Electrifying the Factory: Agentic AI For Faster Clinical Trials with Pamela Tenaerts

The clinical trial system is starting to look like a motorway at a standstill: more and more drugs are lining up, but the lanes ahead are already full. We sit down with Pamela Tenart to unpack why the industry may have reached the limit of human-only clinical development, and why “adding more people” is no longer a realistic plan when experienced CRAs, site co-ordinators and trial operators are already stretched thin.

From there, we get practical about what agentic AI could change inside clinical trial operations. We talk about AI agents that can pull context across CTMS, EDC, safety reporting and TMF, helping monitors prepare for site visits, flag missing fields, propose next best actions, draft communications, and reduce the dead time where teams are simply waiting for updates. The point is not to automate judgement, but to automate the admin and co-ordination work so humans can focus on strategy, enrolment, quality, and patient safety with a clear audit trail and human-in-the-loop control.

We also zoom out to the bigger bottleneck: sponsors, CROs and sites are part of one connected system, so optimising one party can just shift the pain elsewhere. Along the way we explore why better, more sensitive endpoints can shorten trial timelines, how policy and regulation might need to evolve, and a memorable metaphor from industrial history: electricity boosted individual productivity first, but factories only truly improved once they were rebuilt for the new technology.

If you’re working in clinical research, drug development, biotech, or health tech, this conversation will help you think more clearly about where AI genuinely shortens the critical path and where it simply adds noise. Subscribe, share the episode with a colleague, and leave us a review. Which clinical trial workflow would you redesign first?

Visit Labtolives.com

Do you want to try the quiz from this episode? Find the link on our LinkedIn profile: https://www.linkedin.com/company/labtolives/ 

Would you like to join the show as a guest or collaborator? Find out how on our website: https://www.labtolives.com/ 

Transcript

SPEAKER_00 0:00

Welcome to another episode of Lab2Lives. Today we have a guest in the studio. We have Pamela Tenart with us and we are going to have be having an interesting improvised conversation. But before we kind of get into the meat of the episode, I have a check-in question for today's co-host with me, Dimitri and Pamela. I am curious, how did you sleep last night?

SPEAKER_01 0:23

Do you want me to go first? Yes, let's take you, Pamela. I slept in a hotel. I'm in Washington, DC, and I never sleep right. I can never get the ratio of the heat in the room and my blankets right. So I always have notifications

Sleep Check And Opening Mood

SPEAKER_01 0:37

that temperature is high or it's low. But yeah. So I could be better, but it wasn't my worst experience either.

SPEAKER_00 0:44

Hey, Dimitri, I know you have a you have a baby.

SPEAKER_02 0:47

I don't know. Today it's been problem under the weather a bit, but I slept wonderfully because I slept and then I slept extra in the morning and I slept extra in the afternoon. And in total, I slept like two hours during the daytime like I'm a year-old baby, and it was fantastic. Or maybe I slept poorly in the night, that's why I'm like that. I don't know.

SPEAKER_01 1:03

Naps are great. Yeah.

SPEAKER_00 1:05

Big fan. Well, for me, I had an interesting night's sleep in the sense that my 11-year-old daughter woke me up during the night because she had a nightmare. And this almost never happens. She's a solid, solid sleeper. So when she came to my room, I thought it was one of the cats. And she was having like a nightmare about bombs falling on our building and us having to evacuate. And like we spent a good hour afterwards kind of debriefing the situation in the world and how it was impacting her and how we were safe. But like that that was a long time since I've had to deal with a nightmare of this caliber. Yeah, that's it. And like in the world around us right now, like I understand how nightmares like this can happen. Absolutely, yeah. Yeah. Good. Well, uh on that very happy note about the state of the world. I would like to get a little bit closer to you, Pam, and we're going to go into our introduction round. So on this show, we we use the five key value streams in a pharmaceutical company to kind of place ourselves on where we are at individually, at least what is occupying us right now. And just to remind our listeners what these are, we have the innovation stream, we have the manufacturing stream, we have the compliance stream, we have the commercial stream, and then we kind of have the corporate stream. And together they form an end-to-end pharmaceutical company. And uh myself, right now I'm heavily focused on uh the innovation stream in the company that I work for, and how do we ensure that the drugs that we're working on actually work? And most

Mapping Roles Across Pharma Value Streams

SPEAKER_00 2:34

of my career I've spent uh in the RD space in life sciences, mostly on the technology and process side, but very interested in how all five work together to create a difference for our patients. So, Pam, perhaps I could ask you, where are you operating right now?

SPEAKER_01 2:49

Well, I would say for the last 15 years, I've been squarely into the innovation space. I've been in clinical trials for a long time, but right now that's where I'm at, like improving clinical trials. And it feels like we're at an inflection point, and I don't know if we're just getting better at describing the inflection point, because I feel like we've also been saying that for 15 years. Like it's different now. But I think this time it is different because of the capabilities we have with Agenc AI and other things to really make some improvements, if done responsibly and if done right.

SPEAKER_00 3:20

Very interesting. And Dimitri, where would you place yourself?

SPEAKER_02 3:23

I'm definitely also on an innovation spectrum close to that. I work mostly with startups, and so for me it's and and I work hands-on on research projects or on the IT infrastructure behind it. So it taps maybe a little bit in a sort of in a business value, uh just the outcome that we produce because we need to always think about how it brings us closer to the clinical trials, to investments, whatever the next thing we need to earn money from. But fundamentally it's it's very much on the innovation side for me.

SPEAKER_00 3:49

Yeah. Well, this will be a very innovation-focused uh episode then, since we're all kind of operating within that stream. Well, Pam, maybe you could tell us a little bit more about you mentioned that now we're at the inflection point where we can really make a difference with uh agentic AI for clinical trials. Would you care to elaborate on on what you're thinking of here?

SPEAKER_01 4:08

Yeah, so for the last 15 years, I think we've done a lot of work on improving the process of clinical trials, you know, better protocols, better, more patient access, making sure we can have enrollment. And I think we're getting better at a lot of those things, and we have evidence to show that decentralized clinical trials can improve access, can accelerate the speed of a clinical trial and all of those

Why Agentic AI Changes Trial Work

SPEAKER_01 4:33

things. And I can talk about that a little bit more. But what we've not looked at is sort of the process, the human sort of the human hurdles about getting improvement in that space. We're I've been in clinical trials on most every side, and I'm actually gonna start participating in clinical trial again next week, which I'm very excited. Yep. And I've not been at a regulator, but I've kind of been on most sides, and everybody is always waiting on somebody to get something done. Like if you're at a site, you're waiting on a contract, or you're waiting on a budget negotiation, or you're waiting on something else. If you're at the sponsor, you're also waiting on things. And I think with Agentix AI and agents, we've created these helpers that can help us make that space go away a little bit more because they can bring different systems together that would take a human, a lot of brain power to connect in their minds. You know, you have a CTMS system, you have an EDC system, you have adverse events. An agent could pull that all together. And if you were, for example, a c a monitor, a CRE in a clinical trial, they can help you prepare for site visits in ways that you couldn't do before. And what you can now focus on is the strategic part of that CRE monitor role. And I've been a CRE as well, so to kind of look at where the site is at, how it functions and acts compared to other sites, and sort of help the site with those things to maybe help with enrollment, have more uh strategic discussions rather than tactical about there's this thing missing here, because your agent already pre-visit sends a query to take care of that and those kind of things. So I think we're able to solve for a lot of the inefficiencies in the systems in a way we haven't before. And we're going to need that because there's truly a huge bench of new drugs and biologic waiting to be discovered. And right now, it's kind of like, you know, they're sitting at the races as on like horses at the races, and they're waiting for the gate to open, and there is no free gate. All the gates are taken with clinical trials that are going on, and we need to create new gates. And what we're going to be doing is we might not need more people because the more the people we have right now are going to be able to do more things. There's also been a workforce issue. So thinking we're going to be able to throw more people at it, and that's been our typical MO, right? More people. We wouldn't have enough enrollment, more sites. We need more, you know, we need to have more site visits because enrollment is going fast, more CRAs. We've always thought about it that way, and that's maybe not the best way of thinking about this.

SPEAKER_00 7:02

No, I I completely agree. And and also just there's a lack of deeply skilled people within clinical trials. So however we can augment them would be extremely helpful.

SPEAKER_01 7:10

We want people to work at the top of their licenses and not sort of doing things that, you know, an an an agent or a helper can do.

SPEAKER_00 7:19

I am curious, the way that you describe agentic AI, kind of combining a lot of workflows and pulling information for, again, the people that are are rare in in the clinical trial space. Do you see that across company borders? Or is it like within one sponsor? This can be helpful, or because I'm usually uh the way I have experienced clinical trials, we have the sponsor company, but then we have the CERO, we might have a different services provider, we might have like a whole host of different companies involved. And one of the challenges that at least I have seen is getting agentecai to work across company borders can be a little bit more complicated than just within the same company.

SPEAKER_01 8:00

I think we're focusing a little bit on both. We're definitely going to one company and saying this can enhance the way you do business, so to speak. The business being how you manage CRAs, how you manage filing of TMF documents into your TMF system. So there's definitely that aspect of it that can enhance one of the stakeholders. But we're gonna have to think about it more broadly because if one of the stakeholders is optimized and the other ones are not, that's then you just push the can down the road and now you push like you push the balloon and it pops up somewhere else, right? So we're also definitely seeing that uh if a sponsor is going to

Crossing Sponsor CRO And Site Borders

SPEAKER_01 8:38

optimize CRA management and they work with a CRO, they both need to be optimized to the same extent. And eventually, potentially the sites will need to be optimized because otherwise they're going to get a barrage of questions coming at them. There might be a world, and I don't know if this is realistic or not, but where the site has an agent that then automatically fills out the queries if it's a missing data field and they can connect into a system that says, actually, we have that data field, I can just give it to you, then maybe there's a world in which that could all happen. And Michelle Longmeyer, our CRO, has described this as the equivalent of a self-driving car where sort of things become like there's this automation layer that happens, and then there's these conductors that that sort of manage all of that. That's in my opinion a little ways away, but it it might be something that that could happen.

SPEAKER_00 9:31

I'm curious because this problem is one of the things where I would be a little bit skeptical before in applying technology, because if we don't fix our promises, then how can we connect these different dots and systems? Is that less of a problem now when we have smarter technology?

SPEAKER_01 9:47

So yes, we we often get people say things like we need to optimize our data first. But I think with some of the agents, they could could they can do that harmonization across systems. I think that's part of the beauty of it that they can. Now, human in the loop will remain a big important factor in a lot of this stuff at the moment. And there's, you know, the fully automated things at the moment might be things like there's a data field missing. Having a human look at that, there's still a data field, a data field. You know, it's kind of a black and white situation, whether a data field is missing. So those are, you know, we have systems where together with the company that we're working with, you can look at those and go, all right, this can be fully automated and none of this other stuff can, but it can give insights that then CRA still makes decisions on. So yeah.

SPEAKER_00 10:38

Great. I think we are ready for the dysfunction round. During this round, we ask our guests what is the key issue or dysfunction that is currently occupying your mind in life sciences. It can be related to what you're working with, it can also be something completely different. I'll take what you have.

SPEAKER_01 10:56

So I think it's a continuation of what we were just talking about, but this fact that we've reached the limit of human-only clinical development. I think we're not gonna be able to deliver what needs to be delivered based on the the prediscovery tsunami that's coming at us. And I'm not sure that the industry has fully

Data Harmonisation With Humans In Loop

SPEAKER_01 11:16

accepted that. This idea that to just, you know, add more humans to the system and that's going to solve it. I think we need to start accepting that that is not how we're going to solve for this.

SPEAKER_02 11:26

Humans like patients of humans like the people working working on the process.

SPEAKER_01 11:30

Good clarification. The people working on the process. Yes. The people who are trying to make all of this work and to conduct clinical trials. You know, we're going to add more sites when we don't have enough enrollment. We're going to add more CRAs when we need more site visits done faster. We're going to, you know, those kind of those kind of knee-jerk reactions that we often have as an industry, right?

SPEAKER_00 11:53

Yeah. That is does square with uh my general experience of how we solve problems in the clinical space. It is usually more people. And if we don't have enough people internally, then we outsource uh to get more people externally. That's exactly right.

SPEAKER_01 12:07

And I think that combined, there may have been a time where we even could find those extra people that we needed. I think right now it's a question of whether that workforce even exists for us to tap into, especially since the pandemic. There's been a numerous amount of work on how to develop workforce, how to get more people, be coordinators at sites, how to get more people, be part of the programs at the sponsor side. So I think we need to revisit this completely and think about how technology can help us in a way that maybe we haven't thought about before, or frankly, that wasn't possible before either.

SPEAKER_02 12:45

So what you're saying, Pamos, is that you think there's also gonna be a wave of new drugs, which is like much more coming from the drug discovery stage that that we will need to work with in the coming years?

SPEAKER_01 12:55

Yeah, exactly. I think there's been a huge amount of investment, both AI and not AI, on the discovery side, and there's a huge stable of new molecules and biologics that are ready to be put through the system, and we're not quite ready for that. And also, I think you mentioned you were an engineer, even on the engineering side, right? You can do a lot more with one engineer now than you could do in the past, if you think about it.

SPEAKER_00 13:21

I do know you have some thoughts about that, Dimitri. How much can you do with one engineer given the tools available?

SPEAKER_02 13:27

Well, I don't know. Nobody knows these days, right? You can produce more lines of code, right? I'm working in software, right? So you can produce more lines of code. Does it correspond necessarily to have better results and more results? That's where things get trickier. There is like some cases where you do get better. There are some studies that show that you actually no, you lose productivity. You think you gain it, but you actually lose it. But also some other studies show that no, no, actually, if you start using them well, you do get increasing productivity. That's an interesting topic, right, as an engineering. But I also think it's an interesting topic in the context of life science and drug discovery, right? Because now we are finally getting to the stage where we get all these big models. In a sense, if you wish, AI boom helped us having more GPUs, right? And these GPUs mean that now, well, now they're super expensive, but in principle they're also super powerful, which means that any company who needs a big computer to run any simulations or any any, I don't know, AI models to do anything on the drugs, today can buy a GPU which is literally as powerful as the best supercomputer five, seven years ago. And and that's that's kind of insane if you think about it.

SPEAKER_01 14:25

And I think you brought up a good point. There are going to be a lot of this needs to bear out and needs to have evidence around that as to how much is it actually improving? Can we have those metrics? And I think, you know, you kind of start off, and we did this with decentralized trials too. You start off with throwing the kitchen sink cana because you don't you kind of do like, oh, we'll do everything. And then you start figuring out that actually, if we want faster enrollment, we need to do these things. Actually, if we want, you know, maybe a more representative group of people that actually represent the people that will eventually take the medication, we need to do these things. And it's a sort of a more fit-for-purpose approach. And I think that will be the case here too. There's going to be things that AI cannot do and that we need to leave to humans. And there's definitely, as we learn about AI, and I don't know about you, but I've gone down many rabbit holes and uh like raised my head an hour or two later going like, oh my God, where did the time go? Right? Because we were trying to do something and kind of went further and further.

SPEAKER_00 15:24

I heard this joke the other day that it takes about uh 10 years to get a medicine uh from the lab uh to the market with a gentic AI. You can spend those nine years uh querying your AI.

SPEAKER_01 15:37

Well, that assumes that the AI is functioning at intersections that are part of the critical path. Yes. Some things are not gonna be part of the critical path, right? Yeah.

SPEAKER_00 15:47

Yeah. Yeah, absolutely. And I think like the challenge for industry right now is figuring out, well, yes, like there are all these things we could do, but which of the things are actually going to be useful in in shortening the timelines?

SPEAKER_01 15:58

Exactly. And we have a lot of these discussions about what will change the timeline of a clinical trial or you know, the trajectory of that 10 years. Like what is the thing that's going to make it shorter?

SPEAKER_02 16:10

Do you think it's gonna get shorter, by the way, or is it gonna be one of these cases where we're gonna spend as much time but just getting better results now? Uh because now we have better tools and we can put more effort in in every step. So but but then like the mental capacity is gonna stay old and instead of you know, I heard it with AI that this is the experience people often have now that they have AI and they can let's say develop 50% faster, they don't leave work at 5 30. After five hours of work, they stay all the eight hours, but now they just have to chart more work and more productively. And I think my question here is do you think with clinical trials are we gonna see shortening of the clinical trials or we're gonna see increase of requirements and demands and maybe even paperwork needed to produce, but also higher quality, maybe results, or are we gonna see them shorter? So like are they gonna get five years, or are they gonna stay ten years, but demand a higher quality?

SPEAKER_01 16:57

I actually think they're going to get shorter. And one of the reasons I say that is a lot of different things are going on at the same time. We will have less empty space within a clinical trial where we're waiting on things, but there's other stuff going on too. There's a big push for better endpoints, more sensitive endpoints. And for example, what that means is that you will see a meaningful difference in a clinical aspect faster because what you're trying to show in an endpoint is that you're making an improvement in how the patient feels, performs, or functions.

The Human-Only Model Is Maxed Out

SPEAKER_01 17:29

That's the goal of a drug, right? There's an improvement. And that improvement is based on an endpoint. And it could be that now you have like a six-minute walk test. That if the six-minute walk test gets better, you know, that's your endpoint. And that six-minute walk test only has a certain sensitivity for the disease that you're studying. But if you could add other things to that six-minute walk test, you actually get an answer quicker and you may need less patience. So it's not only like you're picking at it from all sides. If you need less patience, you don't need as long enrollment because you can do it faster. And those other patients can participate in other trials. So I think the throughput will become much bigger. But it's going to take a concerted effort to put this all together. And it's it's gonna be a lot of those things happening at the same time.

SPEAKER_00 18:14

I just want to tag on to the thing they said earlier about more drugs coming into the system. This is one of the things that has been occupying me lately, is both like we have these this new kind of company that's scours well what is actually possible using AI, generating more molecules, but also we have this trend of finding additional applications for uh existing molecules. And I I I think you're right. I think there is like a deluge of new drugs who are going to be moving through the system, generated faster than we could before, at least on the conceptual stage. And I I think they're going to create both like shortages of people, but also shortages of like people who have to review all the science uh with the regulatory authorities and generally like strain our system because we've got that part right, but we haven't figured out well, how do we follow this deluge downstream? And I think that's a really good point.

SPEAKER_01 19:04

Because if you look at the last 10 years or so, 15 years, the number of drugs, new new drugs approved by the FDA has kind of hovered between 45 and 55, say for like 10 years. China just had 78 drugs approved last year compared to I think 46 in and I may be wrong on that, but somewhere between 45 and 50 in the United States. So if we want to get a better throughput, all these systems, like all the all the people that are part of our ecosystem are going to have to improve. And I know the FDA is already using AI in some cases. I don't think they're actually doing it for the review part yet, but there's going to have to be big changes. And I actually think there needs to be a national policy around clinical research to, in the United States at least. I was at the National Academies not too long ago, and they were I mean, if we have a good national strategy of how we want to deal with this, I think that also would be helpful. Like the policy side needs to change, and then the all the people that have a part of this also need to adapt. Because again, if you only improve one part, it gets stuck at the next, next thing, right? Yeah.

SPEAKER_02 20:15

That's actually what I wanted to follow up on Pel. How do you think it's gonna manifest? So, like, does it mean that now all the startups from Boston, I don't know, gonna start coming with the new drugs, gonna come to FDA with it, or like it to clinical trials or to wherever, and then just wait for two years until their paperwork is done, or for three years until their paperwork is like how is it gonna manifest this this flooding in in new drug drugs that are coming to the pipeline that's kind of not there to overheat?

SPEAKER_01 20:40

I mean, if you think about companies having a certain number of people and a certain number of capacity of doing trials right now, they're mostly at capacity now, and there are hundreds of drugs waiting to get to become part of that system, and we just don't have the intake capabilities of doing that. There just isn't a lane open for them to swim in right now. And I think what we need to do is free up the people we have so that they can less CREs can work on one trial, for example, so that the ones that were working on the same trial can now work on another trial. I think that's how you create capacity within the constraints that you have. Because the workforce, we still also have to train new people, but that's a long and arduous process too. But then we also need people that then can review the trials and we need sites that can be put that can then function in those trials, and there's a big push about going like putting trials in clinical care, and if we can so there's a lot of things that need to change.

SPEAKER_00 21:40

I think it's an interesting like visual that like we we have been able to apply AI in the early stages, and like we have sped up that part, or at least like increased the volume. But then we kind of have a block in the next phase.

SPEAKER_01 21:53

And this and the investment of AI is hugely different, has been hugely different. The investment on AI on the pre Clinical side has been tenfold the investment of AI, at least on the development side. And I think that is where we're at right now, that we need to increase that investment of AI on the development portion, on the conduct of the trial, so that we can create a system that can handle all these new molecules that have been invented because the AI discovery almost put put on the discovery side, looking for, you know, receptors and drug targets, and then make the drug targets for those, make the drugs for those targets, that now we need the AI and investment to come into the conduct of the clinical trials. I think you're right. It's a good visual. It's like a big balloon here and then like a skinny string over here.

SPEAKER_02 22:43

That's a good one.

SPEAKER_00 22:43

And and it's it's interest but there are two different problems too. Like in early stage research, it's it's more mathematical. You can and and perhaps it lends itself easier to some of the activities that generative AI can can do very well and that is like recognize patterns, is like find adja adjacent meanings. Whereas when we get to the clinical stage, w it becomes more human uh-centric. Not that generative AI can't be supportive there, but it's a different kind of problem that requires more interaction, perhaps.

SPEAKER_01 23:13

Maybe, but if you think about agents creating those insights that you just mentioned, then that is the same problem, it's just applied somewhere else. So if you think about creating insights and doing those kinds of activities, there's still in a clinical trial, you live by dashboards, right? Yeah. You live by how fast are we enrolling, which side ha where are the adverse events, how fast are they happening? Still, that is all insights if you think about it, right? Yeah. So there is a way that you can apply that into have the agents work on those things because you live by insights in the data. You know, we we were part of, I was part of a clinical trial at one point before computers and all that. And we knew that there was an increased bleed in one of the arms. We had to call all the sites and shut them down. Like that insight was probably could be found earlier now because of connected systems and connected data flows that back then we didn't have. And we probably could have avoided people being exposed to that arm sooner. And then sort of so I think if you think about it that way, clinical trials are all about insights and how you react to them. That is an interesting idea. I'm gonna write this down.

SPEAKER_02 24:24

I'm very curious here, in a sense, in a historical perspective, because you already mentioned it, Pam, that we already have hundreds of drugs waiting for clinical trials. And then if you believe in a free hand of market, well then the companies that are assisting in clinical trials, all all the participants of the market, should have an incentive because all the big, rich pharma companies come and say, please test our drug. We're willing to pay double the amount, triple the amount, just take us in a queue. So it seems feels like there should be a strong monetary incentive for more scaling already, even before AI, there should have been an incentive for bigger scaling of clinical trials, making things more efficient. And also technology kind of was there, right? You kind of touched on it. I didn't

Shorter Trials Through Better Endpoints

SPEAKER_02 25:01

think about it. But true, there was a world where computers were not existing, and we still did clinical trials back then. And in a sense, it's just a big coordination project of many sites and many people involved, which feels like something computers, even the classical computers, pre-AI, should have already solved. And so I think my question is here what's the historical perspective, in in your opinion here? Why do you think we have been unable so far to scale the clinical trials if the technology was already quite advanced? And also there's a strong monetary incentive for that.

SPEAKER_01 25:31

So I think, and uh just to be clear, I'm on the science side of things. I'm not I have an MBA, but I'm not really on the business side of things. But for the investment to work, clinical trials are already so expensive. I don't think it would work for trials to be three times as expensive to be able like, you know, that I don't think that works for at the end of the time frame to then be able to have drugs that you can sell at a reasonable price in most cases, or for a company to be bought. I think people still need to get the clinical trial cost under control. And this inflection point, like for the past 15 years and longer probably, I've only been working in it for 15 years, people have been trying to improve things and things are better than they have been. I think we also need the data handling capability of computers now and servers now that we didn't have 10 years ago, we didn't have five years ago, that I think that are required to make these problems handlable by the systems that we have. We did a lot already, but I think there has been improvements and trials are better than they were so many years ago. But I think we're now at a true inflection point, even though we've been saying for so many years that it's changing with the agents that are coming in.

SPEAKER_02 26:44

Quite interesting because I feel that clinical trials should be just a classical sort of database problem, right? I'm a software engineer, so I go into directly in solution mode maybe. But like it's basically a database where you put data and then you need to query it and and it's something that feels like we've had for like 20 years already.

SPEAKER_01 26:58

I think the pr the you're right, it is a data issue, but it's how the data gets into that system that is still a human pro that's still a human thing. It's coordinators at the site finding patients to be part of their trials or patients finding trials. So that is still a human interactive kind of thing. So there's a lot of human pieces to this still that are still ongoing to be happening. It's still a human looking at the data of a site and preparing for a data monitoring visit. And now you're right. Now we can turn that over to a helper in the form of an agent, but there's still a lot of human components. This is a human-heavy industry, and I think over 90% of dollars of this industry goes to humans, like to somebody doing something.

SPEAKER_00 27:44

That makes sense. And so that that's also why perhaps now we are at that inflection point. Maybe if we can actually make those humans more effective, then maybe.

SPEAKER_02 27:59

Yes. I think what what I'm hearing you saying, Pam, is that you know, we have I I need to pick a regulated industry here, and I'll go with construction because construction is quite regulated. You cannot just build whatever you dreamed of.

SPEAKER_01 28:11

I was gonna say it isn't really, but you're right. Okay.

SPEAKER_02 28:13

You need an engineer looking at it, you need somebody to approve your architecture, you need to approve your drawings. And kind of what we're experiencing now here is that suddenly everyone can go to Gemini, not sponsored, and say, generate me a 3D visual of my house, and then like much more people come with an idea of how they want to build their house, and they all go to this, what would it be, architecture agencies to say, hey, can you please approve that? And then they also bring this to, I don't know, government regulators who also need to approve this, and it would flood them, right? In the same way. If if suddenly every single householder in Denmark decided, I want to rebuild my house, here is here's a Gemini generated project, it would completely overwhelm the industry of the current way architects are doing it and the current uh regulators are doing it. And in the same way, a similar thing is happening with drugs here. It's not individuals who design drugs, although this big story about the guy who designed a Marnay vaccine for his uh dog's cancer, I'm pretty sure you've seen it. But it's still companies using AI to increase the amount of things to generate that still hits the same old way approach to how we approve things and actually design them, and actually there's a lot of humans involved into building all these things.

SPEAKER_01 29:21

There's an interesting analogy I saw last week and I was talking about it to some of my friends. So when electricity came, think about when electricity came. So electricity was invented in 1898 or something, like late 1800s. So that was like fabulous, right? Because now you could make machines do things. So there was individual productivity gains at the time electricity was invented. The factories did not get better until the the factories themselves were rebuilt. Because the factory, unless they had an electric machine that used electricity instead of what they had, there was only so much electricity improvement you could be had. And if you think that analogy through, there's a little bit of that going on right now, too, because you're sort of tacking new systems onto old ways of doing things. And so for this to work better, eventually, the whole system will need to be redesigned because we're now working in a system that was designed for humans and a system that was designed for AI. So I think we're gonna have to re-rework the workflow, like redesign what it means to be a CRA coordinator. We may not need anybody to file documents anymore. We may need other people to do other things. So we're gonna have to. This is just the beginning of a revolution, let's call it. And I think we're there too. But you you bring up a good point. There is that individual productivity will move and that will make trials go faster. But for trials to truly go faster, we'll need to redesign the system itself as well.

SPEAKER_02 30:53

I really like this uh phrase that the factories had to be rebuilt before they would Yeah.

SPEAKER_01 30:57

Yeah, because you know, if you don't have an electric machine that does something and you still have a steam whatever thing, then electricity, okay. People can stay longer in your building because there's light, maybe in the building, but you're still constrained to the system you have.

SPEAKER_02 31:13

And you can report that you've done a very important big strategic project on uh full-scale implementation of electricity in the factory you're working at, but it just doesn't bring you where these new modern factories that are using electricity from the get-go are getting.

SPEAKER_01 31:27

And I think that's actually one of the innovations Tesla did. They rebuild a a factory differently than any other factory, and their throughput is much bigger, if I mean than other factories. So it's that kind of stuff that we need.

SPEAKER_02 31:41

I feel it also opens a whole can of worms of that maybe the entire process of clinical trials needs to be rethought because it was also built in a different era where there was just like that many companies and that many professionals who could audit that, and the people working in the companies maybe also even had just a lower education. I don't know how it was historically 50, 70 years ago, but I can imagine today there's like so many people who have busters PhD, who work everywhere, who understand how science works, who understand like all these things. We just like talent-wise, also we're in a different world. Everything is a different world. So it feels also maybe the clinical trials themselves is sort of sort of a factory, if you please, because it's a factory

Rebuilding The Clinical Trial Factory

SPEAKER_02 32:18

that produces drugs just in a global scale.

SPEAKER_00 32:20

Yeah, and we we might have to ask ourselves so if we were to rebuild this process from scratch, how do we show that this drug is uh beneficial and safe in humans? And and if we were to rethink that from scratch, what would that actually look like? Would we have the same amount of paperwork? Would that paperwork look differently if we could potentially pull it directly from the data sources? I'm just like at awe with the metaphor that you brought here because my my wheels are turning. That this is we might actually be able to do that.

SPEAKER_01 32:47

Somebody's pieces are already being rebuilt because, for example, when you use our agent, the agent that we have built, it can grab documents out of an email automatically, it can file them for you. So there's already a whole restructuring of certain processes happening, and I think that is what you're gonna be seeing. Maybe the the the difference between the metaphor of the factory and clinical trials is that individual processes can definitely be adapted much quicker than others, and so it's not quite you need a new whole machine. You need the parts of the machines are getting adapted and improved by the use of AI, and eventually the machine will be different. Yeah.

SPEAKER_02 33:24

We can also stretch it the other way around and be more brave. And there I think it could be something like you know, we used to send reports and summaries because that was necessary, right? Because what could a single FDA employee in 1960s do if he got like the raw data of the clinical trials? Not so much, probably. He didn't likely or she didn't even like they had a computer at their disposal. Yet today you could potentially just upload through some sort of API every single row of data from the entire database and deploy agents on the FDA side. And then these agents on the FDA side just transfer the data in a standard ways, in a maybe more creative ways, and they try to sort of, in a sense, red teaming, try to interrogate this data and figure out, oh, is this a really good result from a clinical trial? And then it would be completely the other way around where we make it everything completely electronic and integrated.

SPEAKER_01 34:11

Yeah. I mean, there's a lot of ways this can go. And I think the difference maybe too between a factory, even though I like the analogy, and this is that there's still lives at stake here, right? People are volunteers in a clinical trial. So you can't really just wait for the raw data to be collected and then shipped over to the FDA. Somebody still needs to look at adverse events, and somebody still needs to look at if those adverse events happen, how are we going to let the participants know so they can make an informed decision about staying into the trial or not? So I think there's some ways that the analogy doesn't work, but I think generally speaking, there's parts of it that I definitely like. Yeah.

SPEAKER_00 34:48

But I'm going back to a thing you said earlier, Pamela, that uh clinical trial is insights and how you react to them. And and that does sound like a very much a problem that can be solved with AI agents.

SPEAKER_01 34:59

Yeah, because that's the other thing. The agents they draw insights from the data they see and then they propose next best actions. And so the agents propose an action and the CRA in a chat interface can go like, oh, that's a good action. Yeah, let's do that. I want to create an email and say that this data field is missing. Okay, do that. And then they file it away in the TMF. So there is the next best that the agents don't just do the insights, they also provide the next best action that a human can then decide. Yes, do it, don't do it. But there's a lot less physical

Safety Limits And Regulatory Possibilities

SPEAKER_01 35:31

paper, like if you want to think of the equivalent as paperwork going around because it's all electronic and coming out of the systems directly.

SPEAKER_00 35:38

I wonder what the clinical trial factory of the future would actually look like. Hopefully we'll see it in five years, right? Yes.

SPEAKER_01 35:46

We don't have to wait for the 30 years it took to build to build a new factory.

SPEAKER_00 35:51

And like back to another thing you said about the policies, there was a big push, at least in some countries, to industrialize, but also like electrify. That was like a government investment and policy saying we want to be a country that is fully connected on the electrical grid. That was like one of the things that the Soviet Union had going for it originally, like electricity to the farthest corners. And that did change things and brought a lot of prosperity. And there are no go ahead.

SPEAKER_02 36:16

I just want to chip in and say that I think that I heard a lot of talks about this that AI should become a commodity. I mean, it's maybe a little bit biased because it's normally pushed by the people who work in AI companies and look for investors and for money. But I see the point, right? If you basically have compute as a commodity, that's that that could change the world completely.

SPEAKER_01 36:33

And to your point of regulations and policy needing to be in effect, there's clearly more policy than there ever was on AIs in drug development. Mostly the policies have focused on things that are non-operational. So a lot of what these, if you think about the operations in clinical trials, a lot of the operational is excluded from the AI guidance, for example, from the FDA. And it's when patient safety, adverse events, when it gets data, when it gets to those things, is where, you know, the AI guidance comes into play a little bit more.

SPEAKER_00 37:05

Very cool. I believe this concludes the idea distillation round for now. I love the metaphors that we found and how we kind of tweak them a little bit. I would like us to continue into the game show round. Oh no, I'm scared. So every time on the show we play a little game show, and it's gonna work as follows. This time around, I'm going to ask uh the two of you a question about a drug development cycle. And I'm going to ask you a question and then I'm going to give you some options. And there will be three questions in total, and one of you is going to win the honor of having responded to most correctly. Good. We are going to get into this. The first question that I would like to ask you is which immunology drug

Game Show On Trial Timelines

SPEAKER_00 37:56

was originally started in 1993 by BaseF and the clinical trial started in 1997, and then the FDA approved them in 2002 and the EMA in 2004. So 97 to 2002. Okay, that's bad. So the you can say the total lab to lives duration was nine years. And I'm going to give you options. Was it Ozempic? Was it Humira? Or was it Skyrizy? Humira.

SPEAKER_02 38:26

I know it's not Ozempic because I don't think it was approved in 2004, at least uh to my knowledge. So I will go with the Skyrizy, I think you said, Ivana?

SPEAKER_00 38:36

Yes, I said Skyrizy. And the correct answer is Humira. It's older. That's the only it's an older one. Skyrizy is newer, yeah. And I think the interesting part for me about this drug development journey was uh and what I didn't know before researching is this molecule actually changed ownership three times during its development cycle. Wow. I had no idea. I I always thought it was Abbott Laboratories that had brought it to you said it started with BS B ASF? Yes. Like the the chemical company. Yeah. Okay. I I I I was uh and then it was bought by Knoll pharmaceuticals. Yeah, wow, that's a name I haven't heard in a while.

unknown 39:17

Yeah.

SPEAKER_00 39:18

And then Abbott acquired it. So yeah. Very interesting.

SPEAKER_02 39:24

What does it treat again?

SPEAKER_00 39:25

Well, it treats a whole host of uh autoimmune disorders, such as Crohn's disease, psoriasis, the one with the joint arthritis. Arthritis. Arthritis. There we go. Thank you. Very good. And Skyrizi is actually like the the next evolution of an immunology drug for now, what is called MV.

SPEAKER_02 39:43

Must say it's not an impressive it went for n in nine years. It's also impressive because it went for nine years through three different management teams. Whenever the acquisition happened, the new team was like, so what do we do about it? And it took us a couple years to figure out just that.

SPEAKER_00 39:56

Exactly.

SPEAKER_01 39:56

That's double impressive, right? Yeah.

SPEAKER_00 39:58

The fun part is in fact, like when I was researching for this quiz, a lot of drugs that started their development in the nineties, they took exactly nine years from research until if really rule. I was like, this is like this is interesting. Not not ten, not eight, nine.

SPEAKER_02 40:15

Well, it's nineties, right?

SPEAKER_00 40:17

Yeah, it was the nineties. Very good. Now, next question. Which of the following trials is still running? Meaning I'm going to give you the start of the trial, but it is still running. A single trial. A single trial. So first option, toxin for involuntary movement disorders that was started in November 1985. Or the implanted stimulator for restoring hand functioning in spinal and cord injury, which was started in April 1989. Or the multifactoral intervention for type 2 diabetes, which was started in January 1992. Which of these trials from the 90s and 90s are still ongoing?

SPEAKER_02 41:02

We go with the spinal cord. I feel it must be pretty invasive treatment, maybe, and that's why they struggle to recruit patients, and that's why it takes them so long. And maybe it's someone discontinued by now or something, they just forgot. It feels very much like they, you know, just forgot, submit the click cancellation button, you know, click cancel, and then you're like, I'm done. Then you come back to the computer 20 minutes later, and prend it. All the time there was a screen saying, Do you really want to cancel yes or no? Which you just didn't know about.

SPEAKER_01 41:27

That's that's really clever. I'm gonna go with the diabetes one. Because if you say multifactorial, I feel like it's more of an observational kind of a thing, and they can can keep continue collecting data. That sounds more logical in my head, but who knows?

SPEAKER_00 41:42

Well, the answer is all three of them are still running. What?

SPEAKER_02 41:48

That's impressive.

SPEAKER_01 41:50

That is I don't know if actually that may be sad.

SPEAKER_02 41:55

One of them is 31 year old by now, bother uh uh botulism toxin.

SPEAKER_01 42:00

Or have they given any results in the meantime, or are they still sort of waiting on results? If they've been giving results, then I can see it.

SPEAKER_00 42:07

They have been giving results and uh they just uh continue MSing results. And especially the I I was curious about the the implanted stimulator. Yeah, that feels because it is it is still enrolling patients and and patients just they enroll and then they just stay in the trial forever since uh 1989.

SPEAKER_01 42:29

I'm I'm uh it's kind of like the women's study, you know, where you need follow people for like 50 years to figure out what what happens to them. Yeah. This is an endpoint problem. Yes. Yeah.

SPEAKER_00 42:43

Yeah. And one of them, actually, the the last one has changed centers. It has changed location, but the study kind of just keeps us going. That's the one that started in 92. Very good. Now, the next question. Glebec, which is a drug now owned by Novartis, is an oncology drug, and it received record regulatory approval by the FDA. So for this record regulatory approval, how long did the clinical trial phase last? And I'll give you options. One month, five years, or two years. When you say phase, do you mean like phase three? This one was actually approved on a phase two. Yeah, back then.

SPEAKER_01 43:20

Yeah. So how what were the options?

SPEAKER_00 43:22

One month, five years, or two years?

SPEAKER_02 43:24

I don't believe in one month. I don't know, like it must be a miracle drug to get approved one month. I just don't feel the oncology would get such a treatment. But two years I find believable. So I'll go with two years.

SPEAKER_00 43:33

Same. The answer is, in fact, two years. But interestingly, this is actually the shortest clinical trial phase from initiation of clinical trial until FDA approval that I have been able to find. The results were overwhelming. That was an amazing drug.

SPEAKER_01 43:48

That is an amazing drug.

SPEAKER_00 43:49

Yes, so that was our little game show round of uh duration of clinical trials. Well done, both of you. I believe it's actually a time.

SPEAKER_02 43:57

I think Pamela One because the first question I missed and And next to we both got so thank you.

SPEAKER_01 44:04

Those are fun things though. I love this kind of stuff.

SPEAKER_00 44:08

I I love the research for this. Okay, well, uh, we're going to start rounding off. And as we round off, I would be curious, based on our uh conversation, what was a thing that you're going to take away and that might make you think a little bit extra about the state of our industry the next couple of days.

SPEAKER_01 44:25

I would say the discussion about the factory and sort of like what portions do we really need to rethink faster than sooner.

SPEAKER_00 44:32

I think I am going to take away this workforce problem that we have. And for me, our conversation really led me to think about how it's it's not about losing tasks in people's job subscriptions. People's jobs are going to just be completely different once

Key Takeaways And Where To Connect

SPEAKER_00 44:49

the electrified factory comes into play. And that we are kind of sub-optimizing right now where we should be optimizing for the whole.

SPEAKER_02 45:37

And if you just open a tap, it's gonna spill all everywhere. And it's very interesting for me personally because I work a lot with startups myself. So this is an interesting problem because many companies, of course, when you're a startup doing design uh drug design and stuff, right? You probably think about, oh, how do we get it to the market? And you expect the process to kind of work like it used to work. And if suddenly it gets completely clogged, that's a whole new sort of threat vector for this company's existence that I never thought about. So that was actually my main takeaway from today.

SPEAKER_00 46:06

That is a good metaphor. Very good. Well, where can our listeners find more information about you or reach out with any follow-up questions?

SPEAKER_01 46:13

Well, I'm on LinkedIn, so that's probably the easiest way, and it's just my first name and my last name at LinkedIn, and I don't think there's plenty there's more than one of me.

SPEAKER_00 46:22

Isn't that wonderful? Having a unique name?

SPEAKER_02 46:25

On that note, can I tell a quick story?

SPEAKER_00 46:27

Yes. Tell us a story.

SPEAKER_02 46:28

I have a daughter, Augusta, very young, a year and a half. And apparently Augusta is a very rare name in Denmark, because in Denmark you can go online on a database and see how many people have your name. And there's like 18 Augustas in the entire Denmark, at least according to this database. And it just so happened that our dentist had exactly two Augustas book after one another last week for an appointment. Because they came out called Augusta, they were like, Yeah, yeah, she's coming, and then they came back, like, no, we're looking for an 18-year-old Augusta.

SPEAKER_00 46:57

That is insane.

SPEAKER_02 46:58

So speaking of repeating names, you never know. You never know, I guess.

SPEAKER_00 47:03

This is true. Yeah, this is true. Never say never. Well, I I would just uh add on to that that uh my daughter's name is Kama, which is a pretty rare name, and mostly uh people who are 85 or older in Denmark have that name. But she did attend a summer school where there were four Kamas. So they had to like put them in different groups to not get confused. Name fluctuate.

SPEAKER_02 47:24

Crazy.

SPEAKER_00 47:25

Great. Well, this is the end of this episode. Pamwa, thank you so much for joining us. Absolute pleasure. Tima, always a pleasure. And Elena. We'll be looking forward to sharing this episode with our audience.

SPEAKER_01 47:36

Thank you so much. This was very fun.