transcribe

Purpose-Built Enterprise AI Agents: What Actually Works

Everyday AI · 32m · transcribed Jun 2026
More from Everyday AI Business
𝕏 Share ▶ YouTube 📥 PDF 🤖 .md

Transcript

0:01 This is the Everyday AI show. [music] The everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical [music] advice to boost your career, business, and everyday life. If we were to believe the hype, 2025 was supposed to be the year of the AI agents. And sure, it was. I mean now you have fairly powerful and robust both AI agents and agentic models that even non-technical people can go and deploy.

0:36 But when it comes to the enterprise, there's not necessarily one general use agent that can go out and accomplish meaningful work out of the box. It's a little messier than that. And I think one of the reasons, and it's something I've been talking about for multiple years on this show, is when people think of AI agents, I think they think of that one general agent that can go out and do anything. Hey agent, go do my work for today. Or, hey, go do this entire project. And that's not necessarily the way that agents, at least today in early 2026, are built. I think the most uh, you know, purposeful ones are ones that are built around purpose, right? on a narrow use case with very uh defined and measurable set of goals. So that's exactly what we're going to be talking and tackling on today's edition of Everyday AI on purpose-built enterprise AI agents and what actually works and learning from someone that's been building them for a while at a high level. All right, I'm excited for today's conversation. I hope you are too. Welcome and what's going on?

1:44 Welcome to Everyday AI. My name is Jordan Wilson. If you're new here, this is your daily live stream podcast and free daily newsletter helping everyday people learn and leverage generative AI to grow their companies and their careers. It starts here with the unedited unscripted liveream podcast. But if you want the real cheat code, you got to make sure to go to our website at your everyday.com. We're going to be highlighting the keys uh the key insights from today's conversation as well as all of the other AI news that you need to stay ahead.

2:14 Right, it's hard. We do all the heavy lifting for you. So, make sure you go do that. All right, enough chitchat from me. I'm excited for today's conversation because we can't talk enough about AI agents and how is it going to change in 2026. So, uh I'm excited live stream audience, if you could please help me welcome to the show, uh Pashanti Padmad Banan, uh the VP of engineering at LinkedIn. Pashanti, thank you so much for joining the Everyday AI show.

2:39 >> Thank you, Jordan. This is like the best way to start my day. So, super excited to be here. uh and I would say that we are in an age where all of us are learning and building and scaling at the same time. So I learn every day through podcasts, through newsletters, through blog posts and whatnot. So happy to sort of like do a little bit of like giving back from my end and coming and you know having this casual chat with you. So excited. Let's go.

3:06 >> Oh yeah. So let's let's get into it. But before we we dive too deeply into this this concept of, you know, narrow versus general agents, can you first maybe explain for our audience a little bit about what you do at LinkedIn, right? Everyone knows LinkedIn, but they might be thinking, wait, agents at LinkedIn. So, describe a little bit your background and then then maybe we'll talk about a use case for agents that you guys uh recently uh unveiled.

3:30 >> Awesome. So, uh yeah, so I work at LinkedIn. I've been with LinkedIn for about seven plus years now. And uh so I lead the engineering team for the enterprise portfolio of products uh under the LinkedIn talent solutions business. What this really means is like we have a variety of products that are customuilt for talent uh hiring and career development and learning purposes. And for example, the LinkedIn recruiter product is like a very very well-known uh recruiting product that's been in the industry for a very long time and uh trusted and used by thousands and thousands of customers. So now when you think about it, well we had a great product that's working so well for our recruiting enterprise customers.

4:13 So why did we get on this uh journey of building an agentic product? For two reasons, right? One, well, we all saw the AI revolution unfold in front of us and agents becoming like a front and center actor in that play. And when you really think about agents and use cases, what you really want to think about is like, hey, is there a set of workflows and uh processes that today customers do to achieve a goal, right? If you really think about recruiters, what is their primary job? Their primary job is to make sure that they uh are able to find the best talent for any given role that they are set out to actually fill. And what happens behind the scenes for a workflow like that? You have to first make sure that you're coming up with like this really well structured way of defining what this role needs. What kind of talent would actually fill the fit the role and then from there you're going and scouting like thousands and thousands of réumés like from LinkedIn from ATSS from multiple places and trying to find like this golden match.

5:18 Here is a role. This is what this role needs. you've had a lot of back and forth conversations with the hiring manager to come up with this perfect definition of a role and then you have like these millions of candidates probably right on the on the platform off the platform and you need to your job is to find this perfect match and make your hiring manager really happy with that. So that's the that's the job. But if you really think about the job, everything from defining the role, sourcing, finding the short list of candidates, outreaching to those candidates, getting them interested, adding them to the pipeline, sending that back to the hiring manager. There's like so many parts of this job that can be done by an agent in a really, really efficient way. Because what are agents really good at? They're super good at data mining. They're super good at reasoning and understanding and matchmaking and whatnot. If we did that, what is what is the end goal there? You will free up your recruiters to do actually the parts of the job that needs that human touch that needs that human uh ability which might be like making the candidates feel welcome, establishing that rapper with the candidates, getting them excited about this role, pipelining them for this opportunity and whatnot. So that's more of the I would say if you ask recruiters they like to do more of that they like the strategic aspects of finding that talent that human aspect of building that relationship. So that's what motivated us to really do this which is like we have a fantastic product already that you know a lot of our customers love and use. Now how can we now bring the power of this technology the LLMs the agentic tech and take away all this manual you know sort of like repetitive parts of this job give that to the agents to do do that really well and then leverage your human intelligence the recruiter's human strength to do the more strategic and uh you know things that needs more more of your emotional intelligence to do that job. So that's the that's that's why we built hiring assistant.

7:24 >> So I I I do want to dive in a little bit more deeply on the hiring assistant and hear about exactly how that went. But before I want to hit pause on that and go back to something you said there because I think it's really important uh you know and it's it's funny right I've been talking to a lot of people about agents and when I talk to a a CEO or someone in leadership it's very different than when I talk to someone maybe with an engineering background and I think it's important uh because it can frame how we think about agents right when I talk to sometimes people in leadership it might be more about controlling headcount or scaling revenue but when I talked with you right there I don't know if everyone caught this so I think it's worth repeating. You literally just talked about essentially reverse engineering a current workflow and defining what an AI agent actually has to do. What does that process look like? And maybe right because I know a lot of times uh you know uh agentic implementation to be successful you have to have your engineers and your leadership meeting in the middle. So how do you do that? How can you actually, you know, uh, meet in the middle and define both how an agent should and can work versus what maybe you want it to do versus someone in leadership.

8:35 >> Yeah, I think you you sort of like uh caught like the most important aspect of this topic, right? In many ways when we talk about agents even internally within LinkedIn, we look at that as a human plus. It's not the replacing the human. It's not replacing a headcount. It's like, hey, how can a human who's doing a job do the job many many times better if you gave them an agent as a tool, as an assistant, as a partner, a thought partner, a work partner. If you put a human and an agent together, whether it's efficiency of the outcomes, the quality of the outcomes, all of this, how can you 10x that? You know, we talk about 10x engineers who are using coding agents. How can you create like 10x recruiters if you give them an amazing recruiting agent for them to work with?

9:26 Right? So that's the crux of the thinking and for you to do that really well. If you really think about the process of how do you define the product requirements for something like that? How do you define the user experience for something like that? How do you define where intelligence is important, where experience is important? How do you blend these things together? I'll tell you one thing Jordan, you cannot do that by just sitting in a boardroom and writing specs. You just cannot, right? So, one of the things we learned early in the game of building LinkedIn hiring assistant is we started working with our customers from day one of doing this, right? So, we set out to to with a goal to build this, but we were very clear. We have a amazing technologist in our company. We have amazing product managers and designers in our company [clears throat] but this cannot be done by just us. So from the get-go we picked like a a meaningful set of customers that we decided we going to work from the beginning from the get-go iterated on the product like crazy with them. So the version of the hiring assistant that you're seeing today in the market is not where we started like everything from uh we started with a version of the product that was much more asynchronous in nature. You'll come in, you'll define what you want to hire for, you'll walk away and then we expected the agent to go and do the job and come back with results. That didn't work. Like what we realized along the way is you need a much more of a conversational interface and experience where the hiring assistant and the recruiter can tag team on that job can work together can bounce off ideas bounce off feedback and say oh this is what I want to look for in this for this role and the hiring assistant will give you some options you can correct it you can fine-tune it right because if you really think about it that's how hiring works you never just start with like a spec and say Oh, let me go and find the perfect person. It evolves. You might look at 10 résumés and you might be like, "Oh, I like this thing that I'm seeing in this particular resume. Why didn't I not think about it to add it to my role description? I might pick pieces of it and I might evolve my what I'm looking for. I might see something that I don't like." So, there's a lot of iterations, mental iterations, mental model iterations a recruiter goes through. Now, you need to shift that to the agent. The agent needs to learn that about you as a recruiter.

11:50 What you care about? What are your preferences? What are the negotiables for this role? What are the non-negotiables for this role? >> Right? So this agent, the beauty of the LinkedIn hiring assistant is that it gets better over time because it's learning you as a recruiter. It's learning your preferences. It's learning what you know what you grock, what you don't grock, what what gives you like the aha moment when you look at a resume or a portfolio, right? So this is this is why uh I feel like in any agentic product it's kind of like this relationship building between the customer and the agent that happens over a period of time.

12:25 >> The agent learns you, you learn the agent because the best way to work with an agent is to think about it as a thought partner. So that was the process where we realized we knew what we wanted to do but the how was not clear and the how evolved as we talked to our customers as we rapidly iterated the product experience. We traded on the models we traded on the experience and that's how over a period of time we actually got a winning product and the results show that it's a winning product. Uh but I I'll be the first one to tell you like we didn't get this right on day one.

13:00 >> Yeah. And and and speaking of that, um you know, because I know a lot of people listening are probably in the same position that you were before you launched it, right? You know, you're scoping at it and you're saying, "Okay, maybe maybe we have something that's broken that we need to fix. Maybe we have something that's working that we just want to make better." Uh you know, or maybe that there's a certain thing that we have as humans have been doing for maybe decades that's better to hand over to very smart and capable machines.

13:28 So, you know, both I'd love to hear how you uh scoped this internally at LinkedIn, but also maybe how our audience should be thinking about this as well because when they're looking at the future of work, right? Because that's what when you're talking about agents that are more and more capable, the models are getting better, the scaffolding is improving, how do you go about scoping it and finding that purpose-built use case that you can ultimately measure and and look back at and see if it's working or not?

13:57 >> Yeah. So I think uh you want to sort of like discern what are what are like the parts of the agent right you have like the model which is so good at compute it's so good at data mining it's so good at like look for example if you have to you know in uh in a in a very short span review and parse and review and analyze like thousands and thousands of résumés and profiles like the agent is going to be far better at that than you are >> right and it's so good pattern matching and looking for like repeatable patterns and it's looking for like the you know we all our dream and it's it's true in a lot of senses that the hiring assistant will be able to find you portfolios and résumés that you will never be able to find yourself just because your capacity to parse and do pattern matching is much more limited right so we always think about in the recruiting world like what are those hidden gems of talent that we're missing out like how do we find those hidden gems right like and machines are really really good at doing that right? So we want to make sure that when you look at a use case or a problem space like just separate out the pieces that are good for LLMs to tackle, right?

15:08 That are good for agents and orchestrators to tackle versus elements of the product experience that you actually want to get right in the way the experience feels. Why is conversational better than an asynchronous way of doing it? Right? Where do you want to make sure you're using the right copy, the right language? One of the things we learned early is uh you know customers have to build trust on agents. The trust doesn't get built overnight. It takes time to build. And when we talked to customers about like what will help you build trust on this agent what they wanted was they wanted to understand how the agent thinks and reasons. They didn't want it to be a black box because hiring decisions are important decisions. You just don't want to completely offload that and just trust the outcomes. You want to know what's happening behind the scenes. It's like we asking the kids to show their math work, right?

16:01 So we evolve the experience so that we actually show the process. We the agent will show you what it's doing. It will tell you what it's looking at, how many résumés it's looking at, what it's finding in the resumeumés. And when we find a match, we show you the evidence. We show you, hey, we think this candidate is actually one of the top fits for this role because this is what we found in the resume. This is what we found in their uh screening back and forth. This is what we found in probably in the future, their GitHub work, their you know patterns that they've published, right? So showing that evidence was very important in the experience for the customers to build trust around it. M >> so it you know we all want agents and LLM apps to be magical that magic just doesn't happen overnight in my mind that magic is going to be a combination of the power of the compute and the models and the complexity it can handle and like just the infinite data mining it can do and the reasoning it can do all of that but equally important is the experience you build on top of it which will be which you can think about it as like that app layer right that UX layer like why was chat GPD so successful just because it made that experience feel so natural right you felt like you were talking with someone you felt like this thing understands you it understands the nuances of what you want right similarly for any agentic experience it's like as as important is the power of the model and the compute and the you know the billions of parameters it can handle and the latency and all of that equally important is like this experience that you build on top of it.

17:43 >> Yeah. And and you know you you hit on something obviously extremely important there. You know one of the key words of you know agentic implementation is is trust right. So you kind of gave the analogy of, you know, a kid showing their work on their math, right? Or, you know, if you're using a front-end large language model, you can always, you know, look at a summarized chain of thought and at least kind of see and understand a little bit what the model is is doing or trying to do aically, right? But on a platform like LinkedIn, right, with more than a billion users and someone's obviously, you know, their career could be one of the biggest things in their life and it is for a lot of people. Um so how do you find that delicate balance right between autonomous uh you know agents going and do doing very important work uh with that piece of the human element right about this could be a make or break for the person hiring this could be a make or break for the person getting hired or maybe the person getting skipped over.

18:43 How did you balance that? [music] Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft, and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for chat GPT training for thousands or just need help building your front-end AI strategy, you can partner with us too, just like some of the biggest companies in the world do. Go to your everyday.com/partner to get in contact with our team. Or you can just click on the partner section of our website. will help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Genai, >> which is exactly why we don't call our agents autonomous agents, right?

19:58 [clears throat] It is an a it is an assistant with the human in the loop. The assistant is doing all the heavy lifting for you. It's doing the parsing. It's doing the analysis. It's doing the matchmaking. But at the end of the day, you as a recruiter, as a human, are the ones looking at it and saying, "Okay, do I agree? Do I as a recruiter agree with this talents uh with this agent's uh assumptions?" And the way the agent is uh trying to build that trust with the customer is by showing evidence, right?

20:27 It every step of the way it is exactly telling you what it's doing. It's also telling you why it it did the match that it did through sheer evidences. And this is also why I think domain spec domain specific and purpose-built agents are important because that is what we experienced. If you just take offtheshelf uh you know state-of-the-art models, it's not going to work for the specific use case like recruiting. So what we do is we use a combination of LinkedIn's own platform insights and the data that we have and what we know about this candidate based on their own resume and their own uh actions on the website right and what additional evidences we're able to gather about this person's expertise and their experience because that's a data like what people put about their resumeumés and their work experience on LinkedIn is you know they they take that seriously right because that is their professional uh you projection.

21:25 So by using a combination of the unique unparalleled insights we have about professionals on our platform, their network, their activities, their experience, their expertise and then using the power of the models that's that combination that works. So you know you call that like very domain specific fine-tuning and you you call that purpose-built agents. By purpose-built, what we really mean is we take something that is a general purpose uh model. You blend it with your own platform's unique data and insights and you fine-tune that model for that use case, which is a very specific use case here like the the sourcing use case and the talent matching against a role as a use case and you're you're iterating that model to get this right. And that process was very important for us to you know to iterate over a period of time. But every step of the way we have been very clear that no none of the decisions are going to be done by just the agent. Every step of the way the human is in the loop. It is a human in the loop experience.

22:30 Even if the agent comes back and says this is my top candidates, the recruiter has every right and uh you know opportunities along the way to say no, I don't think so and this is why. So the the recruiter can give feedback. That's how the agent is going to get better. It's a it's a birectional relationship. >> Yeah. And I think that piece is important to talk about, right? Like uh you know, you kind of referenced this this concept of of co-working and not just handing something off uh to an agent, but you know, I I I want to dive in a little bit more on uh this this concept of, you know, not just trust uh but even beyond that purpose, right? So, we talk about purpose-built agents, but I keep thinking like, and this is even something I'm going through myself, right? Uh when I'm working with these agents that are more and more powerful, I'm sure that there's many recruiters out there who maybe viewed the work that they do in two different ways. Maybe they said, "This is so, you know, this can be monotonous. This can be hard.

23:30 This can be difficult. I I want part of this to be automated." And I'm sure there's other people out there that enjoy that and they feel a lot of purpose in that, you know, that laborious manual steps. So, how do we grasp, how do we tackle that in the future, right, for those people that find a lot of purpose and being great at their job and being able to find the needle in the haystack with now all of a sudden if an agent or an AI assistant can do that in, you know, 1% or 5% of the time, how do you find that balance?

24:01 Yeah, I mean it it's it's such an important thing that you're raising and we actually deliberately talked about it. I mean I I remember in one of the conversations in one of the execs asking what if the recruiters enjoy reviewing hundreds of réumés like as tedious as it sounds maybe that >> there got to be at least one person right >> I know exactly right uh and I think this is where how you build the agent and how you launch it matters right so in in LinkedIn hiring assistants case we didn't just build the hiring assistant as a standalone product we brought it as a capability on top of the recruit recruit a product. So what we are doing by that is like we're not suddenly pulling the rug under the customer and saying you know what forget everything that you did all these years. Forget the fact that you used our recruiter product and you really professed how you're using it. Here is this new thing. Start start from ground zero. That's not what we did. We said your workflows are still your workflows. Recruiter is a product that you're using. LinkedIn recruiter is a product that you're using. Amazing.

25:02 Now we're introducing this agent on top of that product and we are slowly easing you into change of habits, change of workflows, right? At any point of time, you still have the ability to go and pull as many rums as you want and read it. Even in the hiring assistant flow, as we are giving you like the top candidates, we do see the recruiters still opening every one of those profiles and taking a look at it, right?

25:30 It's reducing the number of profiles you're viewing. Like we have a 62% reduction in the number of profiles that the customers are viewing because that's efficiency because the agent has done the hard job of telling you look at these profiles. You don't have to look at hundreds of profiles and the customers are seeing that benefit. It's not like they're not reviewing profiles. They're reviewing fewer profiles, right? So that 62% efficiency win is amazing, right? And why is that amazing? Because the result is that you're getting 70% more inmail acceptance rates. What that really means is you looked at lesser number of profiles, but when you reached out to those people, you saw a 70% boost in the number of people responding to your email, which means you're now reaching out to the right people.

26:20 >> And the agent has helped you with sorting that list down to that the top list that matters. So, your workflow changed for the better. It's not like you're stopping to look at profiles. You're still looking at them. You're looking at fewer of them and you're getting better acceptance rates. That's where the efficiency comes in, right? So, what we did right here, which again, it's a process that we landed on. We didn't completely change your day. We gave you an agent to use as a as a plus on a product that you're already pretty familiar with. So at any point of time you wanted to do the things uh the way that you're accustomed to you can still do that but over a period of time we're showing you where the efficiency is. We are showing you you can spend 4 hours less amount of time on a role and use that time for things that are much more strategic and human in nature while without compromising on the quality of the candidates. In fact you're going to get better candidates. You're going to find them in shorter amount of time.

27:23 Even the outreach is personalized. You know, the hiring assistant helps you build personalized messages for your outreach. So, every step of the way, our our philosophy is we're going to give you the magic. At the end of the day, you still have a choice whether you want to use it or how much you want to lean in. You we're never taking anything away from you, [clears throat] >> if that makes sense. >> Yeah. No, it it it does. And I think this is a good uh a good transition to to wrap here. So, we we've covered a lot in today's show, but you know, maybe if we zoom outside of your role at LinkedIn and step aside of the role of recruiters, but for enterprise decision makers who are making those important decisions right now at the beginning of the year, you know, and they want to get more utility uh out of maybe narrow or purpose-built agents, what's the one most important decision that they should be making right now or the most important next step to make Agentic AI work in 2026.

28:24 >> Yeah, I think um I've come to the realization after working on this product and this journey that enterprise is a messy business, right? That's the truth of it, right? Enterprise use cases are never like this single lane use cases. Enterprise customers need to use multiple systems in order to get their jobs done. Even recruiters use a combination of they use the LinkedIn recruiter tool, they use an ATS, they use a hiring uh you know an HR system, they might use a CRM. So every one of these customers in an enterprise world are using multiple tools which means your context and source of truth is actually fragmented across multiple systems. So when you set out to build an enterprise agent, you just have to walk in with a clear understanding that you're going to need uh you know two or three very important aspects. One is having a having an evolving a purpose-built domain specific model is going to be very very important. So you need to continue that road map. The second part is getting this context engineering done right. Which means you want to over a period of time build deep and broad context through the right memory sophistication through the right agent orchestration architecture so that your systems can continue to work together in an orchestrated fashion, right? So that you you're not duct taping an agent on a complex system and expecting the magic to work. It doesn't work like that. It needs very thoughtful context engineering to bring these systems together. And most importantly, the experience part is so important because it's the experience. The application logic and application layer that you're building on top of these LLMs is super important to get right because guess what? Your experience is how you're going to build trust with your customers. Customers don't care about your models, compute power, and whatnot. What they care about is what they see and what they interact with.

30:21 So, can your experience instill trust? >> Can your experience instill flexibility? Can your experience instill confidence in your customers that this change in paradigm is not going to turn their world upside down? They can still get their work done. they can still get their work done efficiently, high quality, without feeling like that they're completely giving up the agency and what what they might consider their superpower and their professional assets and strength. If you can hit that balance, and that's not easy to hit, it takes iterations over time, then I think you have a winning enterprise agent product.

30:57 >> Great, great piece of advice from someone that's been building it uh for a while now. So uh Brashanti, thank you so much for taking time out of your day to join the Everyday AI show. We really appreciate it. >> Thank you Jordan. This has been a fun way to start my day and hoping the audience uh got something useful out of it. Thank you. >> Right. Well uh if there was a lot that uh Pashanti just shared with us. So if you miss anything or if you're like I need to dive in a little bit more on that, we're going to be doing that in today's newsletter. Uh giving you the insights and summary in case you missed it. So, if you haven't already, please go to your everydayai.com. Sign up for the free daily newsletter. Thanks for tuning in. Hope to see you back tomorrow and everyday for more everyday AI.

31:36 Thanks, y'all. >> And that's a wrap for [music] today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. [music] It helps keep us going. For a little more AI magic, visit your everyday.com and sign up to our daily newsletter so you don't get left behind. >> [music] >> Go break some barriers and we'll see you next

Summary

The Everyday AI show discusses the evolution and implementation of purpose-built AI agents, particularly in the context of recruiting at LinkedIn. The conversation emphasizes the importance of integrating AI as an assistant rather than a replacement for human recruiters, focusing on enhancing efficiency and trust through collaboration between humans and AI.

- AI agents are not general-purpose; they are most effective when designed for specific tasks with clear goals.
- LinkedIn's hiring assistant is an example of a purpose-built agent that enhances the recruiting process by automating data mining and candidate matching.
- The development of AI agents requires close collaboration with users to iterate and refine their functionalities based on real-world workflows.
- Trust in AI is built through transparency, where agents explain their reasoning and provide evidence for their recommendations.
- Successful AI implementation in enterprises requires understanding complex workflows and integrating multiple systems effectively.
- The user experience is crucial for building trust and ensuring that users feel comfortable adopting AI tools in their processes.
- AI should complement human strengths, allowing recruiters to focus on strategic and interpersonal aspects of their roles.
© transcribe · For agents Built with care and craft by Gokul Rajaram