Customerland
Customerland is a podcast about …. Customers. How to get more of them. How to keep them. What makes them tick. We talk to the experts, the technologies and occasionally, actual people – you know, customers – to find out what they’re all about.So if you’re a CX pro, a loyalty marketer, a brand owner, an agency planner … if you’re a CRM & personalization geek, if you’re a customer service / CSAT / NPS nerd – you finally have a home.
Customerland
Trust, Agents, And The Future Of Work
The excitement around agentic AI is loud, but the real story starts where hype meets accountability. We sit down with Eric Karofsky, CEO of Vector HX, to separate narrow task agents from agentic systems that plan, reason, and negotiate across steps—and to ask the uncomfortable question: who do you trust when your agent starts talking to someone else’s agent?
Eric walks us through the gaps leaders often miss. It’s easy to mandate AI from the top; it’s harder to redesign the workflows that make measurable impact. We explore why trust breaks down when autonomy spreads across tools, teams, and vendors, how hallucinations become career risk inside enterprises, and what guardrails look like in regulated sectors. The conversation stays grounded with two practical wins. First, a pharma literature review pipeline that shrinks from six months and $250,000 to roughly two weeks by structuring extractions, adding human checks where accuracy matters most, and instrumenting the process end to end. Second, a document discovery platform that stitches together siloed repositories with smart metadata, natural language search, and relationship mapping that surfaces parent, child, sibling, and multilingual versions of critical procedures.
If you’re wondering where to start, Eric’s advice is simple and hard: map the jobs to be done, pick a high-friction workflow with measurable outcomes, and redesign it around human needs. Save broad agentic autonomy for bounded domains with clear policies, identity, and audit trails. The best KPI for the next 12 months might be whether your core processes actually change; if you’re doing the same work the same way next year, you probably missed the point.
Ready to rethink your approach to AI beyond better emails? Listen now, subscribe for new episodes, and share this with a teammate who owns process change. Your take: where would agentic AI actually earn trust in your organization?
There's massive opportunity for process reorganization. I mean, McKinsey is putting out, it seems like a report a day on um AI. And, you know, the the reason being is they see, you know, massive technology spend there, but massive process reorganization.
SPEAKER_00:Today on Adventures in Customer Land, Eric Kurofsky, who is CEO of human experience firm Vector HX. This is our second official get together, but Eric and I have had a handful of conversations in between this time and last time, largely because it turns out there's just a lot, a lot, a lot of overlap between what customer land is all about, where we think that world is going, and what Eric and his company already do. So with that as a rambling introduction, Eric, thanks for joining me.
SPEAKER_01:Yeah, thanks, Mike. Pleasure to be on and appreciate you having me on again.
SPEAKER_00:Sure. So just to set some context because it's been a while, can you tell our listeners a little bit about Vector HX and what you're all about?
SPEAKER_01:Yeah. So um, you know, the what if all of your apps were built on deep user insights that were intuitive and engaging in every touch at every touch point. Um and, you know, think about the difference between the Department of Motor Vehicles and Apple. Um, you know, and that's that's not even a fair comparison. What that experience has looked like. Um, and you know, that that discrepancy that is really where we focus. Um, we really focus deep in understanding what are the customer needs and then develop develop experiences that meet those needs. Um and you know, very often it's you know, either an online experience, um, an app. Um, more and more we're talking about um agents and a gentic AI um and and things along those lines.
SPEAKER_00:So the work you do is not so much theoretical, it's strategic. But I take it that a lot of what you're doing is actually production work. It's it's the actual making of the things. Am I right?
SPEAKER_01:Yeah, absolutely. We we are making the sausage.
SPEAKER_00:You're making the sausage. So you you you touched on something a moment ago. I'm not sure if that was inadvertent or intentional, but um agenc AI and the fact that it is, um I don't know, you know, in in business, every every moment has its um its term du jour. And I think agentic AI is this moment's term. It's it's just what everybody's talking about. And um, I don't want to get too far ahead of us of this conversation because there I think there's a lot of kind of bridge building to be done between where we are right now, what you and your team are doing in the here and now, and what agentic AI really is. And there's a there's there's a pretty big gap, knowledge gap, understanding gap, and maybe even like production gap in there.
SPEAKER_01:Yeah, absolutely. I mean, that it's funny, the um the call for agentic AI is coming from the CEOs of you know large companies and the small ones that understand anything about AI um are calling for it. And you know, they they say they want agentic AI. I I think most of them don't really know what the term means. How would they distinguish between agentic AI versus agents um versus just a workflow? Um, you know, so uh there's a lot of education needed out there, there's a lot of discussion and um these mandates coming on from high, and there's very little actual building of this stuff. Um, and whatever is being built, you know, really is has yet to be adopted.
SPEAKER_00:Can you walk us through? Because you you mentioned a couple of different variations on the big thematic idea here. Agents, workflows, agenda KI. You're the person who works in those realms, this big realm. Would you mind just for a moment defining those for listeners?
SPEAKER_01:Yeah, so you can kind of think about an agent um as you know, a nice helpful assistant. Um, they they follow instructions, they complete defined tasks and hand the results back. Um, they're they're kind of like the doers, um, and they reduce manual effort. So things like um generating blog articles, marketing copy, um, automating expense reporting and compliance, things along those lines. Um on the other hand, there's the agentic side. Um, and you know, again, that's where all of the buzz is right now. Um, and that's much more of a collaborator, um, you know, maybe with initiative. Um, and what they're doing is they're executing tasks, um, but there's some thinking behind it. They can reason behind, you know, what needs to be done in what order and why. Um, so you know, that there's a lot of discussion about different types of identic apps that that can be built, um, a lot of different use cases out there. But again, um, I see very few actually doing it um with with success.
SPEAKER_00:Yeah, at the moment, I um the conversations that I'm having with people who bring up the term, it's uh it's spoken of in terms of look at the age we're about to enter. Look how good life is going to be when I can ask my my agent to do this thing and it'll go execute and bring these things back, which is which is pretty cool. Seems like just a bit of an extension from where we are right now, but there is that kind of promised interactivity with other agents to bring back what would be more complex solutions. And I frankly, I don't remember, Eric, if it was you and me that had this conversation because it's happened so many times. But one of the things that I find missing out of those conversations is the what's the right word here? The diminution of trust, because you know, we're trusting our one agent to do our one thing, but it's now interacting with other agents to accomplish multiple things, or may just it may just amount to one giant complex task. But in so doing, we have commissioned our agent, given it our own agency and a certain amount of trust that went along with that. And that agent is now giving away that trust to whomever it's interacting with. And it seems to me like the people who are developing these great technologies have only taken that idea of trust to the first step. We're just learning how to trust AI that we can see and feel, so to speak. What happens when that agency is then given out to the next, you know, exponential permutation of agents who are going to work on things? What happens then?
SPEAKER_01:Yeah, I I mean, I I think there's a long way to go. Um, because that that trust issue, I mean, even with the AI that we have now, we've all experienced, you know, whether hallucinations um and getting inaccurate information. Um, and you know, so we've experienced that as individuals as we use chat GPT and Claude and others. Um, but in the corporate environment, um, it's a lot worse because then you're kind of betting your job on some of these recommendations and insights that are coming from these systems that you may not understand how they work um and they may be wrong. So, you know, if if the large corporations aren't getting it right, um, you know, how is that going to trickle? How soon will that trickle down to us? Because doing the creating these um agentic apps, as you said, there's just so much trust involved. They're they're managing so much information. Um, and there are so many edge cases um that it seems hard to figure out that, you know, how to do that. So, like one of the um, you know, an agentic app like for travel could be um, you know, reschedule any meetings that overlap with my son's volleyball games, or book a hotel if I have more than two meetings um uh a week in my next trip. Um, and the agent can go ahead and look at the schedule, check um all of your rules, book the hotels, um, you know, leverage your credit cards, all within certain guardrails. But to do that, it requires a massive amount of information, a massive amount of um uh of trust, a lot of connections. So it's it's hard to do this.
SPEAKER_00:Yeah, it's um I think it's and I was talking to somebody who who runs a division of a company that builds these agents early stage. Um, and it's really it's really quite easy, apparently, to develop an agent that is a single or maybe you know, single-digit number of tasks it has to perform um and interact with. But um you get into a more complex situation and uh the decisioning technology has it seems like has yet to catch up. But also you lose you lose the ability to really track and understand the veracity and integrity of those other agents really quickly. Like you're out in no man's land in a hurry.
SPEAKER_01:Absolutely, you know, and especially if you think about like what's what's the future look like. Um, so you know, I have my personal agent, and the brand that I'm working with has their sets of agents, and the agents are talking to each other, um, which is kind of weird, um, and then and making decisions on our behalf. Um, and then, you know, that there that requires an awful lot of programming. And I'm not talking about code, of just understanding what my desires are, understanding what the brand's desires are, and then to go ahead and create that handshake of a deal, whatever that deal might look like, is right that there's a lot in there.
SPEAKER_00:So you and your team at Vectorate Checks do a lot of work in the healthcare and pharma worlds. And, you know, if you're gonna pick out one realm where trust and data integrity or process integrity are vital, that would have to be right near the top of the list. You know, maybe that and financial services, but healthcare, you know, literally your life depends on the integrity of these processes. And um I'm really interested as a as a person who runs a company that designs processes um within that highly regulated and uh world of healthcare. Like, okay, it's one thing for for Mike Giambattista who needs to uh have his agent go order a pizza and then coordinate that with the uh you know, ordering the beer at the 7-Eleven. I don't order that way, but you know, uh those are those are simple steps. Entirely different idea if your agent is running uh behind the firewall, so to speak, uh performing complex tasks in a regulated environment like that. Like, you know, there have to be some, and I'm hoping you're working on or somebody is, there has to be some guardrails that um that can be put in place so that you can you can trust the entire chain of of interactions there.
SPEAKER_01:Yeah, uh absolutely. Um and that it that's gonna be a huge challenge. Um, and when you're talking about healthcare and financial services and having these agentic um apps doing things, I I think we're a long way away from that. Um, you know, it's it's something that is, I would bet anything is the future, and there seems to be a lot of discussion and planning and talking about it. But um doing anything meaningful, uh, I think is still a bit away, and that's where you know people's imaginations, um, you know, the capability is there, um, but the maybe the ability is not there just yet. Right. Or the desires and and the willingness to accept that risk.
SPEAKER_00:Yeah, right now it's uh it's just a really neat thing that could be, but but the realities behind it are fraught with all kinds of of security safety issues. So when you talk to when you're talking to your clients, potential clients about agentic AI or or even about your kind of non-agentic uh projects, do you find that leadership is aware of some of these hurdles, pitfalls, problems, and you know, potential risks? Or is it kind of like they have to be enlightened as to?
SPEAKER_01:I believe they need to be enlightened as to um, you know, they're they're they're setting these mandates that we need to be, you know, we need to be leveraging AI in every single possible way. Um, and you know, they're setting these lofty goals. And then, you know, the people below them are setting their goals um, you know, based on that. And you know, there's still a lot of discussion of saying, you know, how do we use AI, you know, beyond using it for to improve my email and write a better sounding email, you know, which certainly does have value, but it's a very small amount of value spread across a lot of different people and interactions. Um so it's it's almost impossible to measure. Um, you know, the the real types of opportunities are in AI is when you go, you know, really vertical and redesign entire processes. And to do that, you know, getting back to the CEO, you really need the CEO or certainly C level to be supporting it because there's massive change that's that's needed. Um processes change um and you know, uh all sorts of process changes is needed to go ahead and make that happen.
SPEAKER_00:Um the this is a conversation that I think you and I would do well to just spend the rest of this conversation on and and maybe many future ones, because um the idea that AI is coming for our world, you know, is it a is it a doom and gloom view, or is it, you know, all um all roses out there? Depends on your perspective. But the the honest truth is, well, a couple of factors. One that you see report after report that says that um enterprise level AI projects are either failing or they're not being adopted into their fullest capacity, and there's lots of reasons for that. I contend that a lot of that is cultural, a huge amount of it is because hey, I got the thing to do the thing, but wait a second, if we want the thing to do all the things, it's an entire reworking, redrawing of how this company is set up, or at least that division. And I don't think I'm exaggerating there.
SPEAKER_01:Oh no, I I don't think so at all. Um, you know, that there's there's massive opportunity for process reorganization. I mean, McKinsey is putting out, it seems like a report a day on um AI. And, you know, the the reason being is they see, you know, massive technology spend there, but massive process reorganization. Um, and how do they go ahead and, you know, you know, help the sea level go ahead and and recreate these organizations? And it's just a tremendous opportunity for a company like a McKinsey. Um, and that's why they're all over it.
SPEAKER_00:Yeah. On the one hand, you see that uh AI is coming for all of the consulting and advisory jobs. Uh I'm not sure that's true. If you're a really if you get this, there's huge opportunity to advise companies on how they're gonna have to fully rewire their organization. Yeah, absolutely.
SPEAKER_01:But you know, the there are some areas that that where AI really is, you know, taking off. And there's a project that I'm working on that that's really cool. Um, and it's definitely um AI enabled. Um, and this is with one of the large farmers, um, and it's around the literature review process. So at the start of, you know, the the cycle to go ahead and and develop a drug, you know, eight to ten years, billions of dollars. Um and the first step in that is doing a literature review. And that literature review consists of, you know, once they have an idea of the drug that they want to develop or the thing that they want to go ahead and try and cure, they'll go ahead and they'll um work with a vendor to find all of the different academic publications that have been printed on this. Um, and they'll come back, depending on the type of disease, with, you know, 500 different thousand different articles, um, peer-reviewed articles. So the first step is how do you whittle that down? Um, and that's a lot of discussion and bringing that down to like maybe about a hundred articles. The next step is this excruciating process of looking through these really deep academic papers and pulling out point by point and putting it in an Excel document of really long rows and really many columns, um, all of the data points. And they go through that for the entire hundred different um, you know, uh articles that they're that they're looking at. And then on the pharma side, they do that again because they want to double check the work because all of the follow-on billions of dollars and years spent is based on this. So it's got to be right. So that process takes right now about six months and about$250,000. Um, and you know, for big pharma, they're doing a lot of these. It turns out to be a lot of money. So, you know, we're in um uh V1 right now of the product that we're putting together. And We've gotten it down to about two weeks, and it's going to be about twenty thousand dollars in in soft costs.
SPEAKER_00:Wow.
SPEAKER_01:And those soft costs are, you know, because they're not spending extra money. It's it's work that people are already doing. And the AI is doing all of the extraction. It's it's really cool, a really cool use case of AI that will really benefit everyone.
SPEAKER_00:Yeah. Yeah. What a great use case. That's that's actually pretty phenomenal. I'm I'm the reason I'm quiet is because I'm thinking through all of the LLMs, chat GPTs and the like that I've been using over the past year and a half or so. And I'd call myself a heavy user. And um, you know, you figure out the nuances and subtleties and hiccups of these things along the way, and you kind of mitigate for those in your processes and you know your workarounds, and and for the most part, most of the tools that I use have gotten better over that period of time. You know, they're learning and it's working, but um they're still not perfect, and I I uh daily uh get some sort of you know prompting request that has to be completely kind of reworked because either I didn't articulate it correctly or clearly enough, or and I'm capable of that because I'm you know I am the human here in the loop here, or the machine just came back with something that made no damn sense, and it happens. It happens. I guess the one thing I would ask for in there is that it would come back with a little bit of humility, it never really does.
SPEAKER_01:I never get the time to say I'm sorry, you know, and it's always encouraging you like, hey, that was a great question.
SPEAKER_00:That was a terrible question. I know. I was talking with somebody recently who uh runs a division of a company that that makes a lot of this technology, and um we we kind of took a forward view of you know, okay, take you know, take us 12 months in the future, and what is what does your world look like? And he kind of put it back on me and said, Well, you're a user of this technology. What does it look like for you? And I was like, Well, I'm putting it back on you because I'm interviewing you, but he said, you know, internally, he said, We have we're very clear on this idea that 12 months ago we weren't using chat GPT and our own internal LLMs um the way we are now. There, there we are rewiring processes, we're constantly thinking about how we can make what we do work better with these technologies. He said, Um, to the point that my whole staff is very clear on the idea that next year, if you look out 12 months from now, if we're doing the same things the same way, we shouldn't even be working here because we've missed the point. Right. And and um he didn't call it a mandate, but he said, let me just tell you, we're hyper aware of that idea that we're creating change, but we have to be prepared for the for the change we're creating.
SPEAKER_01:Yeah, that that makes perfect sense. Um, and you know, that that's actually kind of a nice way to think about it. Um uh of you know, next year we shouldn't be shouldn't have the same processes that that we're doing this year. And you know, that that might be the best type of KPI because you know, all of the other KPIs are uh a bit unknown right now, right?
SPEAKER_00:Um, if I can, I want to go back back into your world of pharma and healthcare. And we talked about that one use case, which I think is really cool. I mean, from six months down to two weeks or whatever that delta was, that's pretty phenomenal. Any other use cases that you can talk about where you and your team are designing or developing AI-based solutions for some of your clients?
SPEAKER_01:Yeah. So um, I mean, another healthcare one uh for a different company is um one of the big challenges is um finding the right types of process documents. So in pharma or in all business, you have to go through processes. Um in pharma, you know, as you can imagine, those processes need to be really locked down. Everything from the way you manufacture a drug to the way you interact with the FDA and what your legal processes are. Um so the problem is that all of these process documents are in tens of different databases. And those databases were developed at different times by different um people in the organization, by different vendors with different mandates. Um, and then you know, there's been acquisitions, so there's a whole new slew of different types of databases. So each one has its own unique way of trying to find a document. And some of them, it's almost impossible. They have these cryptic names. So the only way you can find some of these documents is either if you know the document name or you know someone who knows the document name. Um so one of the projects that um we just completed is we're working, we work with the team to go ahead and um they are scanning all of the different databases, coming up with metadata um based on specific keywords that we generated, as well as what the computer generated and it organized it. So now you can find it using natural language search or by keywords, you can find certain documents. But the real aha moment was yeah, it's great that you can find the specific document that you're looking for. But if you're going through a process, you also want to find what's the parent document look like? What's the child document look like? Um, because you need to know all of that. So the AI is able to identify that. And that's something that was never able to do before you could do beforehand. In addition, you can also find the exact same document in Chinese if it exists, or in Spanish, because these are all global companies and they have different manufacturing plants and different processes across the world. Um, so you know, being able to come up with the parent child and um sister documents, you know, was just the big win.
SPEAKER_00:Yeah. Think about the complexity that you just solved for or your team. Give them full credit there.
SPEAKER_01:Yeah, and a lot of other partners as well. Um, you know, it it was a big project, um, and it's still going on in in different versions, but um it's really made a big difference where they can find find the documents that that they need.
SPEAKER_00:So that that's I would guess that's a client project for a specific client. But sounds like, you know, depending on who owns the IP for that one, that has that has utility across not only the rest of pharma, but just think about the rest of yeah, giant.
SPEAKER_01:Yeah, and that's not, I mean, that that's not a unique problem um um or or solution. Um, as you stated, I I think there's a lot of different companies developing custom solutions and a lot of different vendors out there offering solutions along those lines.
SPEAKER_00:Yeah, interesting. All right, so I'm gonna I'm gonna put you in the futurist seat and uh let's let's see what you think your world is gonna look like 12 months hence.
SPEAKER_01:I think there's um gonna be a lot of time spent on process redesign. Um, and you know, and that's certainly a core component of the work that we do in finding out what are the personas and what are the journeys that they that they go through and what are the jobs to be done. I think there's gonna be a renewed attention to things like jobs to be done of what's happening today or yesterday versus what's happening tomorrow, um, once this new these new systems are in place. I think a tremendous amount of time is gonna be spent on that. And a lot of it, I think, is you know, we're gonna get it wrong. Um, and you know, that that's both a little scary um because of the power of these systems, um, but I think that's gonna be necessary to do and to constantly revise.
SPEAKER_00:Yeah. Um now I want to go down a whole nother rabbit hole on jobs to be done. Because um, and if you're not familiar with it, if you are familiar with it, you can skip right over this part of the podcast. But if you're not, there's uh there are theories and strategies that have been uh what would you call it, like 10, 15 years out in the wild there, driving the way a lot of more progressive companies kind of operate and decide on things. Really fascinating stuff. Well, Eric, um, it's a ton of fun jumping into the future with you and uh look forward to the next time we can do it. In the meantime, just encourage everybody. Uh Eric's company is called Vector HX, which I believe means human experience. Stands for um you'll see a link to that uh with this podcast. But as always, Eric, it's a ton of fun. Thanks for doing this, Mike. Thank you so much.
