thing. You're putting it into one system and everyone has an agent today. So I buy an agent for my CRM system and that agent does certain things okay. But then my CRM system needs to engage with my HRS system and my finance systems. And there may be agents over there and those agents don't work well together. And so they hit a wall just as humans hit a wall if they're not able to sort of navigate an organization efficiently. And so I think it's just that. And it stems from, again, not having sort of this top-down strategy to say, hey, here's what we're actually trying to accomplish.

10:26Now let's introduce these agents into our workforce, just as we would new workers. And that's why I like to call agents digital workers, not because I'm trying to personify them, or I think they have human level intelligence today. But I think if you start treating them as if, hey, we should qualify the agents that we're bringing in to the organization, we should train them, we should test them, we should hold them accountable, we should be able to fire them, we should manage them just like we do humans, then all of a sudden it's a different game versus I'm trying to install software. I'm installing software with the high hopes of having this human level intelligence and it's just not realistic.

10:59Yeah. In Arm potentially, you were saying that, you know, ADECO is running a business. Let's run a parallel track to figure this stuff out. Is that something that you advise companies to do to set up a team or a, you know, a parallel track that tests these agentic systems or AI systems before integrating them into the main enterprise? I don't think it's a bad idea. Again, every company is slightly different, but it allows you to move quicker. In this world of AI, as you know, things are moving so rapidly, and it's just not realistic for any big company. Even if you're a tech company, You can't move that fast.

11:48I mean, we're here in the middle of San Francisco and Silicon Valley, and it's a completely different world from where ADECO is headquartered in Europe. I mean, there just couldn't be two different universes. And so I just can't imagine doing what we're doing every day here, you know, still within that environment. And again, risk tolerance and just speed and resources focused on, you know, the core business. So it's not a bad idea to at least consider that, you know, for large companies especially, to be able to move quickly. And again, when things don't work, hey, we can shut them down, you know, the next day.

12:20And it's very low cost. It's very low risk. And then we get to try it again and again and again. And so it's allowed us to iterate very quickly and build some interesting things for the enterprise, but doing it sort of at startup speed. Yeah. And then, as you said, things are moving very quickly. This is something I've wondered. I mean, for myself personally, looking at all these tools, it's just overwhelming. How do you decide what tool to spend money on? For an enterprise, the tech keeps changing. How do you advise people to deal with that? I mean, how do you commit to something? and then if something new comes along, how do you swap it out?

13:12Or is enterprise always going to be 10 years behind the market? I don't think they can afford to be 10 years behind the market because I think some of these tools now, as you mentioned, are so incredible and sort of bring a level of efficiency that I haven't seen before how fast it sort of materializes, I don't think the enterprises can afford to wait. And just as an example, so what we're doing is we have these things called units of potential. And essentially it's what we believe is going to be sort of the future configurations of workforces. It's, hey, what am I trying to accomplish as a company?

13:52What are the real world constraints? what are the human capabilities that I need to get this work done and then what are the digital capabilities I need to get that work done. Whether that's right or wrong, don't know. But what we're saying is, hey, for companies, let's test those in the smallest scale possible. And by the way, it can't be fully digitized. So there's a lot of handholding from us and hey, we're going to work with your teams to try to implement this. And over time, we hand it over to you. And over time, we start to automate and digitize more and more of that. But that allows us the flexibility to say, okay, if we had a hunch, this is what you're trying to accomplish.

14:29We're either accomplishing it as planned or, hey, it didn't go quite as planned. We had to make some changes. Now, the next time we're going to do it slightly different and slightly different. I think that allows you to sort of go fast, do things that are related to the business, adjust as new technologies come, adjust as you learn before you make these big commitments to say, well, we're all in on this particular agentic platform, or we're all in on this particular technology, and by the time you get it implemented, it's irrelevant or it doesn't work anymore. And so I think it's more test and learn and work fast within the enterprise, but doing so with scale in mind.

15:03Yeah. And so you work alongside the enterprise. Is that relationship then ongoing? Because as the technology changes, as new tools come along, either the enterprise is going to have somebody dedicated to figuring it out on an ongoing basis or they're going to hire someone like our potential to do it for them. Yeah, that's right. So, I mean, ultimately, our business model is we're creating something what we called a year ago is the total talent marketplace. So how do we aggregate the workforce supply on one side, whether you're human workers or AI agents or robots or whatever comes next. And as I described before, hey, let's go try them out and test them.

15:49Ultimately, I don't want to build another consulting firm to have all these professional service people running around the world implementing these things. Ultimately, the company should be able to do this and it should be as automated as possible. But for what we call design partners, sort of our first enterprises that we're working with now, we're right there with them and we're doing it together. We're learning, they're learning, and it's kind of a, we're on this journey together approach. So we have a very small team doing that. I envision that team remaining small no matter how big the company gets.

16:16But it's really important for those people who aren't necessarily technologists. These are people that know how the enterprise works. They know how to navigate to enterprise. They're learning alongside the customer at the same time, and then we're feeding that back into our platform. Yeah, and that's a good point that you don't want to build a consultancy. but how do you scale? So you're trying to develop like this CEO, co-pilot or chief potential officer. You're trying to develop something that can adapt to any enterprise. What does the enterprise have to do if they're not working hand in hand with you to adopt something like that?

17:08Is there a menu of what kind of data you're going to have to feed it? Are these models that you fine-tune or is it a RAG system that you load up with company documents? I'm just curious how that works. Yeah, so like any marketplace, which I relate to all my background before building in the retail world. And now we're talking about a talent marketplace. You know, you can never build it all at once. So when we first spun out of Adeko, our focus was we have to go build the technology platform. And we built sort of this enterprise platform that had, you know, an experience layer, a trust layer, what we call cognitive services, which essentially is, you know, proverbial brain.

17:52And we have data querying services, which what we were able to do spun out of ADECO is we took about 60 years of labor data. We took massive amounts of, so let's think about like a billion resumes, think about every LinkedIn profile that exists today, think about almost a billion job descriptions from 10 million companies, real-time sort of labor regulations and laws from the major countries around the world, skills ontology, salary data, et cetera. And all that data we were sitting on forever at the ADECO group felt powerful, but it was completely useless. It was fragmented, dirty data that we sat on forever.

18:29So the first thing we had to do with our potential is build this architecture that allowed us to sort of structure that data and make it usable. So first six months, all we did was build the platform. Now this next phase is, hey, how do we take this chief potential officer, contextualize all that data, and provide it back to the CEOs to say, hey, we already know a lot about you and your industry and what's happening out there in the world of human and AI. let us at least provide that as a starting point. So at least that gives them a plot on the map to say, okay, here's where I'm standing today.

19:00And so what we do is we provide them that plot, if you will, to say here's how many people we think work at your company. Here are the roles that they work in. Here are the skill sets of the people in those roles. Here's what we know about what you've said publicly or what we know about those types of jobs and AI. We think your level of automation is this today, 18 % by different job function. hey, we think there's a theoretical ceiling of automation, you know, of whatever, 42 % in this particular role. Now, we're not suggesting that you go to that and get rid of those people. We're saying, hey, if you can get to this level of automation in these particular roles, that gives you a lot of free capacity to then redeploy that, you know, somewhere else in the organization, more revenue-facing roles or driving efficiency, et cetera.

19:47So that's a long way of answering your question is the first data set that we have is just our own data set that we were able to structure and prioritize. Now, if the company says, well, if I provide you my proprietary data, how much more accurate does sort of this assessment get? And the answer is quite a bit more accurate. And then so a new company comes along, once you've got this product deployed or the chief potential officer, What is, is there, um, and if you don't want to have to work with them, uh, one-on-one, how do they know what data to supply the, the system in order to refine the, the, uh, the profile that it's built of the company?

20:40So it'll make a recommendation. So the chief potential officer say, hey, look, here's what we know about you. Here's some recommended workforce configurations based on what you're telling me. Hey, if I had access to HR data, I can give you a much better answer. And because we were able to build this sort of enterprise platform in part, so Salesforce is one of the early investors in the company. And one of the reasons why we wanted to work with them is because they're enterprise-grade security and using tools like MuleSoft. So now all of a sudden we have all the API capabilities for any major system that an enterprise would have.

21:15And so within a matter of days, we can plug into that HR system if they agreed to that, get that data, and turn it around within a very short amount of time to say, okay, we gave you one answer. Now the answer is this based on that data. And who's doing this at the enterprise? Is this a CTO's team or the IT department? and who takes on the product and works with it to figure out what data it needs access to. For now, we're working with a combination, a cross-functional team of both business people and the IT department. However, for our design partners, and right now we're going live in January, we're only working with design partners.

22:03The criteria is that if we don't work directly with the CEO, then we're not working with the company at this time because the CEO is helping us sort of create this experience specifically for them. And we've been so lucky coming out of the ADECA group and then having Salesforce as the other investor that we've had sort of unfettered access to the C-suite over the last few months. And we've been able to talk to dozens and dozens of, you know, the biggest CEOs in the world. And so we're saying, look, let us give you this chief potential officer. It's not a tool because CEOs don't use tools. Think about it as a relationship.

22:36like you or I may have with ChatGPT or another tool. CEO can't use ChatGPT to make decisions for a public company. They need something that's built for the enterprise. It's more secure. It's more private. But they're appreciating to have this thing relating to them, not as the CEO of company X, but as this individual. So the most important person involved right now in the process is the CEO. Now, when they say, hey, let's go and put a real sort of use case into the world, this unit of potential, then that's when we work with a cross-functional team of theirs. But it requires them to start the whole process because we have to build the trust with them.

23:12And this idea of working with the CEO,

23:20what's the interface? Is it like a text chatbot or is it a voice chatbot or is it an avatar that they talk to? I'm just curious. how that's going to present to the CEO. Yeah. So right now it's purely web-based and it starts with what we say, here's a workforce sort of assessment or report about your company. And then there's a chat interface that allows them to go into any element of that report and say, okay, tell me more about this. Hey, you came up with this number. What does this mean? Or, hey, this number doesn't seem accurate, you know, whatever. And so it starts to interact. But what we've done is we've created these composites.

24:01So it's not just a chat GPT sort of mirroring back what it thinks you want to hear. And so your design partners, have you sat in the room while CEOs interact with this? Yeah, 100 % every week. And then the idea is once this is in the CEO's office, he would be talking to it throughout the day. Maybe. I mean, we're not hoping to create an app where we're all about engagement. We want the CEO to talk to this one in the moments that matter the most, right? And the CPO is meant to be sort of proactive and reactive. So it's, you know, sort of scanning the world out there and saying, hey, here's some things that maybe you should be aware of, Mr.

24:48or Ms. CEO. You know, maybe it's important, maybe it's not. Or it's just waiting there for the CEO to come and say, look, I have a board meeting coming up. You know, hey, tell me what I should be thinking about or what questions should I be preparing for that I think I'm going to get hit by my board. given everything that's going on. It's still too early. I don't want to say we know exactly how they're going to use it, but it's not meant to be something like social media where we hope they engage with it every day. Yeah, and the advice that it's giving, how do you expect that to be used? I mean, you talked about a cross-functional team for implementation.

25:25is that something that if a CEO or an enterprise adopts your solution, then the chief potential officer would be advising you should ideally form a team with somebody from this group, somebody from that group, and take a look at these tools, or is that left to the human CTO to figure out? Yeah. So as I said, at scale, assuming this all works how we envision it, is it as this talent marketplace, this platform. So the idea is that it starts with a CEO, And of course, the CEO will never only be the only customer, but we think this is who has the biggest problem now. As we understand the demand to need through these conversations, then it says, hey, would you like to generate this unit of potential?

26:35CEO says, yep, great, let's generate it. So again, here's what you told me you wanted to accomplish. Here's what I know the constraints are. Here's what I think the combination of humans and AI should be based on all the things that I know about those two things. You can change that. the CEO can say, you know what, this is a good idea, but I want to go validate this with some more people on my team. So what we have for now, and it's only a Slack integration now, but at scale it'll be other tools. You can say, well, go talk to the CFO, go talk to my CTO, go talk to these other two people because I know they're experts in this space.