Agentic futures: How AI agents will transform research over the next 18 months
Editor's note: This article is an automated speech-to-text transcription, edited lightly for clarity. To view the full session recording click here.
Just as everyone was getting used to generative AI, a new type of AI has been introduced, agentic AI.
Bulbshare from SMG’s head of content, Nina Glynn, moderated a discussion between Seth Minsk, senior director of global insight and analytics at Perrigo, Juan Planillo, head of product at Bulbshare from SMG and Jaisal Mistry, insight specialist at Asahi Europe International.
The group discussed the differences between gen and agentic AI, the effect agentic AI could have on the industry and gave tips on creating your own agentic roadmap, during this 2025 Quirk’s Event – Virtual Global session.
Session transcript
Joe Rydholm
Hi everybody, and welcome to our presentation, “Agentic futures: How AI agents will transform research over the next 18 months.”
I'm Quirk’s Editor Joe Rydholm. Thanks for joining us today.
Just a few quick housekeeping notes that we won't have a formal Q&A with the session, but some panelists will be in the chat during the session. So, feel free to submit your questions or comments using the chat tab during the presentation.
Our session today is presented by Bulbshare. Enjoy!
Nina Glynn
Just got your head around generative AI? Good, because that's now yesterday's headline.
Yes, just when you thought things were getting comfortable again, there's a new contender for your attention, your roadmap and your ways of working. And this one comes full of possibilities.
Welcome to today's webinar on Agentic AI and what it will mean for insights.
Today we're joined by an absolutely stellar panel to unpack those opportunities and challenges.
A big warm welcome to Perrigo’s Senior Director of Global Insights and Analytics, Seth Minsk. Hello, Seth.
Seth Minsk
Hello.
Nina Glynn
Hello.
A warm welcome to Asahi’s Insights Specialist, Jaisal Mistry. Hi Jaisal. Pleasure to have you.
Jaisal Mistry
Hello.
Nina Glynn
Hi.
And last but not least, a very warm welcome to Bulbshare from SMG’s Head of Product, Juan Planillo. Hi Juan.
Juan Planillo
Hey, Nina.
Nina Glynn
Hi.
I'm so pleased to have you all here today and I can't wait to get into this very new and exciting topic with you all.
But before we do dive in, there is one last intro, a very quick one from us at Bulbshare from SMG in case you're not yet acquainted with us. As a platform of SMG, we harness the power of always on bespoke insight communities to empower real people to shape their experiences with products, services and brands. Co-creating the journey at every touchpoint right from the start.
We are delighted today to offer you more information on Agentic AI, on Bulbshare or any unanswered questions you have at the end of this session in personalized, one-to-one, follow-up meetings.
So, look forward to those. Do reach out. And we'll also have someone in the chat answering any questions you have as you go. So, please do get involved, drop your name, where you're from, how you're feeling today and any questions. We'd love to hear from you.
Right then I think we can kick off.
This is very new and exciting iteration of AI and I don't want to take it for granted that we're all experts on it. We're all deeply familiar. We know what it is, how to wield it. So, we are going to start with a nice definition.
Juan, I would love to come to you to get a bit of a definition on Agentic AI, what it is and how it's different from the AI we know and love already.
Juan Planillo
Yeah, thank you very much, Nina.
As you said, I think that everyone was getting comfortable with the generative AI terminology and suddenly agentic AI comes to play. I think that the basic difference is that generative AI is reactive.
I think that we all have been playing with this technology. We have all got familiar with the terms of prompts. We've been putting prompts into some systems and essentially getting answers. And agentic AI is coming here to change all of that.
Agentic AI is really proactive. It is taking those tasks autonomously and essentially working in a space that we define, given set of boundaries.
I think that a simple way to put it is generative AI is a talented assistant that is waiting for you to give orders. Then agentic AI is like a coworker that is working with you side-by-side helping you to achieve your goals and objectives.
I think that, in the concept of research, that is massive because essentially, we are changing the way to work. And I think that is something that we'll need to get used to and see how our workflows will evolve.
Nina Glynn
Absolutely. Yeah. I love what you're saying about the sort of shift between that partner in research.
We've gone from a copilot before to maybe now autopilot and a project manager moving beyond the generation of images or analyzing data and writing scripts, whatever it may be, to actually autonomously and independently running things potentially. Which is very exciting.
I imagine our minds are really buzzing with possibilities here. Lots of ideas for where this could take us. So, I want to capture some of those initial reactions. I'd love to come to both Seth and Jaisal on this one for a bit of an overview of why this is so exciting for research.
So, Seth, I'm going to come to you first. What makes this next phase of AI so transformative for the research industry?
Seth Minsk
Yeah, and I think in a next phase is really the operative term. It's just the next phase. AI was not invented three years ago when gen AI made such a big splash. It was a leap forward. It was a tipping point certainly, but it's technology that we've been using for years that is continuing to evolve. This is the next iteration.
So, what we're really talking about is advances in the convergence of math and computational power that led us do sort of more interesting, more useful things and it will continue to evolve.
We're talking about agents now, but a couple of years from now there's going to be some other new breakthrough that's going to seemingly change everything that we do.
But the way that I think about it is that if Gen AI helps us with the ‘what,’ it helps us generate content, then agentic AI really helps us with the ‘how.’
How do we bring this into our business? How do we operationalize it? How do we really fully take advantage of it?
After we were all blown away by what gen AI could do the first time we saw it three years ago. But it was still a solid year, 18 months until we saw the commercially available applications of it from an insights perspective.
I think agentic is really going to be key to how we scale this. To how we get tangible outcomes from it, which I think still feels a little bit difficult with generative AI to tell what's novelty, what's fun versus what's something which is really a solid output that I can build my business around.
But agentic is really going to be how we build workflows around the AI and really take advantage of its efficiencies and breakthroughs.
Nina Glynn
Yeah, absolutely. I love that shift from ‘what’ to ‘how.’ And you're so right in saying that obviously AI has been around for a very long time. We're just seeing these new phases and getting to grips with them. So, yeah, absolutely.
I'll pass to Jaisal. What do you think? What's next with this?
Jaisal Mistry
Well, I think from my perspective of everything, there is a lot of AI in research that goes on at the moment, which is very reactive. And agentic AI probably gives everyone within the insight industry the chance to become more proactive in that sense.
AI that's more goal oriented, which is always a really big plus with all of these things, and can fit within the wider term objectives of what you are trying to do as a business, factor that into how you’re working and how you’re researching things.
Again, as Seth was saying, the transformative way in which AI is meant to handle some tasks that you give it and what it can handle going forward is going to massively affect timelines and whatnot. But yes, it's just very much being able to give yourself as the researcher, a lot more power in that sense to make a lot of those strategic decisions in that sense, which is really exciting in that way.
Nina Glynn
Yeah, absolutely. Shifting into more strategic decision making. I love what you said about being more proactive as well. And agentic AI isn't just generating insights, it's taking actions based on them, which is really exciting.
With that in mind, Seth, I'd love to come to you to understand what the impact is going to be on use cases and whether we foresee potentially entirely new use cases coming out of this. Any thoughts on that?
Seth Minsk
Yeah, so I have three that I want to talk about.
The first one, following on something that Jaisal had just mentioned. So, I think automating the proactivity.
When I come into my desk in the morning, I have meetings, I have work to catch up on, I have e-mails to catch up on. I don't have the time to get ahead necessarily. So, I think automating some of the more proactive activities that are a little bit more tedious, a little bit more laborious, environmental scanning, automated discovery, automating the search for signals and bringing those to me. It doesn't mean that that's going to be it, but if it can help lead me in a direction where I know that I need to deep dive on items one, two and three, but not four, five and six, but if it can help surface the signals, that's really what takes a lot of time and effort and I think that is a huge win for us.
Second use case that I want to talk about, which is a little bit more tangible and specific, is from how we bridge from insight into marketing execution very tangibly.
Every morning when I come in, tell me what's going on with my brand, tell me what's going on with my category, tell me what's going on with my consumer. Save me from having to spend six hours going down TikTok rabbit holes, help me at least a little bit separate what's noise from signal. But then give me 10 activation thought starters, which within my current brand campaign, within my current targets are things that tomorrow, today I can go out and I can build with my agency, with my automated AI content tools, and that I can place an online media for activation tomorrow. And oh by the way, go out and automatically test that among synthetic audiences so that you're just delivering a prioritized list to me.
Again, that doesn't mean that we're going to just go take that creative that the AI generates and go out and place it without human thought to it, but it helps us execute faster.
Then the third thing I want to talk about, a lot of this just really sounds like science fiction, but this is where we are. I've been hearing more and more about brands, marketing agents, marketing directly to consumers, automated shopping agents.
So, there's this whole sort of marketplace going on behind the scenes of negotiation that will end up in purchases. And it's interesting to think about. Will I be surveying consumers AI shopping agents rather than consumers? I'm not really sure.
Will this sort of automated exchange, which maybe works like some of the advertising exchanges have worked for the past 15 years. Will that throw off metadata that I'm doing things with, which of course, then I would want an agent to analyze that data and tell me what's going on.
So, I think there's a lot of really interesting stuff that's potentially on the horizon.
Nina Glynn
Yeah, absolutely. I wish I had an agent to help with my e-mails this morning after coming back from holiday. So, really like the sound of that.
What you're saying about bridging that gap from insight to marketing. One of the examples I was hearing about was if an AI detects declining brand sentiment, like you were saying, rather than us going through TikTok and doing all the social listening. It could autonomously brief an agent to create with generative AI, then test a different ad concept and run these sort of iterative tests continuously.
It's really exciting. So much opportunity really shifting from episodic and one-off bouts of insight to this continuous and potentially really autonomous research cycles. So yeah, it's really exciting. So, thank you very much for that overview, Seth.
As we know, agentic AI can operate persistently, and it can gather and interpret data without needing prompts, which is one of those things we've been getting used to for the last couple of years. So, what does this mean for our processes I wonder.
Jaisal, I'd love to come to you on this one. How will our ways of working change in light of this new iteration of AI and what are the practical implications?
Jaisal Mistry
Well, I think one of the things that has surfaced, with what Seth and yourself have been talking about, is the timeframe that it takes to do a lot of research. Be it through scraping or surveys and even just recruiting a database to people and whatnot. Agentic AI will help reduce a lot of that time down and the time it takes to do all those things.
I think a firsthand example of how processes will change is that things will be a lot faster than they were before.
I think the next thing that you're thinking is that Agentic AI is going to be always on in how it works, where if you are scraping for trends and whatnot and for getting an overview, it's able to provide those insights, provide suggestions of what you might need.
So, you as a sort of an insight worker or stakeholder can then make lots of those strategic decisions and strategic planning decisions right off the bat of things. Enabling yourself more time to do those things and work in the ways that you probably would enjoy to work more and become more of an orchestrator and a strategist instead of someone who's processing data is a really big plus. I think having that time back to do some of the thinking skills that a lot of people have basically been employed for and what they specialize in is great.
The other thing with how agentic AI works, and something that excites me, is some of the ways that it can probably react for the consumer, especially if you're doing surveys and whatnot. The way that it can ask follow-up questions in research is really interesting.
I think if you've got a way that you've got a survey and you are oriented towards your goals and you put the right frames and stakes on the AI, it's able to ask follow up questions or automatically within a survey when someone's asked something. And I think being able to gather insights from that, which you may not have already considered in the first place is really useful. But having that already there and maybe having that as an expectation is a really good change in process in that sense.
I think there's ways in which your process has changed, where the timeframe obviously becomes shorter, but how you then think it becomes very different as well.
So, the ways that you want to work in your skillset will change from being someone who is potentially very good at Excel but then being someone who is very good at prompting AI to be good at Excel. So, it is all of those sorts of things where there's maybe some soft skills that we need to learn, how to speak to AI properly and whatnot too. There are minimal ways and also it is a major ways too.
Nina Glynn
Yeah, absolutely. Well quite liberating what you're saying around the creation of hopefully more time to do the things that you set out to do that strategy. We'll delve into that in a little bit more detail in a second.
But yeah, absolutely. This kind of shift from static reports to this kind of living, breathing ecosystem of research that's always on, like you say, that's more continuous. And allowing us to get deeper with the consumer too, those follow up questions that are probing. Loads of opportunity.
We can shift from prompting the tour at these individual stages to seeing this more full cycle research that's running from survey design and deployment all the way to analysis and reporting. Potentially all without human input is the thought, I suppose.
Let's kind of dive into that. What does this mean for human researchers and for our jobs? I'm sure this is something we all want to understand more.
Also, a big question for qual research especially, which is so hinged on human understanding and empathy. How will we see that evolve?
Seth, I'd love to come to you on this one and kind of dive into how the role of the researcher will change when we can run research on autopilot presumably. And how do we upskill and future proof our roles in light of this?
Seth Minsk
We've been going through a long-term transition in what it means to be a researcher anyway. This is really just another evolution.
When I started my career, the role was market research or custom research and very focused on executing primary research. We've seen primary research become a little bit less important. We've seen the time and the effort needed to go into primary research, be cut with automation, platforms and DIY.
And the profusion of different types of data sources really evolved the consumer insights role anyway to one of data detective and storyteller and somebody who's a co-equal business partner with the brand team in driving the business forward.
We don't have that many caddy phone interviewers anymore. We don't have that many punch card programmers anymore. There's constant shakeup in roles in our industry and in every industry.
So, I really think that agentic AI is the next level of the toolset that AI has started to introduce that helps us become faster in matching the speed of the business, deeper at the same time, and agents are a force multiplier.
We're all being asked to do more with less. Agents will let me be in multiple places at once. My biggest barrier to really getting the right, real-time input into my business to affect business decisions is time. And so the more that can be delivered to me, that cuts down on the Excel work, on the time I have to spend scrolling through things. That lets me jumpstart my thought process to get to what does this mean and what do I tell the business to do, is a good thing.
So, if an agent can deliver that all to me so that I can get to the provocative thought starters that I can build on, then I think that that's a positive evolution.
Nina Glynn
Yeah, absolutely. I love that term data detective. And his idea that the roles are constantly evolving. And agentic AI helps you become better at your job and able to almost have more opportunities at the same time.
I think we are shifting to see that the AI is running the mechanics, as you say, but the researcher can really level up into the interpretation. That's the ethics, the strategy, as Jason was saying. And also what you were saying, Seth, about working on that provocative side of things.
Seth Minsk
And if I can just add to what I said before was that if gen AI gives us the ‘what,’ and if agentic can give us the ‘how,’ humans still need to give us the ‘why.’ And agentic AI really frees us up to do what humans do best.
Nina Glynn
Yeah, I love that. We need to clip that. That was great. I completely agree.
Also, there's this importance of, I think, our roles will shift to be masters of AI and we need to really understand its capabilities and its limits. So, we might see that shift because we need that oversight. We need the understanding of where the AI might fall short.
I think it'll be a really interesting next 12 to 18 months to really see how human empathy, human storytelling, human strategy fits into this wider set of tools that we are acquiring.
Of course, that's one of the concerns that we see everyone talking about on LinkedIn. Everyone's questioning what's the future of our industry, of our roles. But there were lots of other concerns and considerations I'd love to dive into here as well. I want to get into what the watch outs are.
The question is, what are the challenges to implementing agentic AI in our work? I would love to come to Jaisal first, then go to Seth on this.
So, Jaisal, if you could kick us off with what keeps you up at night with Agentic AI? What are the concerns here?
Jaisal Mistry
I think there's a few, but I think some general concerns with AI that are probably very common across businesses in the way that we implement them. And I think as AI becomes more integrated within our working society, that will slowly be ironed out in each form. And as AI develops, we will then probably have the same concerns and the same ethical, ethical concerns, but we'll be able to answer those better.
I think one of those is probably that AI can only be as good as what it's working with.
So, if you are integrating AI with your own system, your own data, there's always some challenges there. There's maybe other challenges in terms of how much you can trust something from AI. And how much then you are able to present something back to either stakeholders or to the rest of the business. How much will people believe that AI is telling you the right thing and AI is providing the right prompts.
So, that's where Seth is, right. You need the human interaction of the ‘why’ or the human inputs to be able to control what it's doing.
I think the other thing from my side when we are thinking about positive data and consumer interviews is if you do give AI it's own power to have the opportunity to answer follow up questions where you are researching towards a goal. In that sense, you don't have the power to control what AI says unless you put the right guardrails on and you are not able to control that.
You don't know if, say a consumer's been going into an interview and there's potentially something negative their AI may come up against. And these are all sort of, I guess things that will probably be answered over time, potentially by another AI agent that can control this. But yeah, there's things that will be raised in that sense.
Nina Glynn
Yeah, absolutely. I totally agree on that point of trust and sort of stakeholder buy-in as well and putting in the right guardrails. But also what you're saying around what we feed into it is only as good as what it's got. It kind of comes back to what Seth was saying in this kind of loop of AI feeding on AI again. So yeah, lots to unpack there.
Seth, anything you would add in terms of challenges, concerns or considerations?
Seth Minsk
Yeah, I have just three quick things. Some of these will build on what Jaisal said.
Number one, how much control am I comfortable giving up to the technology?
And then hand in hand with that as I'm designing it, not only what are the guardrails and how do we program those guardrails, but where do I make sure, even if it's a little bit inefficient for the automation workflow, where am I making sure that the human is inserted? So, that's number one.
Number two is this is going to have implications around fundamentally redesigning business workflows. It's not just building an insights agentic workflow, then there's output that fits in with something else. If I, as an insights team am able to deliver, go from insight to validated innovation concept instantaneously, overnight, whatever that is, but if my product development cycles are still 36 months, what does that mean?
Or if my creative cycles are still six to eight months, what does that mean?
Do I really even need to be doing that? How does that mesh together?
The third thing, the thing that I wanted to add around the input data. We know AI is only as good as the data that feeds it. And I think part of our role evolves a little bit to data curators. Where we're making sure that we're continually feeding fresh and relevant data into our AI algorithms.
What do we need to do to keep back testing our data to make sure that the data is still up to date? When we need to add more data into our algorithms, how are we doing that?
I think that the role of custom research changes a little bit from going out to do research to answer questions, to going out to do research to generate a dataset that's going to feed into the AI algorithms.
Nina Glynn
Yeah, yeah, absolutely.
So many considerations as we enter this sort of brave new world. You are totally right about this idea of how do we make sure that the data is fair, is relevant, is up to date? How are we feeding that? What are we feeding it with? We've got to be asking those big questions like who's responsible for AI's decision making? How much control can we give up? Yeah, I totally agree.
There's lots to be working through. And as you say, bringing the whole business along on the journey and doing that change management and making sure that everyone's working at the same pace.
Thank you very much for sharing all of those considerations.
Now, one of the things I'd like to address while we have the time is the impact on roadmaps and what this means for products in the future. We've got Juan here luckily to help us unpack that.
Juan, how are we prioritizing AI in our roadmap and what is the impact of agentic AI on future planning for products?
Juan Planillo
Yeah, I think that AI has been something that culture always has been at the forefront and we've been always looking at innovative ways to use AI.
I think that whenever the tools from OpenAI were deployed into the market, we were one of the first ones to test. We were in partnerships with Microsoft, with OpenAI, throughout all of this journey. And we are right now going into the next step in our roadmap, having AI front and center.
Whenever we are developing a roadmap that includes AI, you need to think about what the key pillars are or essentially what is going to be anchoring your roadmap.
For us, it's really about efficiencies. I think that everyone has said doing things faster is something that everyone is always after. We also right now more and more are looking at a roadmap that is an agentic roadmap that is autonomous.
We are seeing this element of AI doing things on our behalf even whenever we are not looking at it. This means agents that can operate independently, but still keeping that human in the loop that we are saying that is really, really important.
Then also there is the empathy element or the element around, we don't want to be building an AI that feels robotic. We want to build an AI that really takes into account the storytelling, the nuances and essentially help us deliver an output that is fully human.
Those are the core principles that we are looking at when building our solutions. In reality, these are taking shape as a network of agents.
Here you can see some of the agents that we are planning for next year. Some of them we are already working on. Essentially imagine an agent that can be doing that exercise of moderation, of following up with consumers in real time when the interaction is happening. Or an agent that is able to scrap all of that data that is all over the internet scattered identifying those signals or how we can keep even communities engaged.
Communities are a big part of what we do and we don't want to be losing those individuals that are actually experiencing our products. So, we want to build an agent that keeps the community happy, healthy and engaged at any point.
Essentially then we want someone crunching those numbers and present them to us in a sensible way for us to make decisions.
The result is, I believe what Seth was mentioning, something that is constantly going on without even us having to worry about. It's running autonomously in the background, helping us focus on actually those provocative thoughts that really will shape our businesses.
For us, agentic AI is not an upgrade or a next step, it is actually the foundation that we are going to be building everything from now on.
Nina Glynn
Amazing. Thank you so much Juan. Always on, always an early adopter. Thank you for sharing that.
So yeah, efficiency, autonomy and empathy, sort of the key pillars there. Thank you very much.
I'd love to come to you to sort of wrap up with a final question then, which is what would you advise for anyone who is thinking about building agentic AI into their roadmap? Any final thoughts or advice?
Juan Planillo
Yeah, so I think don’t focus on the tech. Really is focus on the problem or the ‘why’ of you are actually building something. I think that is the most important product lesson. Like always start with the ‘why.’ Once you have one ‘why’ keep asking even five times until you understand why you need it and why you have to focus your efforts.
I think that is really easy with all of the news that we have that all talk about agents, how you are missing out on the opportunity to try to get something delivered fast. But I think that you really need to understand what the pain points or the problems that you are trying to solve are. You need to see what the bottlenecks in the data are, the time things, that you are experiencing and then try to solutionize on that.
I think that element of building an agent that keeps the human in the loop, I think that is more and more important. We see all of the big players providing you with the reasoning of all of their agents so you can really understand what is happening in the background and really verify that what's happening is correct.
I think that whenever you are building any sort of agentic solution, I think that you need to keep human in the loop at any point just to make sure that you understand the reasoning process and that you understand that the tools that you are building are essentially solving the real problems.
Then finally, I think that I would start always small and then be prepared for iteration. You don't want to be solving all of your business with one agent, you want to focus on that specific area that you can make the biggest impact.
And then these agents don't really work very well on sandbox environments. Those environments that don't represent realities. So, I think that you need to deploy them fast, see what you learn and then constantly keep it training.
I would say that those are the sort of secrets for building a successful AI roadmap and really something that we are taking more and more into practice.
Don't be scared to fail. If you fail, yes, understand why and then try to develop something else and see if you can find a better route to solve the same problem.
Nina Glynn
Fabulous, thank you. What great words to finish on.
I love that we've kind of come back to that, always start with the problem and go from there. It's the kind of golden thread to all of this. Keeping that ‘why’ at the heart of things.
Thank you so much for those words of wisdom, Juan.
Big thank you to everyone. Thank you Jaisal, Seth and Juan for your time and insights. It's been very eye opening and elucidating as we enter this next phase.
And a very big thank you to the audience too. Your engagement and your time is very much appreciated. We hope you learned something new today. I certainly did, and that's that I need to Google how to pronounce ‘operationalize.’
A reminder to everyone listening today that we are going to be offering these one-to-one Q&A sessions with the team. If you want to learn more about agentic AI, you can chat with us at Bulbshare.
Get in touch by scanning this QR code or you can reach out at our website, which is bulbshare.com/demo and we'll be very happy to run through any questions that you have for you and your business.
Thank you everyone. Thank you very much. Have a fantastic rest of your day. And yeah, that's bye from me.