AI is the defining technology of our age, offering near endless opportunity and transforming how organizations find talent and get work done. The rollout of “agents” that can perform tasks and make decisions is further pushing the envelope of what is possible. That said, separating hype from reality and understanding how to leverage and operationalize these new capabilities is more important than ever, and many organizations have struggled to achieve ROI despite significant investments. In this research-focused webinar series, you will learn:
The webinar discussed the role of AI agents in HR, focusing on Eightfold’s AI-native platform and Talent Tech Labs’ research. Key points included the evolution of AI from bespoke models to generative AI, the rise of multimodal AI workers, and the importance of responsible AI. Eightfold’s platform supports talent acquisition, management, and resource management, with AI agents like the AI interviewer. The discussion highlighted the need for IT-HR collaboration, the growing AI spend, and the importance of agility in AI strategy. Case studies showed significant improvements in recruitment efficiency and quality using AI.
Host 0:00
Hello, and thank you for joining today’s webinar. We’re excited to have you join us on this great topic about AI agents in HR. If you have any questions during today’s webinar, please put them in the chat widget located at the bottom of your console. And if you experience any technical difficulties, you can also place those in the chat, and our team will help you troubleshoot. And with that, I’ll pass it over to Amy to kick us off.
Amy Tilles 0:25
Thanks, Sue, and good day, everyone. I’m Amy Tilles. I’m the director of partnerships for Eightfold. I am joined today by David Francis, the global head of research at Talent Tech Labs. Thanks for joining us, David. We’re going to be discussing AI agents: hype versus reality. So, really thrilled to have you here, David.
David Francis 0:42
Excited to be here. Thanks so much, Amy.
Amy Tilles 0:44
You got it. So it’s no surprise that AI is happening everywhere right now, and there’s so much noise around it. Our goal today is really to just cut through that noise and give you a grounded, research-backed picture of what’s actually happening and what you should be thinking about for your organizations. We’re looking forward to an active and engaging conversation, and we would love, by the way, for all of you to put your comments or your questions in the chat. We’ll be taking those as we can and rounding those out at the end.
In the course of today’s conversation, we’ll be focusing on the following four points:
We’ll be talking a bit about that toward the end. But before we jump into that, I want to take a moment to highlight what Eightfold AI is all about. Eightfold is the only AI-native platform for all talent, backed by an incredible ecosystem of world-class partners, which you’ll see at the bottom of the slide. We’re the global leader in talent intelligence, built on patented deep learning AI, and purpose-built for every kind of talent, including recruiters, candidates, employees, contractors, and more.
What sets us apart isn’t just our platform; it’s the fact that our ecosystem includes trusted global leaders like Microsoft Azure, Salesforce, Deloitte, Accenture, and others to give customers the support to drive the type of transformation you all want at scale. Our technology is protected by the foundational patents that you see on this slide that no one else can claim, covering everything from bias reduction to candidate success prediction, which is something I know, David, we’re going to be talking about with you in just a moment.
So, just a little bit more information about how we think at Eightfold. This slide is a concept we talk about a lot called the infinite workforce. It’s a situation where we see things moving from human scale to agent scale, and that actually requires support on both sides of that coin. On the left, you’ll see humans: they’re the orchestrators. They focus on strategy, judgment, nuance, creative problem solving, and complex decision making. On the right, you have agents: they’re executing. They’re autonomous, high volume, and data-intensive, working with that infinite consistency that we as humans just can’t compete with. This combines to create an infinite workforce. This isn’t just about AI getting smarter; it’s a fundamental change in who or what executes work, and that has massive implications for every person on Earth, along with everybody on this webinar today.
Just to explain a little bit more about how Eightfold is different from generic AI, I want you to take a look at this visual. The three data layers on the left are where we begin. You’ll notice at the top enterprise data from human capital management, HRIS, applicant tracking, learning management, vendor management, Salesforce, or similar systems, and more. Then there’s the external layer, gathering 1.6 billion career trajectories and skills anonymized from market data and public sources. All of that combined with user interactions within our platform ensures that AI learns and improves from actual decisions that are being made about talent. This information drives true intelligence. When individual and organizational intelligence from those different sources and layers combine, it supports finding the right career for everyone, which happens to be Eightfold’s mission. The complexity of connecting all these data sources and keeping them current is the business that we’ve been in for over 10 years.
So, what exactly do we offer? Our platform covers talent acquisition, talent management, and resource management. The agents that you’ll see at the top of the screen are where AI lives, and that includes our AI interviewer, AI interviewer companion, our agent companion, our candidate agent, and newer capabilities like our Interview 360, which we’ll talk about a bit later, and Talent Forge. On the right-hand side is the build-your-own layer, a common topic these days, for organizations that need custom talent workflows on top of the resources and agents we already provide.
Amy Tilles 5:10
So, how do you know if your AI is responsible? Well, a strong foundation and powerful data only matter if you trust the platform behind it. For a lot of organizations, safety and risk are the things preventing them from moving forward in the AI space, and we at Eightfold understand and agree with that perspective. You need confidence to make sure that AI isn’t going to introduce unnecessary exposure for your people, your brand, or critical data. We’ve been built with responsible AI at our core, which has informed every line of code that we’ve written over the last decade, and our algorithms were trained with specific parameters to interpret information from the very beginning. I think you’ll see that and feel that as we move through the conversation.
On the right, we have the things that we hold dear: the right data, algorithms, evaluation, and product are really the definition of trust for us. If you can’t trust your AI, you can’t use it. David, I look forward to hearing your perspective on leadership—CIO level, HR leadership—and sort of how all of that plays into the conversation.
Last but not least, the proof that we have in the success of our product is really in the definition of the logos that you see on this page. We work with the most innovative brands in the world in 155 countries, 19 industries, and over 30 languages to deliver a premier talent experience. The organizations we’re working with are small and large, but their talent challenges are very real. What we’ve found is that responsible AI and effective AI aren’t really in tension with one another, but rather the same thing. That’s really the context, David, that I want to bring to today’s conversation, and I’m really looking forward to understanding your research and priorities.
Just to introduce you a bit further than I did at the beginning, David at Talent Tech Labs is an independent, non-vendor-funded research firm focused on talent technology. Your research spans talent acquisition, talent management, and the extended contingent workforce, and you have done an enormous amount of research on this topic. So, David, tell us what we need to know.
David Francis 7:18
Brilliant. Appreciate that. Thank you so much, Amy. Appreciate the kind words and the warm introduction. Happy to be here today and excited to talk about this topic. Lots going on, so lots to get into.
This is our obligatory “about us” slide. I think you already gave us a pretty nice intro: research and advisory firm. You mentioned our practice areas, and we’re quite proud of the organizations that we’ve had the privilege of serving. The only other flavor or note I’ll add here is that one of the advantages we have, given our position, is that we have had the privilege and opportunity—in the process of working with our clients through some of these issues—to see behind the scenes in the sausage factory, so to speak, how some of these dynamics have come to play across many different organizations. So I’m going to try to thread that in without mentioning specific organizations, but what we’re talking about here is backed by research and also by the work that we’re doing directly with these organizations as they navigate some of these challenges.
In terms of what we’ll be talking through today on the agenda, we’ll start with a walkthrough of the provider landscape—basically the solution set of firms out there actively pitching their products and services to your function and organizations—and try to give you some insight into how that market is evolving, along with best practices for making decisions given the amount of optionality in the market today. Then we’re going to take a look at how this has gotten adopted across the enterprise, examining some of the different stakeholders—talent being one of them, but also other stakeholders responsible for making decisions. Then we’ll end with some case studies and takeaways for talent practitioners specifically who are looking to get started on this journey or get better results on the journey they may already be undergoing.
So I want to start here by condensing 70 years of AI history into one grossly oversimplified slide. The main point is that AI isn’t necessarily new. In fact, we’ve had AI in the talent space for the past 20 years or so. Historically, the way that AI came into talent solutions was via bespoke AI and deterministic models, which meant you had all these different vendors focused on one specific point solution or part of the talent process—maybe somewhere in recruiting or talent management. They had trained a custom model with a team of researchers on a specific dataset in order to do a specific thing, like reading text or pulling skills out of a resume. It was mature, but very niche. You needed a team of researchers and a custom dataset, which was the reason there were all these different point solutions.
In 2022, with the release of generative AI and large language models, it was essentially the first example of something similar to generalized artificial intelligence, where it was a model that was good across all different domains. It didn’t need to be specific to talent or finance, but was good across all domains and generally operating or performing at the level of a human expert—although how exactly that’s defined has some discussion around it. The challenge with large language models as they were initially deployed was that they were basically just chat interfaces. So they could do things like data retrieval, synthesis, ideation, or content creation, but they didn’t actually execute work.
David Francis 11:26
They didn’t actually interact with corporate systems or even your personal devices. So while they were extremely powerful, there were limitations to what they could actually accomplish. The latest iteration or most advanced evolution of this is what we call multimodal AI workers or agents. The distinction here is that they have significantly deeper, more complex reasoning capabilities, are able to string together more complex tasks, and importantly, they have access to systems where work is actually performed. They have the ability to perform work, update systems, grab data, write emails—whatever the organization decides to give them permission to do, they actually have permission to complete. So this is transforming the workforce at large, and in some cases, talent acquisition and talent management specifically.
Amy Tilles 12:28
So, David, I love this slide. I think it’s an excellent way to set context about the history of all that’s happened. So, I’m happy that this question just came in. The comment is that marketers have done an excellent job of blurring the lines, especially late in the journey here. Are you seeing talent leaders or businesses being able to recognize the difference between, say, a conversational chatbot and an autonomous agent, like what you just described?
David Francis 12:55
Yeah, I wish the answer was yes, but not always. One of the things I’ve observed—and it’s not just talent leaders, quite frankly, it’s basically everybody—is that depending on who you ask what an agent is, if you ask 20 people, you’ll get 20 different answers. What I would say is important is to make sure that you’re aligned around a common set of semantics. So when we go into a conversation with an organization or a group of talent leaders for the first time to talk about agentic capabilities, we make sure we define what we’re talking about first. What’s a workflow? What is just generative AI, and what is an actual agent? Make sure we’re aligned around that to ensure we’re comparing apples to apples, because agentic isn’t the answer to every challenge organizations have, and it’s important not to confuse what’s been marketed as an agent when really it’s just a chatbot.
Amy Tilles 13:53
Fantastic, thanks. So attendee, sounds like you’re not alone.
David Francis 13:59
All right. So here’s a visualization we put together to try to help make sense of some of the noise out there. The space right now is moving quite quickly, so don’t take this as gospel because the players are moving and expanding into new areas, but take this as how we got where we are today and a starting place for how the ecosystem is evolving.
At the bottom, we have foundation models. It’s only a handful of organizations with the infrastructure and capital to build what we think of as AI today, or commonly called frontier large language models that power the entire ecosystem. These companies are investing billions of dollars to train models on essentially the entirety of digitized information across a variety of sources, competing with each other to create the leading model that’s then distributed to other platforms, point solutions, and vendors who use those capabilities inside their applications.
AI infrastructure is a middle layer—more of an enterprise IT layer—where this is like a “build your own agent” type of experience. A lot of the large infrastructure players have built low-code or no-code tools for organizations to be able to put together agents themselves in their own infrastructure using these foundation models, designing, building, and deploying them themselves. This is the domain of Microsoft Copilot Studio, ServiceNow, Moveworks (which ServiceNow recently acquired), Salesforce, and the like.
Then at the top, you have point solutions. These are domain-specific applications sold typically to the talent function that use the capabilities of the bottom two layers. We’ve broken this out into a handful of different areas which we can explore in the next slide.
These are talent-specific domain solutions, and I’ll talk through how these differ and the commercial model, as well as who’s responsible for making these decisions later. For foundation models, up until recently, nobody in the talent department was thinking about which model to use. There’s a little movement here as model providers try to build domain-specific agents in talent, finance, and other enterprise-facing domains, but historically these companies just build the foundational AI used by consumers or application developers. One important thing to note: as a result of this ecosystem, when a vendor says, “Oh, we use AI,” normally what they’re referring to is large language models. Since there are only a handful of these companies, there isn’t as much differentiation in the fact that AI is available as there once was when each vendor was responsible for bringing its own algorithms. When Microsoft or a new vendor pitches you their AI, a lot of times it’s powered by one of these foundation models.
Amy Tilles 18:10
Suffice it to say, there’s a lot of choices, sounds like, David.
David Francis 18:13
That’s exactly right. The way I would describe AI workers is: think of an occupation you might talk to your kids about getting into or going to college for, and there’s a group of people in Silicon Valley trying to build agents that do the majority of the tasks associated with that occupation. In the early days, it was largely software engineering, but now it’s market research, sales, finance, medicine, legal—across the board. If you can think of an occupation, there’s probably a company trying to build agents verticalized around those tasks.
AI recruiters is a category that is technically a subset of AI workers, but since we’re in the talent industry, we broke it out uniquely. These are firms that take on some or all of the tasks associated with talent acquisition and recruiting: intake, pre-screening, assessment, sourcing, outbound activities, qualifying, and matching.
AI and algorithm audit is a category that’s risen largely on the back of legislation. The concept is a self-regulatory mechanism where vendors go to these providers to measure inputs and outputs, ensuring decisions aren’t inadvertently introducing bias or running afoul of laws in relevant jurisdictions.
AI infrastructure we talked about—the build-your-own, choose-your-own-adventure tools to build agents, deploy them, and integrate them with corporate systems.
Candidate AI tools are the bane of many a corporate practitioner’s existence right now. In their most benign form, they assist candidates with job applications or resume writing, but they also do things like helping candidates cheat on assessments or generate fake experience and profiles to get through an application process they otherwise wouldn’t be qualified for.
So that’s the solutions landscape as it exists. Now, it’s not quite so simple; there’s quite a bit of evolution happening across the space, with a few different turf wars being fought over commercial models to capture enterprise budget. At the top, foundation model providers like OpenAI or Anthropic (the provider of Claude) use a cost-per-token commercial model. The more you use their AI, the more they charge you, almost like a toll booth. If you do a lot of prompting or AI inference, you pay more; if less, you pay less. As noted, these are typically bundled into other products, which is why many vendors are moving to a consumption-based model. Instead of paying a fixed price per seat (the historic SaaS model), you pay based on usage—price per interview, price per coaching session, or price for the volume of data analyzed. The model providers themselves are getting into verticalized solutions.
David Francis 22:29
Possibly as a defense against becoming a commodity, they are trying to sell directly to enterprises. In Claude’s case, with what they call plugins, they have an HR plugin—an AI agent that can help enterprises automate tasks. Generally, these tools are bought and purchased by IT.
In the middle, infrastructure players also use consumption-based pricing and give LLM access, working inside existing corporate environments and bought and governed by IT. One challenge is scaling and standardizing this. What a lot of organizations experience is many people building different agents, resulting in an “agent of one.” I might build an agent useful for me, but my colleagues in research don’t benefit because it hasn’t been standardized and distributed. Keeping corporate control becomes an issue here.
There’s also a turf war happening where ERPs like Workday, SAP, Cornerstone, and others are trying to become agentic systems of record—the infrastructure layer for how agents are built and deployed. So there’s a shift from historical HCM/ERP platforms toward becoming AI infrastructure platforms. If you look at Workday’s recent acquisitions, they all follow that exact theme.
Then you have point solutions with mixed pricing—typically a fixed license plus consumption depending on specific AI capabilities used. One advantage of using a point solution over building it yourself is that vendors bring workflows out of the box, designing the interface and specific workflows for you, making speed of adoption faster and more standardized. This set of solutions is generally purchased, bought, or influenced by HR, and certainly governed by HR once deployed.
Overall, three macro trends drive the solution provider market:
To illustrate this evolution: on the left are Claude’s recent HR plugins, and on the right is a set of acquisitions Workday has made—all AI- and specifically agentic AI-related.
David Francis 26:27
The speed and pace at which announcements are happening and new products are released make it challenging just from a getting-started perspective. It feels like as soon as you’ve made a decision, something else happens that you need to adjust to.
This slide synthesizes the state of the market and provides guiding principles on making decisions in an AI world across two flavors: evaluating vendors and addressing the build versus buy conversation.
In evaluating vendors, the question is: where is differentiation among a solution set of companies that look very similar and may use the same underlying AI?
David Francis 29:58
This is a commercial evaluation as much as a technical one, and it’s essential for long-term success. Lastly, consider actual commercialization and cost depending on the use case, as spend is a core part of any ROI calculation.
Amy Tilles 30:19
David, I’d love to jump in quickly with a comment and a question that came to mind regarding the flexibility of providers. We’re encountering that at Eightfold within our own four walls. It’s no longer just build versus buy; it’s build and/or buy with the products we’ve generated. For example, our core talent acquisition and talent management platforms are there, but if an organization is interested in our new AI interviewer product, they don’t necessarily need to invest in the other two. We can integrate and connect with their existing talent acquisition systems. It’s not all-or-nothing for us. On top of that, our build-your-own capability supports unique needs if an organization wants AI interviewer alongside custom workflows without making a wholesale system change. Vendors are going to have to be more dynamic because companies want to ensure they have the right tools.
David Francis 31:31
Yeah, that’s exactly right. We’ve always been advocates for technology systems that are more open than closed, and it’s even more important these days. Being closed can be extremely limiting to long-term success. The puck is moving toward open ecosystems, so it’s good to see Eightfold responding that way.
Amy Tilles 32:04
With that said, how are organizations managing the compliance and security challenges of allowing internal teams to build their own tools? There’s a side to this that risk-averse organizations are concerned about. How do you advise clients? Where does IT step in?
David Francis 32:27
Yeah, that’s a can of worms. A good use case for allowing what we call citizen development of applications is proof of concepts (POCs). If you want to run a POC to see if an idea has legs, you can do so relatively cheaply with proper guardrails and training. The challenge comes when it’s time to scale across the business. You need a process involving IT and legal to graduate a proof of concept into something deployed enterprise-wide. The best use case for citizen development is identifying good POCs to double down on and distribute more widely.
Amy Tilles 33:46
Great, thank you.
David Francis 33:48
On build versus buy: it has gotten significantly easier to build and deploy using AI coding assistants and the rise of “vibe coding,” even by non-technical teams. A guiding principle is to evaluate the complexity of management, security, and scaling versus your pilot. Often, 5% of the effort goes into the proof of concept, while 95% goes into managing and deploying it enterprise-wide. You also need to evaluate whether building software is a core competency for your team. Given that employment is a regulated industry, the risk of running afoul of laws can be quite steep, so exercise caution depending on the use case.
Now I want to shift gears to the mindset of practitioners through the eyes of different stakeholders. The AI revolution isn’t a talent-specific phenomenon; it affects the entire economy. Stakeholders across the organization are wrestling with what AI means for the future of work, each with slightly different views and desires.
David Francis 40:39
For those hoping AI is a passing fad, data from a CIO survey by Andreessen Horowitz shows where AI investment money is sitting. Initially, AI was funded predominantly by innovation budgets for experimentation. Now, it’s sitting inside essential IT and business unit operating budgets.
On actual enterprise spend, investment in both large language models and AI business apps is growing significantly faster than initially estimated. In fact, token utilization is up about 10x year-over-year. Spend is up 2x while utilization is up 10x because unit costs for intelligence are dropping, but overall utilization is surging. This isn’t a passing fad; it’s becoming a central part of how organizations manage their business and IT spend.
Amy Tilles 42:50
David, a quick question on the far right of this chart: the projection for 2026 shows 60% growth from the prior year. Do you have insights into how this will evolve beyond what’s projected? Will the appetite for AI apps continue growing at this rate?
David Francis 43:17
There’s historical tension between business units wanting tailored tools and IT/finance wanting to consolidate into an ERP. During this period of experimentation, there’s been great openness to trying AI apps. However, we’re entering a period of rationalizing pilots down to core platforms. Total spend on AI apps will continue to rise because it’s still early days, but organizations increasingly want to consolidate tools to keep things simple.
Amy Tilles 44:58
Great answer. Thank you so much.
David Francis 45:00
Happy to share. This diagram illustrates how these solutions are put together across the enterprise to help you communicate your goals to IT partners.
Here is a real-world example of an organization using a similar architecture, showing how specific vendor logos fit into each layer of their enterprise ecosystem.
David Francis 48:56
Data from our global talent leader survey addresses who is responsible for AI strategy and governance. Is it HR or IT? The survey results show it’s a shared responsibility. Organizations that partner early between HR and IT achieve far better results than those where IT manages 100% or HR operates in a silo.
Amy Tilles 50:05
That’s exactly what we see at Eightfold. Organizations with innovation coalitions uniting HR leadership and IT achieve far greater success, as both groups align their priorities around built solutions.
David Francis 50:28
Moderna formalized this into a shared function. If organizational inertia takes over, the default owner becomes IT, so talent leaders must stay involved in these conversations to avoid being passive consumers of IT’s decisions.
On adoption levels across four categories: generative AI copilots show high adoption. HR service delivery uses agentic AI to resolve inbound HR requests without human triage. AI recruiters and AI workers are also growing rapidly. In our recent survey update, the percentage of organizations actively using these solutions has increased dramatically.
Regarding strategy: two-thirds to three-quarters of organizations report they don’t have a formal AI strategy yet. There is a lot of activity happening before the master plan is finalized. Rather than trying to figure out what’s possible, many organizations now feel overwhelmed by options and suffer from analysis paralysis amid constant vendor announcements.
Amy Tilles 54:06
We’ve received several questions asking for case studies and success stories. People are facing the dilemma of deciding too early versus waiting too long. Do you have examples of clients or research showing how to do this right?
David Francis 54:47
For large companies, agility is key. Large organizations tend to plan budgets years in advance and move at a snail’s pace, but they need to operate more like startups here. Making a decision and adjusting as you go is better than standing still due to uncertainty.
On specific case studies: here are results from two pilots run with AI recruiters. One was in a high-volume, lower-skilled beverage manufacturing environment; the other was for medium-to-high-skilled seasonal tax providers. In both cases, the AI recruiter outperformed historical benchmarks in candidate contact rates, pull-through rates, and management efficiency. For seasonal tax hiring, a process that previously required 72 specialized staff experienced a 90% reduction in required management personnel, as the AI recruiter natively handled specialized language capabilities. We see similar strong results across other organizations running thousands of interviews monthly.
David Francis 1:03:15
To wrap up with key takeaways:
With that, I’ll hand it back to you, Amy.
Amy Tilles 1:04:51
David, thank you so much. You’re a wealth of knowledge, and we could go on for another hour. We really appreciate you sharing your research and insights today. To our audience, thank you for joining us. Registered attendees will receive a recording of this webinar. For unanswered questions, our team will reach out post-event with specific insights. Thank you again, and have a great rest of your day!
David Francis 1:05:36
There you go.
By submitting this form, I consent to Eightfold processing my personal data in accordance with its Privacy Notice and agree to receive marketing emails from Eightfold about its products and events. I acknowledge that I can unsubscribe or update my preferences at any time.
