Qlik Connect 2025 Delegate Roundtable on Agentic AI

We are here at Click Connect 2025 with the Tech Fuel Day delegates, and we had, uh, some great in-depth presentations on the first day of the event. We've also seen some incredible, uh, keynote presentations, uh, deep dives breakouts. We're actually here on the, uh, expo floor where they have all the breakout sessions, and we've been watching those for the last two days.

One of the points though, that I wanna make is, um, well everybody's talking about Agentic ai. You'd think that, uh, if last year was the year of AI and, uh, large language models, this year is the year of AI agents. Qlik is no exception.

That was one of the two main focuses of the, the major keynote, um, the other day. And the question is, agentic ai, is it real? So I wanna start by throwing this over to Brian Boin, who, uh, is, uh, maybe, uh, most up to speed on what Click is doing, and then let's open it up to the rest of the panel.

Uh, is AG agentic AI real? I think that's a very good question. Based on what we saw in the keynote yesterday, I think that Qlik had a big driver last year in terms of AI and the AI Council and making sure that was a very public thing.

Um, AG agentic has kind of crept up on us in a way that ai, we had a bit of precursor to it, but Agentic has just kind of appeared out of nowhere, and it makes you think what is real when you're watching it. Um, and you're seeing the keynotes. And I think what, from the internal click community, I would say one of the big things that people wanted to see was that seamless join between structured and unstructured data.

And with Mike Capone click, CEO and stage talked about how that, how that segs and how that ties together. And I think for a lot of people, that was the big announcement on the keynote stage yesterday. So I'm keen to dig into it a little bit more, but I think I do need to see a little bit more before I can make a field decision on that.

That's my initial thoughts on it. Yeah, I think there was a fair amount of, uh, let's say messaging and talk about Agentic AI and enabling agentic AI here within the click platforms. Um, but unlike a lot of, and maybe this is to their credit under, unlike a lot of vendors, they're not coming out and saying, look, here's our marketplace.

Here's our platform here, our four agents templates or what have you. I, I always think of Qlik as taking a bit of a more measured approach. And I do think also it's such a data focused company.

Um, you, you mentioned the, the, the structure of the new data lakehouse, right? I think in general, AI is as good as the data that feeds it and as good as the data that uses it when in inference. And so there's a lot of focus on that right now, and maybe we'll see more later, a little bit more practical, more, you know, platform producty type stuff.

So I think there's a question between, is ag agentic AI real at click and is agentic AI real AG agentic AI is real? Like this is not a question of as, as to whether or not companies are actively pursuing production products or have production, uh, projects in around agent ai. They do.

I talk to a company that is replacing their SAP system with Agent AI in a combination of gen generative ai. Is that a smart thing to do? Maybe, maybe not.

We'll, we'll see how that works out for them. The question is where's the opportunity for, uh, click click customers and Qlik Partners? And I think they laid out as good, if not better than average, uh, plan of how they're approaching generative AI in general, and specifically ag agent ai this year.

One of the things I really liked was talking to, uh, the vendors that are here and getting to see some of the presentations here on the show floor as well. And what I got from those was, was two things. The vendors all talked about how they've got customers now because they ran to do generative AI and are looking to do agent ai and their data sucks.

It's not in order. Um, and we've seen, at least here in the US what happens when you try to use generative AI to solve a big hairy problem with very old data that you may or may not understand. We get lists saying that there's people 150 years old that are on social security still, right?

So there's a lot of data cleaning that has to happen, and lots of vendors here are very excited for that because they can help with it, with using Qlik as a platform. And then one of the presentations I saw that was really good was from one of the Qlik, um, employees laying out, okay, what is this looking like? And, and how do we, all of it rolled back to your data.

It, can this be real? What are all the things around it we need to think about before we ever think about having a marketplace for our ai um, agents? So I thought that was really good.

Going back to basics, making sure that we don't make AI mistakes on the scale of 50, a hundred times with these agents. Yeah. Yeah.

I think ag agentic, you know, what's in a word in the first place, right? To me, AG agentic is more an evolutionary concept beyond large language models. I mean, we all have used large language models.

They are very good at what they do, but nothing more. And so what, what really the evolution is, is is to use multiple large language models and to add some reasoning to it, right? A large language model still does a lot of hallucinations.

And I think one of the, the, the requirements or description of, of Agen AI is to kind of add the reasoning from a human perspective is to reduce the amount of hallucinations. And so from that perspective, I do think that agen is is right. Uh, but again, what's in a word?

Yeah. From my perspective, what I really saw here, uh, was back to Gina's point, a focus back on data quality. Uh, I actually spent about 10 minutes talking with one of the clicky as they're called here I, what is this trust score thingy?

And he was able to explain it relatively well. Um, the interesting thing is, especially with what they're trying to do to get to AG agentic, is we are going to go and be able to merge together more effectively structured and unstructured data because everybody's been going after unstructured as like, this gold mine of material. Well, it came from somewhere anyway.

Where do you think a lot of it came from? It came from the structured data. How, where do you, where do you think you got your annual reports that you're referencing?

You know, it must have come from there, but it's also that we are spending the time to go back and do honestly as data engineers what we should have done in the first place to build out the metadata so that when you look at a field that says, EQ ID, what does that really mean? And is that important to my customer? So it was good to see that focus, the trust score to me was just about the most fascinating thing I saw here.

Yeah. So from my perspective, I do believe that agen AI is real. I believe it exists in pockets.

And like that William Gibson quote, the futures here is just not evenly distributed. I think one of the most sobering analysis demos I saw was, uh, idea of marrying structured unstructured data, tell a story, gimme five executive bullet points. What do I need to know?

Yes, I could look at a sand key diagram. Yes, I could read reams of charts and stare at 15 different dashboards, but what are the five things I would need to know? But the precursor to that was because they brought in comments from the field.

They actually ran spell check on the comments. That was one of the functions applied to that data set. Because people misspell things.

People have different sentiments. And so before you even get to the sentiment analysis, do you have that scrubbing of the data? And I think a phenomenal demo for maybe 2020 is, is there a version of that function for gen alpha speak because your workforce is changing, your customer segment is changing.

And so these models will have to evolve. And so I don't know if you've seen that YouTube video where the professor gives the gen alpha talk and the kids are all grimacing. 'cause unfortunately he got a lot of it right?

But he's a professor, so how do we professionalize these other kinds of new functions? We'll need to scrub the data of the future. So yeah, You're reminding me of a, of a demo that I saw, um, or you know, one in one of these sessions, um, where they were, uh, you know, showing a use case of, you know, uh, what amounted to, uh, I think it was customer data or order data, um, being entered.

And the fact that they could use a little AI agent to go poke around and see like was the, was the state of being in the United States or some other, other country, province, whatever, was it entered with the correct spelling or the correct code mm-hmm. And correct it. And that's like a fantastic, you know, little menial job that, you know, speaks to me, especially 'cause I hate to see those kind of misspellings and I worry about how much we can trust the analysis when there may be that kind of dirt in the data.

But, so let's get back to the, the point though about agentic ai is what you're describing is any of this agentic ai, because truly is a, is a periodic batch job that cleanses data a agentic ai, oh, You want us to define agentic ai? Yeah. And I'm not picking on you, but I'm saying like a lot of the things that I hear when people talk about agentic AI is not agentic and not ai.

And frankly, I'm, I'm here for that because not everything has to be ai. Mm-hmm. Like, I think that a proper agentic workflow should include regular expressions, should include conventional database lookups should include like really boring normal stuff.

Yeah. I think as well as truly autonomous AI agents. Yeah.

I think you're overthinking if we're extracted away from the geek talk and the question of how do I get my cash faster as a, as a business owner, and how do I reduce the number of errors, SAP Oracle and all of these, uh, ERP software companies have been promising the ability to take a invoice, scan it in, and get that paid, but invoices come in different sizes, et cetera. And we've tried to use machine learning in the past to correct that problem, LLMs solve that problem, the ability to do this transform, act upon a, uh, invoice that correctly identifies handwritten notes, et cetera, and translate that into business logic. That's agentic, uh, in a sense that I no longer have a business, I don't have a, I can take my business analyst and put them on a much more difficult problem than this mundane solution of trying to, uh, triage when OCR doesn't work.

So when, when I think of vient, number one, I think the name is stupid. It's just creating an adjective out of a noun and just to do it for marketing. And I'm a product marketer and I don't like the name.

So we'll start with that right there. But then when I think about like, what they're trying to do, I, I, it's almost like it is building bots. You know how when a OL was around, I don't know if there's anything more modern that did this, and you would have a bot that you would use to kick everybody outta the chat room.

And it was hilarious, especially people you hated was in there. So that's how I see it. Like, there's a specific job I need done by a specific bot, and there's almost needs to be a, the, the bots, I'm gonna call 'em bots.

The bots create, you create a workflow outta the bots to go do certain things. So if I have to do a lookup in a, in a traditional database, you should, if we're gonna use a gentech, and that's how it's gonna be, you, you should try to use it in the mo use, um, the tools against your data in the most efficient way. You don't u need to blast AI and do an additional data when you just need to know the facts.

This is the country and the state of a certain place. Um, and then as things go on, maybe you need an LLM to do something. And so you'll have a little agent to run, do something over here, and it's gonna be a little workflow that's contained by.

And that's, it's, it's a whole thing, but it's gonna take design and that's gonna depend on the data. And so just tagging onto that concept, I think there's an a that we've not talked about, which we need to, which is automation. So automation is another a that gets confused in this whole AI agent circle.

And something that Qlik does has done really well is automation. Now, uh, by, by their own admission on stage during the keynote click's definition of agent AI is different. The topic of the conference is due data differently.

So again, I think we have to try to like work out the wood from the trees here. Whether does AG agentic mean an automation that has ag agentic properties or does it mean true egen AI that sits underneath? And for me, that's one of the open questions that wasn't quite answered yet in the demos that I saw.

Well, I think with agen ai, not everything is ai, you know, to Steven's point, right? The only thing in my book, what Agen AI says is that there's one or more large language models under the cover. But if you look at some of the blog diagrams that vendors are publishing a lot of those agents, because agentic is just a, a bunch of agents, some of those agents are just regular expressions or simple tasks, but the whole is basically agentic, um, labeled agentic.

I Don't know. You know, I I I feel like it's, so I, at the beginning of all this, I, I used to say, I think I wrote a column on it, that there's no such thing as an AI application or an AI app. I think some of you have heard me say that before.

There are applications that are powered by ai, one or more AI services Yeah. A chat bot has is predominated by the large language model, AI service behind it, but it's still a chat for all. You know, it's a person behind it.

The application is the chat. And in the same li way to me, an agenda AI is an agent, and it uses, might not even use a large language model, but it uses one or more AI services to carry out its few. These, there's just one difference that I would say, one difference, which is that agents have goals, and those goals or their methods for achieving those goals can be shaped by a little AI loop, right?

That makes it morph a little bit in how it carries out its goals. That's the only difference. I, the thing I, I would be impressed with ag agentic ai if I said one of our examples was, uh, here, right?

Coffee sales are down in wherever, right? In in Finland, uh, Finland or Buffalo, New York, what can I do to increase those sales? If the AG agentic AI came back and said, Jim, you're not asking the right Questions.

Right? And that's what I think is still what, where, where I am holding back. Uh, we just saw a demo here today where they were talking about using this in manufacturing and there were a bunch of, you know, bullet points being returned about, uh, the sales volume in Buffalo, New York for whatever it might have been.

Right? But again, is that what I really need to know to answer the question is why coffee sales are down in Buffalo, New York this week? Because there's so many different questions that I could be asking.

Is it a supply chain? Is it warmer temperatures? Is it, you know, and that's what I'm, I would expect an agent to say something along those lines.

Here's the right question to ask. So, so to make that, like the ledecky section where she talked about, she started with her first coach. Mm-hmm.

Then there was the second coach picture, then the third coach, the fourth. That's interesting. Apparently she's gonna stay with the fourth coach.

Right? So maybe AG agent AI for it to be real is instead of us thinking of it as the player on the field, it's actually the coach. Yeah.

So maybe that's the arbiter of like, now we're an ag agentic, it's a coach we can rely on. Thanks for bringing in Katie led. Okay.

I just, It was an important part of click connect. But I wanna point out too that, um, I was watching that same demo, and again, I don't wanna pick on this demo, but its suggestions were asinine, its suggestions were, um, a, a happy hour special mm-hmm. On Right?

What I think it was week beer or something like that. What's Really interesting though is imagine instead of you asking why are coffee sales down, you get a note saying, the agent noticed that coffee sales are Down. Yeah.

And see that's, that's exactly it. That's when it becomes, I think, more useful and more agentic when it's actually continually monitoring things and alerting you. I think there was a better demo where it showed like, um, proactively from the stage it showed like, you know, sales of this are down, and I noticed this, and so I sent you an alert because I don't know about you, but I get a lot of useless alerts and I would much rather have an agent keeping out from keeping a look at my stuff, sending me useful alerts.

So I think the, the expectations for agentic AI are too high. That's not what people are actually using AG agentic AI for today. Yeah.

What I'm seeing people use agentic ai, when you're talking to a CD or a chief data officer, and he or she is complaining about that 80% of their data scientists time is wrangling data. It's cleaning up data. What happens when I can put a agent against my 18 data sources to, so when I'm doing my ETL, I get clean data on the other side.

Yeah. Is that not agentic ai? That is.

And, and in my past in companies I worked for, we really had a team called the Dallas Six. It was literally six people suffering the details of the data, right? Six that was eventually replaced with what we now refer to as an iPads.

And the great thing about that iPads is it worked until it didn't work. So to your point, there was no loop that was changing its goal as the business needs changed. Mm-hmm.

And so even iPads had its time in the sun. And now if it's agent, does that mean you're coupling like an iPad like function, but giving it more agency? This is why AI projects fail, is because there's this desire, I want to ask the AI why our coffee sells down.

There's no business leader, even the small business leaders that's gonna ask an AI agent, why are coffee sales down and take that advice? They're going to, uh, get lower level reports and do some of this analytics up upfront. The companies are, the companies that I talked to have failed in their AI journey is because they bid off more than what they can do.

They've, they've watched these demos and that be, that's become the goal. Yeah. I can't imagine Why they thought that they could do more with AI than they actually can.

I think that's the whole point of this, because everything you guys are saying, honestly, we said 10 years ago and we solved with, we solved with scripting, we solved with different automations. So this has been done before. And, and so for this to be an an AI agent experience, maybe it's just an agentic experience.

We're gonna build little bots that can do things and we're gonna trust them, which I don't, and I don't think business leaders will either. And I, I think that's the whole part of this, is how do you get the data to a place? Can you use, um, different agents, like Keith was saying, to get the data cleaned.

Can we trust that they got it cleaned the way it's supposed to be cleaned? Mm-hmm. Can we then use that data to build pipelines that, that do things to the data?

And a continual is continual basis that can give us insights to our business that we can act on. So there's never gonna be a time when the humans aren't involved. But there will be times when we can get six humans that are very smart, that understand the datas and understand how to query the datas, and they can, and they can be moved on to something that's more worth their brain power time to solve what, but this has to be looked at, you know, this is what we got in trouble with with ai, especially generative AI thinking it can do a bunch of things that it cannot do.

So if we're looking at building agents to do things autonomously, that's way different than letting agents do things, um, automatically. Yeah. Right.

So, so we gotta figure out like, how can we trust it? When can we trust it? And none of it's gonna happen until the data's problem solved in the front space.

And I think there's still quite a serious market pacing issue here that, that, that, that we, we were talking about AI last year, and now we're talking about AgTech, and now we're talking about AgTech in production. Now, first of all, who asked for that? Second of all, who's ready for it?

Third of all, who's ready to implement it? And I can tell you from multiple discussions on the floor, that many of these customers, and it's not just native to click, are sitting on-prem with data that is not cleaned, and they don't know how to identify it and clean it up. So to take that leap and talk about merging AI and identification and automation, we're Still Like two or three steps away from that in most cases for some customers.

And I think there's a gap there that hasn't really been discussed before. We hear the announcements and all of those things. So I just wondered what you'd seen based on your conversations on the floor.

So the one demo that still sticks with me, I guess it's now two days old in my brain, so I've had some time to reflect on it, and it still bothers me. And it was a demo that showed a person, and it, you know, I'm not sure if it was a Slack or a Teams, but it was some kind of a chat, but it was them, it was their little avatar, their, their face, their profile, their name, and you were asking it questions, but it wasn't him. But how would you know it was him typing or the agent doing the typing?

And that's the, that was the, that was the uncanny valley moment for me. Is does the bot, to your point, you know, does to Genus one, does the bot have to have a name? So we know that's a bot that is not Brian, that's a bot.

It was me by the way. It was, yeah. But, uh, that's the one that kind of, that's a slippery slope for me right now.

So what you and Gina are talking about, and Brian, are critical aspects of a successful AI project, which is observability, and it we're norm it, and even in, uh, six Sigma processes we're accustomed to observability getting these, uh, short feedback loops, feeding it back into the system, and monitoring for exceptions, et cetera. Business at large is not good at that. And, uh, we need, I think, uh, Jay, to your point, labels that says, Hey, this result was via ai, that long dash is a really good indication that Keith's blog post was rewritten by ai.

Yeah. Or, or is like, is it Keith during the day? And then there's the Keith version after dark.

And then do you get to be like in, uh, what was it, interstellar movie, you know, set humor to 75% or whatever doing the day is Keith at Knight is playing as Keith in the day, the day. So in summary then, uh, what do I, I wanna give y all a chance to, to kind of quickly respond. We've had a conversation, um, after being here at Click Connect.

What's your opinion? Is ag agentic ai real, and I, and I will take Keith's point that it's not just, is it real generally, is it real for click? Um, answer both of 'em.

Is it real? Is it real for click? So two, two points.

First off Keith's point that customers who are trying to do what they see in the demo are failing. And, but I think Keith, the majority of the customers I've spoken to know what they're up against and they're taking a more measured approach. So that's the first point.

Um, and that would be too, actually having agents that are AI fueled. Right? The second thing though, and, and both Brian and Jim alluded to this early, is this issue of trust and what the Qlik AI Council, like they, they announced a couple of days before the conference started, um, that it made this sort of broad statement that, uh, AI will not scale without trust.

The word trust here can mean a lot of things. But here at this conference, it means trust in the results, trust in the outcomes, and, uh, the trust score. And all these things have to do with being able to observe, understand, get an audited result out of like, why did the AI do what it did?

And I think that's one of my biggest takeaways from this conference as a whole. Ag agent, AI is real, but like all things I, I compare it to when VMware, for those of us in the infrastructure realm calls the software defined data center, and I say, I want to buy be a software defined, I like two or three of those software defined data centers. It did not exist.

It is the, uh, it is the, the halo of the overall capability. And the industry is moving towards that, uh, towards that vision. Qlik is no further ahead or no further behind any of its competitors in that race to achieve agentic ai.

I, I would say it's right now agent a ai, based on what I've seen here and what I've seen at other conferences, it's aspirational. Won't it be cool when we can do this hand this task off to an agent? It is inspirational, but it's not here yet.

It's so, it's not perspiration, it's not per, it's, um, isn't genius 98% perspiration. Okay. I mean, I would say from inside the ecosystem a little bit more, I would say that what Qlik have done in this conference is they've solved a problem that people have been asking for, which is tying together unstructured, unstructured data.

Now, what we still, it's still too early to tell is that AgTech that is doing that, that's what we are being told, and that's what we're seeing in the demos. But until we understand more, we don't know how long that's going to take to actually prevail. And I think when you, I mean, we haven't talked about Open Lake House and Iceberg and all that kind of thing, but that plays a massive part in this.

And how those two things stay on track, does one accelerate ahead of the other? Do they, do they work in a pair? Those things will all come out in the wash in the next six months, I would say.

So I'll go with an analogy. Uh, playing off of the word lake house. I think we have a great lake house.

It's a beautiful lake house. The lake was formed from melting an iceberg. Uh, we're gonna have a party.

Mm-hmm. There's a beautiful lawn, there's chairs on the lawn, but I don't know if we yet trust our agentic caterers to actually serve the food right side up or upside down on the plate. I also don't believe we have our guest list fully figured out.

I think there's pockets of the guest list. We think maybe HR is definitely gonna be that invite list. They're the first pitch, then that invite list.

Second person invite list might be someone in fulfillment or inventory. I don't yet know if I have confidence that there's a difference between like a Chief revenue officer's gut and what the top five bullet points of what beer I should be selling right now is a special deal, is something that the adjunctive speak, you know, trusted with. So I think there's gonna be a Lakehouse party.

It's gonna be an agentic party. I don't know when the party begins. Yeah.

So I, I do really think agentic is real. Um, and it's great that there are companies like Qlik who presents a different kind of solutions to customers. So we can use, but to me, the bottom line is customers have problems and those problems have to be solved.

And in the end, if it's called Agentic or something else, it, it really doesn't matter to me. I think Gentech still is kind of a marketing term. That being said, I, I think also there's a lot of confusion about Agent, where agentic is not one thing.

AG Agentic is a, a series of agents, and one agent can call another agent. And to Steven's Point, the agent doesn't have to be an AI agent to start with. That's A good point.

Yeah. I agree with everything everybody said. Pretty much.

I wanna get, you know, the thing that I saw the most was, um, I, I did see from presentations from customers, I think Penske was one of the ones I watched. And then from talking to people on the floor with their materials that they had, um, people starting to create frameworks of how to move from the cleaning the data to, uh, a real enterprise way of managing, uh, just generative ai. Mm-hmm.

Right? So they're just starting to get their heads around that. Yep.

And I think that framework is probably very good and will be beneficial for moving ahead to figuring out how they would implement, uh, gen, uh, Gentech workflows. And so I, I feel like it, it, it is just the next evolution of computer science. Is this how we're gonna do it?

And, um, we know that we have to slow down a little bit and figure out how to bring everybody to the party. Yeah. Oh, I, I started it all, but I just wanna know if next year we're gonna be talking about Vibe ai.

Ooh, vibe ai, anyone? I, I don't know what the answer is to that one, but I I have a feeling that there's gonna be a buzzword next year. Yeah.

And we're gonna be talking about it. Yeah. So last year it was, uh, rag.

Yeah. This year it's Ag Agentic. Uh, next year it'll be something else.

You know, it it, it gets me thinking though the contrast. So this is my summary, I guess the contrast between where people are to Brian's point and where the industry is headed. It's important that, that companies like Qlik have that they're out leading, right?

The worst possible situation would be to come to the conference of the company and have the company be like, you know what? We've decided not to innovate. We're really gonna focus on solving the problems you have today.

And that's the only thing we care about is fixing all the things and getting you all up to speed with what we announced two years ago. That would be the world's most boring conference. It would be practical, but that company would be, that would be trouble.

Companies have to lead. They have to be out there, they have to be showing a vision. Um, I think they should be clear on what's a vision and what's a product.

I think they should be, uh, clear on what, you know, what the timeline is, what the reality is gonna be. I don't want to see like pie in the sky visions. We don't need a flying cars announced at conferences, but we do need a vision that leads the industry forward.

And also I think companies have to be very careful that they're always measuring the distance between the vision and reality. Because if you get too far out, and some companies do, and, and that's a real problem at these conferences, if you get too far out, then it's basically irrelevant and you end up losing the audience. And I think that Qlik hasn't done that.

I think they've kept us grounded enough that customers can, can come here and they can say, this is where we are, that's where we're headed. Let's try to get there even though we know we're not gonna get there tomorrow. And I think that's a, that's a nice balance there.

So thank you guys so much for this, uh, this great round table discussion. If you enjoyed this, uh, check out the Tech Field Day YouTube channel. com.

Uh, we had presentations deep dives into a lot of these sessions during Click Connect. And of course we've got a lot of other content there as well. And if you enjoy these kind of discussions, we also have the tech field, a plus YouTube channel where we have podcasts and behind the scene videos and reactions and so on.

So thanks for joining us here at Click Connect, and uh, hopefully we'll see you next year.