AI for Finance Leaders: Emerging Patterns to Implement and Utilize AI

Jeff Jahnke: Good morning, good afternoon, everyone. Welcome to Condor's AI for Finance Leaders and Emerging Patterns for Implementing and Utilizing AI webinar. Really awesome to see everyone. Got a huge crowd today, so really excited to show you a little bit of information about AI in general, but also what we've been working on here at Condor.

‍

A few weeks ago, we presented this panel at ABFO, and the session was incredible. It was standing room only, and that surprised even the organizers. And what that kind of signaled to us is that finance leaders in this industry know that AI is coming, and they're done waiting on someone else to tell them how to use it.

‍

And so what we're hoping is, you'll walk away from this webinar with a bit of a blueprint for how you can implement and use AI in general in your workflows, but then also how Condor is leveraging AI to revolutionize clinical finance for clinical trials and R&D. So just a couple things that, that we wanted to talk about that you'll get from this presentation today.

‍

One is the framework that Jennifer presented at ABFO with a roadmap that holds up under audit pressure. A live demo of an internal tool that we have developed here at Condor to help streamline implementations, which I'm really excited to show you. And then a glimpse into our commercial product and where AI agents are going inside that product to help you make better decisions for your portfolios and clinical trials.

‍

So with that, maybe we can kick off with a few introductions. Today we're joined by Jennifer Kyle, founder and CEO of Condor, and also by myself. I'll be hosting. I'm Jeff Yahnke. I'm the director of customer operations. So really excited to be with you all here today. Jennifer, did you wanna give a little bit of your background?

‍

Jennifer Kyle: Sure. And for those of you guys who haven't met me you can also call me Jen, but Jennifer's super formal on here. Jen Kyle founder, CEO of Condor. Prior to Condor, if you guys who don't know I was an auditor at Ernst & Young, where I audited many different industries, among them biotech, and really had a hard time wrapping my head around clinical accruals back in the day pre-COVID back in the 2010s.

‍

And then ended up not staying down my audit path worked in the industry, worked in finance, accounting, operations, and got called to a biotech company to go solve this really crazy problem around clinical accruals and forecasting back in 2017, 2018, and that really started Condor's journey. And so now I'm really excited to share a little bit more about how we saw this future of clinical finance, R&D finance, and how we've really matured, how the industry has matured, and how our technology has matured along with the industry, and what's possible today.

‍

I think what's really exciting and scary at the same time, we'll go through these slides, is we're moving so fast, it's the levels of efficiency that now we can leverage are unparalleled to an- anything that we've ever seen in the industry. And we'll go through as Jeff said a framework of how you move th- through the industry, what's happening today in AI, where AI is going, and how can we be better prepared to really utilize it across our organization and life in general?

‍

Jeff Jahnke: Yeah. And my name, as I said, Jeff Yonke, Director of Customer Operations. Yeah, my background, I've been here with Condor since twenty twenty and prior to that had five years working for a large CRO, so very familiar with clinical finance and kind of the structure of how what levers you can pull, I should say, for analyzing clinical trials and the underlying finance, and really excited to showcase what we've built here today with you all.

‍

Maybe we can go into a little bit about the history of Condor too, for those who might not be familiar with our story. So Condor is an intelligent finance layer that unifies clinical, operational, and finance data into a single source of truth. We manage twenty billion dollars plus of spend on our platform today and have, a, just a whole lot of amazing customers that we work with.

‍

Some of the metrics that we like to say is we enable ninety percent forecast accuracy. We enable a much faster month-end close and up to thirty percent budget savings on change orders. A little bit about our journey too. Jen gave a little peek into her background. She's an ex-EY auditor and saw a lot of the overlapping problems firsthand, both as an auditor and then as a consultant and interim controller for large biopharma companies here in San Diego.

‍

So in twenty twenty decided to take everything that, that she had learned and seen in the industry and found Condor to tackle those problems. In twenty twenty-three we pioneered our knowledge graph, which is built on clinical and financial ontology. For those of you who might not know what a knowledge graph or an ontology is basically it's a web of facts that are all interconnected.

‍

So you can think about it as how a person learns a skill and understands that skill intimately. That's essentially what the knowledge graph is here at Condor, is we have an AI layer that understands and deeply is able to interpret- Very similar to a clinical finance or a clinical accounting expert.

‍

In 2025, we rolled out our enhanced forecasting scenario planning suite. And then very recently, in early 2026 on the, as we have w- worked through all of that journey Insight Partners took notice of all of the hard work we were doing and really excited about where we saw the industry going and what our vision for the next generation of clinical finance is, and chose to invest with us for a $24 million Series A to expand our financial intelligence layer capabilities.

‍

And our vision here is a world where no therapy is delayed by financial uncertainty. In other words, we wanna make sure that finance is not a barrier to getting therapies in the hands of the people who need them most. And with that, I'll kick over to Jen where she'll give a little recap of what we heard at ABFO and how that kind of dovetails into the broader presentation today.

‍

Jennifer Kyle: Thanks, Jeff.

‍

And before we dive in, one thing that I wanna mention, 'cause we see here that in 2023 we pioneered the knowledge graph. It took us quite a few years to build this, and I would say I didn't realize we were building a knowledge graph until today, and there was a little bit of luck and timing where we kept saying we have a lot of deep domain expertise that's coded in our software, but that's as much as I could articulate it.

‍

And what really excited insight partners and the industry is that actually positioned us that when large language models came into the mainframe of the market that we see today, bolstering Condor into that was actually really deep proprietary information that we possessed at the intersection of building out this web this knowledge graph web.

‍

And we'll go into a little bit about what that means. Which, really it was years of deep R&D and product development that we're now able to take to, to capitalize on and optimize and that wouldn't be possible if we weren't where we are today with AI. So a little bit of luck and timing kinda hitting the market where it is and taking this to the next level.

‍

So with that, for those of you guys that don't know ABFO, it's an incredible organization. It's the Association of Bioscience Financial Officers. If you want to network with other peers in the industry and you just wanna hear what everybody else is doing and make friends and have a really good time, it's a great organization to to be part of. And I'm not getting paid to market for them right now.

‍

I truly just have a lot of fun when I go to these organizations, and you get CPE credits. Reason why that's important, as Jeff said, we did this session at the ABFO event a few weeks ago, and it was a hit. And so bringing what we learned from that that event and that organization to you guys, we heard a few different things.

‍

Number one, everybody knows they need AI. I can't tell you how many conversations I'm having where they're like, "We need to implement AI. What do we implement? What's the difference between Copilot and GPT and Claude? What's better? And how do we keep it protected? How do we keep it secure? What's the first use case? Like how do we use it?" Like we all know it's here. We are just really busy in our day-to-day jobs, and we're not sure where to start.

‍

And I don't know if any of you guys feel like that, but I know that there was a one point in time when I certainly felt like that too. And then the second thing that we heard is, you hear all this talk of AI is gonna wipe out 50% of the, white-collar jobs. It's gonna wipe out management. It's gonna wipe out all of these jobs.

‍

And new college grads, what are they gonna do? 'Cause they're the ones that are doing all of the grunt work, and now if AI's automating all of the grunt work, what do we do with those new college grads, right? And so there's a lot of uncertainty around what does it mean for me, my job, and just my career long term.

‍

And then the other thing, the third, was a lot of the conversations were wanting to get some insights on us, with us. Number one, how we're using AI irrelevant of R&D finance, but just how are we doing it. And then number two, how are... what are we doing for AI and R&D finance, and how should companies be thinking about elevating that part of their organization into the new ways of working?

‍

So I'm gonna show you a funny video here. This was the Wall Street Journal, and this is what happens when Claude created a vending machine. So hopefully you guys can all hear my sound. It's a short video.

[Video clip: Wall Street Journal coverage of Anthropic's Claudius vending machine experiment.]

‍

Jennifer Kyle: Now, that's a little bit of when you don't have the right context and guardrails, and we're gonna go into that a little bit more in a couple, in a, in other slides.

‍

And we've got another funny video from Silicon Valley because we are a startup, so why not bring Silicon Valley into the world of R&D finance? Before we do so what we're seeing, what we're hearing, what we're learning, AI is reshaping the industry at an unprecedented speed. It's evolving faster than any of us can adopt.

‍

It's scary, I know. I feel every single time you feel like you have your arms wrapped around AI and you know how to use it, all of a sudden there's a new update and there's new things. And just for perspective it took us about 12 to 14 years to roll out cloud infrastructure that, we all use cloud software, so it was a long time, so we had time to figure it out and get it together.

‍

But now with how fast AI is rapidly evolving and adapting, there's a big change management element at play. There's getting ready with your processes. There's being data ready, and that's really hard to do when it's just constantly moving really fast. And it's like when, if you're trying to implement AI today and you finally get everybody on board in your organization, your systems are ready, all of a sudden, it's a completely different AI six months down the road, right?

‍

And so we're trying to figure out how do we implement and adopt AI and spread it across our organization when it's moving so fast? And I don't know that we have an answer for that. That's why this session is not called best practices. It's called emerging patterns. This hasn't been around long enough to know what the best practices are.

‍

We're all still learning and moving as fast as we can. In some companies, you can move quicker, and others, it's like herding cats and pulling teeth.

‍

So here's some opinions three things that I believe that we believe. Number one, AI is risky, so are people. Who has 100% hit rate on making great hires and training them and onboarding, and they stay with you forever, right? People are risky, AI is risky, and we're gonna talk about that on the next slide.

‍

Number two, prove that it's broken, don't prove that it's right. You're gonna see the... when you put stuff into GPT or you put stuff into Claude, it's always gonna have this confirmation bias. It's your job when you're a user of it to prove why it's wrong and why it's failing and why it's broken, and that's how you get it better.

‍

And then number three, make probabilistic AI a controlled system. And for those of you that don't know what probabilistic AI, you'll hear deterministic and probabilistic. Deterministic is like a system of record. Or you can think about it in a, in an easy way, a calculator. Two plus two equals four. That's pretty fact. That's black and white. That's ter- deterministic. It's math.

‍

Probabilistic enters in a layer of subjectivity and judgment. So clinical accruals, it's really hard to come up with a clinical accrual without some level of subjectivity, which is where it gets really dangerous using AI for an accrual. Forecasting, that's probabilistic AI because you're projecting what patterns may be. But you can use probabilistic when you have a controlled system, just like you have an analyst and you have guardrails and reviews around what that analyst produces for you.

‍

So AI is risky, so are people. So here's some idea, or here are some ways that AI fails. It it fabricates confidence and misses data. It doesn't have it really, it, it doesn't know how to handle complicated trial, like study edge cases. So as you go deep into domain-related verticals, it's really bad at being precise and specific when you need a lot of that context.

‍

Hallucinates. Of s- of course, it hallucinates, and if you think about a large language model, it's effectively getting smarter based on the world data that what populates it, and you can't control what other people are putting in, which is going to affect the output that you get from it, right? So any time, and we're gonna talk about this on the next slide, any time you're just using a large language model to put stuff together for you, and you really need deep context because you need that level of precision, you're gonna get in trouble eventually.

‍

And then to the last point on how it fails, it scales really bad workflows exponentially. So if you build something out in Cowork and you haven't thought about that process end to end, and it runs for you, and it touches different things, it's, it, it could end bad, and there's actually a really funny video that we're gonna show here in a few slides.

‍

Now, these are the bad things about AI. There's a lot of really good stuff that we'll talk to in a second, but people fail too, right? People miscontext. How many times have you said something in a meeting last quarter, and the next quarter nobody remembers what that is, right? We all have a lot going on.

‍

People get tired. They're distracted. Politics happen. People get defensive, right? And so there's, some people are gonna learn really fast, and some people are not gonna learn really fast. They meet your expectations, or they don't. So it's the same. You're gonna use AI. You're gonna use people. Which one is more appropriate? Which one is gonna help you? And then how do you build the right processes and guardrails around AI or people?

‍

For those that are former auditors in here in the room, you're probably really good at this. Prove it's broken. You are so ingrained, and I can only speak for EY, what can go wrong? What can go wrong, right? So any time you build something, building it is not the hard part. That's the easy part. The iterating and fine-tuning it and making it correct and precise, that's the hard part, right? And to do that, it's a constant series of, "Why are you wrong? Why is this not working?" Just keep doing that, right?

‍

So number one, you can build something, then prove why it's failing, then fix that issue, then reprove why it's failing. And when you get... when you do that cycle enough time, now you get to a level where you can actually use it, but it takes some work.

‍

And then three, make probabilistic AI a controlled system. So look, AI accelerates both sides of the equation. Innovation, adoption, empowerment, and control, structure, and validation. But if you do not balance both of those, meaning you're so quick to just innovate and adopt, and you have a, a, somebody new in your organization that's just gonna go gangbusters on adopting AI all over the place, and you don't have the control, the structure, and the validation, that can end very badly for you, which is the next slide that I'm gonna show you.

‍

On the control side though, let's say that you're like, "I don't know AI. I don't trust AI. I don't wanna use AI. I'm just gonna stay in my lane of control and structure and keep our processes and stay manual because, I can trust that." That's a losing game down the road. A- and your organization, your C-suite, people are not gonna be happy with that outcome.

‍

And boards the C-suite, they need to be moving faster. You need to be able to a- use AI to adopt to help your competitive advantage in whatever it is that you're doing. So what, when you have innovation as part of your culture, experimentation, go out, try new things, have fun, put guardrails in place, and you have a good process around making sure that you're validating whatever process or wh- how you wanna use AI, this is where companies and teams work and move really fast.

‍

So what happens when innovation is greater than control in a completely over-indexed way? This is gonna be a little bit... This is a little funny. This is Silicon Valley and Son of Anton. For those of you that don't, haven't seen Silicon Valley, it's a really funny show on HBO. I can't watch it anymore. Hits too close to home. But this is fun. So it's about a minute and a half long. Watch this video, and I apologize for some of the profanity. I couldn't cut it out.

‍

[Video clip: Silicon Valley, "Son of Anton" — the team discovers their AI has deleted working code and ordered 4,000 pounds of beef.]

‍

Jennifer Kyle: Okay what did we learn there? We learned a few things. Not having guardrails in place, that can be very bad. So when you guys are using Claude, for instance, and you are plugging it into, I don't know your inbox on Microsoft Outlook or Microsoft Teams or anything that you have connecting into Claude, there's always settings and it says, "Can I read?"

‍

Yes. "Do you want me to overwrite?" And there's a no, there's ask first, there's yes, always. I think we know as finance and accounting people, very rarely should you ever say yes, always or you end up with a Son of Anton situation and inappropriate emails that go out. Governance, right? We also learn that if you don't have good governance in place, like not telling the CEO when you're doing something and enforcing that kind of culture, that's also very bad news.

‍

And then context, right? The reward system. So he wanted to find cheap hamburgers. Type in, "Find me cheap ham- hamburgers." It goes and does it. It orders 4,000 pounds of beef because the price per pound of beef was probably the cheapest. So when you don't add in the right level of context, you're gonna get 4,000 pounds of beef.

‍

This is just funny. So aside from there, so where... how is AI evolving? I wanted to bring this to you guys because I speak with a lot of the big companies, AI companies, foundation models, tech companies out there on what they're doing, what they're seeing, and what's being built that none of us are using yet.

‍

Or when I say us, I'm talking in particular like biotech companies, pharma companies. You guys don't really have line of sight of what's being developed, what's being funded right now, and it's something that I'm very curious about because I always, I'm always looking to see where is the industry going.

‍

So what we had before AI Chat, right? I think this all just started really bubbling up for us last year, although it was around quite long ago. Now we're in the agentic AI era, and where we're moving now is AI native where every department and process is AI first. And what's gonna be really exciting, and I'll share with you this in a minute, human and AI workflow- workforce.

‍

People, you guys are all gonna be working with agents. You're gonna be managing agents. Agents are gonna be working with one another. Agents, every software company is gonna have agents. There's gonna be a lot of agents, and this is inevitably where the industry is going, right? We've got a little bit of time to get there, but everything that, we're building at Condor, we're also thinking, how is what we're building today going to be in line and really embrace what the market is gonna have?

‍

And how do we make, for instance, how do we make the human and AI workforce with the right governance in place, making sure that the agents can work with one another and we can trust how they're working with one another, right? So all of that is coded into our design and what we do.

‍

I did a a session, or not a session, I did a blog on what is AI and how do you think about it? And AI is not one thing. I think we all know this by now. But you can think about it as that six-layer cake. You've got the layer one, which is the raw processing power, layer two, the systems, pipelines, the infrastructure, layer three, the large language models.

‍

That's your GPT, that's the ability to chat in Copilot. It's Claude Chat, right? Then what you have is the ontology, the orchestration, the agents, and the apps, and that's the most exciting part, right? And I think a really important thing here is when you're using Claude Cowork, for instance being able to being able to automate a workflow, you require context, and you require a lot of context.

‍

And without having that context, now you have an agent working on an LLM model which is taking feedback from the universe or the world as a whole, which is not always the best case.

‍

And so with AI moving really fast, what does this mean for teams? It means more speed. It means more people, oversight, judgment, and strategy. So number one, capital efficiency and velocity. We have to use AI. It's our teams or our boards are asking us, the CEO is asking us, and when you're a biotech company, it means how do we move faster from discovery to development to commercialization with the same or less capital?

‍

How do we learn better business decisions that are gonna drive better outcomes? Then you have Jevons paradox. And Jevons paradox, for you guys that don't know what that is, it means when something is efficient, you use more of it, not less of it, and we're on that Jevons paradox curve. And that third one is on human-centered AI.

‍

You can delegate tasks, but you cannot delegate judgment. And so you leverage AI to delegate all of these tasks. You still need to be the one to make the calls. And so here in a little bit, we're gonna kinda go through and we're gonna show you within Condor both externally what we have in chat the ability to chat with your data, what that looks like, the power behind it, and then what we're doing internally on reading protocols and site amendments. Jeff is gonna show you that, which is called Talon.

‍

So we have three levels of use cases that we're seeing in the market today, and I'm curious to hear where everybody is. Level one, there's the ad hoc LLM tools. That means you as a person, you're using LLM to maybe help sharpen some of your emails, help understand questions that are being asked of you, and then you're taking that information and then doing whatever you're doing in a software platform.

‍

The second use case is where you have a shared LLM model that is, and your softwares are connected to the shared model. So maybe your clinical teams and your finance teams are sharing a large language model together, and you're sharing in some of that context.

‍

The third one, and to be fair, most people are somewhere between one and two, depending on their, the maturity of their organization, how pro they are on AI. It varies widely between small companies and larger organizations. But then number three, this is where the industry is going, shared context across the value chain. And this is what level three looks like, and this is what the future re- really looks like, where you see the value chain from discovery to development, manufacturing, commercialization.

‍

What is agentic orchestration between all of those functions on driving decisions, on transferring data. How do we leverage manufacturing data to inform discovery and development and commercialization? How do we use what we're learning in development to inform commercialization? And so down here you have, study-level truth that rolls up into the program economics, that rolls up into the portfolio roll-up that has now universal company-level context. And this is effectively context engineering.

‍

So we're curious. We're just gonna do a very short poll. What level of maturity is your organization at? You got level one, ad hoc LLM. You're just using your own LLM model. And then we have pockets of AI uses, so maybe you're sharing a large language model with your whole team, and your teams are sharing with large language models with each other. Or are you on level three? Is your team building out level three, which is across the entire value chain?

‍

Okay. That's what I would assume. Jeff, that better not have been you that said level three.

‍

Jeff Jahnke: That was not. Okay. I can't vote.

‍

Jennifer Kyle: That's not... That wouldn't have been fair. I would say, so most of you guys, 60% are at level one, 26% are at level two. One person said they're at level three. Love to chat with you. And then we don't use AI. That's about 12%. So that's about right. We're all right here, and right now we're all just trying to figure out how to do number two for the most part.

‍

Okay, I'm gonna give you like also a quick history and background of what has been being built on and worked and how long it's taken to get to where we are, right? So in 2023 is when prompt engineering was a thing. How do we engineer better questions? What are we... what questions should we ask, and how do we improve the models, the observability in the models, which is effectively the results you're gonna get?

‍

How do we observe or how do we increase the observability of the models through the prompts that we're asking? Which makes the output better. You're gonna get better answers and because it rates how popular that question is and what answers we're giving. So anyways, prompt engineering, that was 2023.

‍

We're in 2026, and for us in finance, we're just figuring out how to do decent prompts. This is how long it's taken us, right? And we're still lurking. We're still working on it. Context engineering, so 2024 and 2025, this was really deep. Con- Condor, this is effectively what Condor has been building, context which is, studying business, studying the intersection of clinical trial operations.

‍

What does a protocol mean? What does a schedule of assessments mean? What does every procedure mean? How does that relate financially? So there's so much you have to understand from clinical operations to be able to predict what's gonna happen financially, so you can't just code all of the finance. You have to understand both sides.

‍

You also have to provide the context around the accounting, the internal controls. How do you communicate things to auditors? It's context. And then harness engineering, and this is next level, and this is what's being built this year. Not just Condor, but this is all of the companies the software companies that you're using. Some of the bigger ones are probably building out some harness engineering so that they can trust their agents to work with other agents, which is the governance and the operational infrastructure for agent-to-agent interaction.

‍

And I mentioned this a little bit, right? What we're building we build AI into everything that we do. We've been doing this for years. I just didn't have the right words, and funny enough, the context to communicate what it was we were actually building besides, "Hey, Condor has a lot of really deep domain information coded into the software."

‍

What does that mean? It means an ontology and a knowledge graph. Very deep IP. But now what we're able to do with that, number one, surface patterns. Patterns that are not as obvious to the human eye when you're so busy doing 10 other things. You're just... You don't have that deep concentration to see everything that's going on.

‍

So Condor's gonna surface that, flags the risk. So we help you think about, again, what's not top of mind for you, and flagging that risk is actually way more powerful for number three, because it bridges the context across functions, and it's also studying the industry as a whole. So what are we seeing in other organizations around CRO behaviors and habits that we can then share with whoever's using Condor, the application, right? This is sharing context and experience across the entire industry.

‍

But no matter what, it's not replacing human judgment. All it is it's decision support. It's giving you information. It's helping you ask better questions, more questions. And then it's up to you working with your clinical teams and any of your teams in your organization on how do we, how do I help surface these insights to drive better outcomes and automate a lot of this laborious task?

‍

If I'm spending most of my day going through Excel schedules and ticking and tying things, obviously we know those are really quick wins that AI can automate. And you don't need Condor for that. You can do that in Claude.

‍

Jeff Jahnke: So I think with that we wanted to open up for some questions. So I know we have a few minutes left, but would love to hear from the group if there are any questions about anything that we've talked about. Please feel free to put your questions in the webinar chat.

‍

And while we're waiting, maybe Jen, I remember at ABFO, one of the big questions that, that people repeatedly raised to me was, "How do I get started? I'm not using AI in my own organization. I'm not quite sure how it works. We've got people that are hungry to use it," but just where do we- how do we get started with all of this? What are your thoughts on that?

‍

Jennifer Kyle: Me. So I would say just get started anywhere you want to. Where the, wherever you spend a lot of time doing something that is just highly manual, that's a good place. So for instance, if you are I don't know, from an accounting perspective, those really long gnarly disclosure checklists, have it fill that out for you.

‍

Jeff Jahnke: Yeah. One thing that I found too is I heard this a lot, was that AI adoption is really being pushed by upper management, but a lot of folks at the lower levels of organization also don't know where to get started. One thing that I did, 'cause I noticed this on my own team, right? I use AI constantly.

‍

I've really embraced it, but a lot of the team just doesn't feel like they have time. I set up what we called a hackathon, where I basically partnered with my team to pinpoint what's three or four of our biggest headaches currently, and worked with the team to set aside, four hours on a Friday afternoon where we didn't have a lot of meetings, and just work through those scenarios as a team, and it really built the team camaraderie and also solved problems.

‍

A lot of what I've found is just getting in there and playing around with it is the best way to start. There's not like a playbook, if you will.

‍

Jennifer Kyle: Yeah. Okay, we have some questions. So I would say this is an easy one. Lindsay, Lindsay asks, "Are you rolling out Talon out to Condor users, or is it just an internal tool?"

‍

We're gonna be rolling out to users but we've got more, Because this is gonna be touching a lot of the data validation, we're very serious about using responsible AI and having controls around it, so it's still under testing and development. But we can do some beta stuff with you. So hit up Jeff after this.

‍

Vinay asked, "Can you explain agents and how do you create one?" I think, Jeff, you could probably take this 'cause you've built quite a few agents. And then-

‍

Jeff Jahnke: Yeah ...

‍

Jennifer Kyle: Ravi asked "How can we leverage AI for quarter-end validation for accrual accuracy?" I would say for that one, probably an easy one to do is to load in load in your current workbook or your current accrual and load in your prior accrual, and then ask it to validate the differences and call them out for you.

‍

It can also ask if you have supporting documentation, if you're a reviewer and your team member gave you "Here's my journal entry," or, "Here's my reconciliation," you can ask Claude to review it for you. What... Where the limitation is gonna hit though is I wouldn't review, I wouldn't have Claude review judgmental estimates on the judgment, because that's where you need context.

‍

And even if you were to add in context, word of caution for everybody, if you add in context and you don't add in grounding elements, then what happens is you start having hallucinations and you have model drift. So you have to keep your context precise and grounded. So you have to engineer both at the same time, which is generally why people don't use generic AI for complicated stuff, because of that grounding element.

‍

But Jeff, can you share just quickly, how you, how do you create an agent?

‍

Jeff Jahnke: Yeah, great question. So I think if we wanna go all the way back to how to create an agent, it's first having a problem that you wanna solve, right? Because all an agent is it's essentially a pre-designed workflow in an AI tool of your choice.

‍

Copilot has it, Claude has it, OpenAI ChatGPT has it. But essentially you're just creating workflows and teaching, almost like you would teach a new employee how to do something that you need solved. And you can create agents directly from chats, and then essentially save the workflow. And then at that point, any time you ever need that expertise or that particular task called u- upon again, you can go select that particular task and then feed it the new information.

‍

So for example you could create an agent that basically puts together your board slides for you. And what you would start off by doing is maybe uploading your board slide template and uploading all of the source documents that go into those board slides, and then say, "Hey, I want to create an agent that helps me create and validate these board slides."

‍

The AI will assist and review all of the documents as well as the template, and then will allow you to save that workflow so that then the next time you can just upload the source documents and it will generate the actual slides for you at that point. That's an a very basic example of what an agent can do.

‍

One thing I would caution against is a lot of agents more agents seem better because they're automating more of your job, but you can think of them as employees. To Jen's point it's not gonna be perfect 100% of the time. You're gonna have to review the work that your agents are putting out there.

‍

And it's okay, if you spin up 30 agents, that's 30 additional employees that you're having to manage, right? Depending on how many people you manage today, that might not sound as enticing. But yeah, that's a little spiel about agents.

‍

Jennifer Kyle: Yeah. Thanks. And Vinay, I would say if you're using Claude Cowork you're not gonna be able to do it in chat. Use Claude Cowork, and you're gonna wanna use Skills. And so look into that, but just quick highlights there. If you have any further questions feel free to reach out to either Jeff or I, and we can go a step le- level deeper with you that's more specific on what you're trying to do.

‍

And then Anne says "How are you tying the AI pursuits to overreaching business goals and making sure that you're not implementing solutions that expose and scale organizational disconnect?" That's a great, that's a great point. And I think in order to go to work cross-functionally, number one, and have everybody use it, you need this top-down approach and change management.

‍

You need a couple things. You need your processes to be ready, you need your people to be ready, and you need your data to be available. If you don't have data that can be accessed and available, it's gonna be really hard to use it, and so there's u- generally a lot of cleanup work that needs to be done, and that's up to the organization to figure out when is the right time for them to go go across the whole value chain.

‍

Right now what we do is we just, we provide business context. So we have a lot of deep clinical operation context that you need to understand to help drive business decisions and outcomes, and we put that context in the hands of finance people today. Now, that context, though, can also be shared with the clinical operations team, and we're working with clinical operation partners on refining those models for their own use.

‍

But it, it really starts with the organization, the culture of the organization, and how they wanna work together, which is no different than implementing any kind of enterprise-wide system where you need to have multiple users using it.

‍

Jeff Jahnke: All right. All right. I know that we're at time here just wanna thank everyone for their participation. It's fantastic. If anybody's really interested in being one of the first users of some of Condor's AI capabilities, please reach out and also scan this QR code. We have a AI takeaways blog that will be helpful in just continuing educating you and giving you more tips and tricks and ideas for how you can leverage AI at your own organization.

‍

Wanna thank everyone again for their time and, looking forward to discussing AI continuously via webinars and other forums here at Condor. Thank you all again, and we'll talk soon.

‍

Jennifer Kyle: Thank you, guys. Thanks for your time. This is only the beginning. It's exciting time. Bye.