AI gets faster, but building systems that last is still engineering
In this interview, Jay Tailor explains how Condor Software uses AI and a deep knowledge graph to rebuild the financial backbone of drug trials, lifting forecasting accuracy near 90 percent. He argues the core basics of engineering do not vanish with AI, and that finding people who build end-to-end systems is the real bottleneck.
Jay Tailor is VP of Engineering at Condor, a biopharma software company rebuilding the financial backbone of drug development with AI and data. He describes an industry where a single clinical trial can run eight to ten years and cost tens of millions, spread across vendors, hospitals, doctors, and patients around the world, with the money tracked in sprawling 400-tab spreadsheets. Condor's mission, as he frames it, is that no drug treatment should fail because of financial uncertainty.
His work centers on turning that tangle into something legible. Condor connects ClinOps and financial data into a deep knowledge graph, layers rigorously checked math on top, and uses a mix of a custom model and governed LLMs running locally inside AWS so that regulated customer data never leaves the environment. The payoff he cites is concrete: forecasting accuracy near 90 percent, accounting timelines cut by 70 percent, and analysis that once took four or five BI engineers now returning in minutes. He is careful to separate the deterministic core, budgets and patient counts and site charges, from the probabilistic analytics that AI layers on top.
“Our mission is that no drug treatment should fail because of financial uncertainty.”
Tailor is skeptical of AI hype that skips the hard parts. He argues that the core basics of engineering do not disappear with AI, that a system still needs context management, caching, vector search, audit logs, and feedback loops to stay useful four years on. His biggest worry is teams shipping bots and tooling loosely and piling up tech debt that AI cannot simply be thrown at, and he keeps returning to a metaphor he and host Josh Rubin build together: making the baby is easy, raising it is the work.
Leading a lean engineering team of ten to twelve, he sees the real bottleneck not as writing code but as finding people who can build end-to-end systems with a product mindset. He points to the rise of forward-deployed engineers who go into a hard problem, understand it, and bridge it to the tools that can solve it. On team-building, his rule is simple: work with people's strengths and around their weaknesses, because in a complex machine every part, from the critical component to the smallest bolt, has a role to play.
The conversation
In this conversation: Josh Rubin (Host, CTO Studio) and Jay Tailor (VP of Engineering, Condor Software).
I am VP of engineering at Condor. It's a biopharma software company, basically revolutionizing the complete financial landscape for biopharma using AI and data.
What does that mean?
Think of the whole drug manufacturing from the discovery part. When a lot of companies invest into the research part of the drug, like, hey, this is the particular disease we want to solve, they invest a lot in research. Then after all the research gets completed, it comes to basically trial. They send it to government for trial approval. The trial is a very complicated process because it runs from eight to 10 years, just for a cancer treatment and things like that. And this requires a lot of financial commitment from the companies. It's very, very difficult because you have to work with different vendors, different hospitals, doctors, patients all over the world. The whole financial system is very complicated, and top biopharma companies basically hire tons of accountants just to make it happen. Now, what Condor is doing with AI is we are bridging the gap between what's happening in the real world and how this trial can be done in the most efficient manner through a financial platform. Our mission is that no drug treatment should fail because of financial uncertainty, and that's what we are trying to do. So we connect all the ClinOps data, all the financial data, and make sure that you can plan a particular treatment trial, like a cancer treatment, instead of it coming in at, let's say, 10 years. You can optimize it with the particular sites, particular patients, particular patient groups, and particular geolocations, and you can bring it down to maybe six years and with more financial transparency. So instead of spending like 20 million, you may be able to do it for like eight million. So you spend way less money, less time to get that drug to the market, to the people, as fast as possible. It's completely revolutionized by AI. In the legacy way, the accountants basically manage 400-tab Excel sheets, all hooked up to each other, their financial system all over the place. It's so hard to maintain. And 90% of the drugs don't get made because of that uncertainty, because they don't know where they're spending money and things like that.
So if I had to break this down, ultimately we're talking about not just financial modeling of drug trials, but the facilitation of building the financial model for the drug trial and then executing on that?
Exactly. It's forecasting. It's budgeting. It's benchmarking. Pretty much everything. For example, a company like J&J or Merck, these big biopharma companies, they can literally say, "Hey, I've got these 20 research programs which are showing pretty good results in research. Which one should I go with?" Even before they start, with our data and the AI, we can actually help them pick and choose. Considering all the data we train our model on, we understand what exactly it takes to get there, what is the chance of success, how much money you would need to get this drug to people, how much time it will take, which sites you need to work with, which hospitals you need to work with. That's not available beforehand without our AI. So that is the real impact, because today it's not available.
Obviously there have been systems in place to do drug trials for years. And for these massive big pharma companies out there, they have in-house teams that are building this, a data science machine learning problem. When you say AI is changing this game, is that really LLMs that are changing the game? Is that part of this AI?
It's a little bit of everything. So LLMs are definitely changing the game, because with LLMs every big company uses their own system, their own data sets.
And that data set is based on historical data, as well as what we're currently testing, all the data associated with it.
Exactly. So the LLM is basically generalizing this data. Now you can do it way faster. One of the things we are seeing is that the accounting, finance, and analytics systems don't even talk to each other, because they are so big that the data is all over the place. What AI does is create the vertices of a graph between these different systems, like site, hospital, vendors, patients, your investigators, which are the doctors, the trial. We create a very, very deep knowledge graph within Condor, all implemented with AI, and that enables us to give you insight which was not possible. On top of that, we have put really good math, a calculation, because it's all financial systems. Not everyone understands that math. So the team has done quite a bit of research to make sure that the math adds up, the calculation is proper. So with that ontology graph, and the math, and all the data, now we can predict way more accurately. Our forecasting accuracy is, for example, 90% today. The whole accounting process is very critical and complex, because you need to be in a regulatory environment, a secure environment. So we were able to reduce the whole process by 70% in timeline because of using AI.
AI systems in general, and LLMs in particular, are probabilistic systems, but you're dealing with some deterministic stuff, math ultimately. So how are you putting a deterministic layer on these probabilistic systems?
The underlying math and the system is very deterministic. When you enroll 100 patients, every patient shows up to your trial, you're giving them some money, and there is a budget assigned to it. It's very deterministic in how you're calculating what the hospital is charging. Everything is very deterministic, and that's all mapped to each other. So you never lose track of what exactly is going on. The probabilistic part is, now you have all the data, and all the calculation is very deterministic. Instead of writing it back into the system, can you figure out an analytics part of it, which can help you get deeper analysis? To give you an example, a lot of times these companies would like to know why a particular trial is struggling, where we are losing, which site is not working well. Today it takes a team of four or five BI engineers to really dig into the data and run queries. With AI, now that is like a two-minute job. Every company has their internal fields, and mapping them to each other is very, very fast.
It's the connective tissue, essentially, that makes this work really well.
It is the DNA, basically connecting everything and flowing everything, and now the precision is the key.
The outcome for this is a successful trial. When they're talking about using AI to come up with new drugs, are they actually coming up with new drugs, or are they doing what you're talking about, making the process for finding new drugs a little bit better? Or is that just a different animal than you're dealing with?
I think it's a different animal, but everything is connected. Manufacturing a drug, or finding a new drug, is a very different ballgame. But I know Google just invested, I think, $2 billion in a company in the UK, where they're literally using AI, because even on the human ontology level, it's like DNA and math. Every protein synthesis, how it reacts to a particular chemical, is all math. So they're using AI a lot to do more experiments, wider algorithms, better outcomes. So that definitely revolutionizes the industry and makes it more efficient. Then the second part, which is a very clumsy part, is this trial. Then the manufacturing part, because once it gets approved. So for example, a company can do 20,000 research programs in a year, and they may spend about $20 billion for that. And then they may spend another $2 billion just for the trials. Out of 2,000, maybe they just submit 20. And in that 20, 90% fails because of this financial uncertainty. Because they spend so much money, they don't have enough way to optimize this, they don't even know what they can get out of it. Their milestones are not very defined. They can't even predict how much more money they need. Sometimes they're so close, but they run out of money. And this is the transparency Condor can provide, basically, to make their whole financial plan.
Is it lowering the barrier to entry for smaller players to run trials?
Oh, 100%. 100%. I think we are already seeing some traction from the market where they're like, oh, this is great. How can I just connect all my data to your system and then start predicting it, start planning my trials?
IP, I imagine, becomes very important here too. Are you a cloud-based system? Are these people building their systems on-prem? We're watching some of those transitions with AI happen. And in addition to that, are you using various models, or are you using an in-house model? There's a lot of people very concerned. Am I just giving Anthropic, am I just giving OpenAI all of my information?
We are using a mix of that. So we have a custom model. We also use LLMs. But even those LLMs are in a very governed environment on a cloud. So it's not like we are sending the data to Anthropic. We're running a localized LLM within AWS. So the data doesn't go outside, and the data is also secure. We work in the biopharma industry. This is the most highly regulated and compliant industry. So we need to make sure that we do not compromise with securing the data of customers. The LLMs train on a very limited environment and give the prediction based on that. The LLM doesn't have the power to write it back.
You haven't had the data leak yet, though. I'm very interested in pharma, biomed, just because they are highly regulated. So if they can figure out how to do this within regulations, that becomes a very important use case for everyone else. However, we haven't had the mass screw-up yet that's gotten reported. How secure are these systems? How confident are you that we're buttoned up here?
That's a great question. I think every company tries their best to protect their data through all the technological advancement and implementation. We are so compliant, going through an auditory process every year. We know who's accessing the systems, who has the changing power, who has accessing power. It is a very well-layered security protocol we follow, so that even by mistake it doesn't screw up. So there are two things. One is through tech, you make sure. Then there's an internal process companies follow. Everyone is tied to a particular ID, and then this ID only has access to this, this, this. Even if it's a customer outside, like the other day we were having a conversation with a customer who has a contractor, for that contractor we know that we cannot give them every access to even their own company's data. We only allow them limited access. So you have to put a system in place which allows you to create those layers in terms of policies and also tech.
What's your revenue model for all of this?
I wish I could speak more on the revenue model, because I'm not in finance, but I think it's more on the trial basis and also all the add-on functionality, premium on how you're using our forecasting capability, how you're using our AI agents. There are forecasting agents. There's an agent which can help you with the benchmarking. So everything is an add-on piece. But it's all about how we can provide the value to our customers to get the medicine out to people as soon as possible. It's a very mission-driven culture and goal we are shooting for.
And a lot of these AI companies, you're not SaaS, you're more solution as a service in that new rubric. How big is your team?
The engineering team right now is about 10 to 12 people, very solid engineers. What we're hiring for is a really smart senior staff engineer level, someone who has enough understanding of traditional machine learning and data engineering. At the same time, they can go into this AI world, running AI agents at scale, because this is where the world is going today. It's all about running AI agents in production.
Are you a cloud shop, are you a codec shop, or is everybody just building how they need to build, just get it in here?
Pretty much the overall high-level architecture is the same for everyone, either in GCP, AWS, or you're using third-party companies. But I can see that a lot of AI companies are in the generative AI field. You're trying to create a new type of image, new type of videos, new type of this. And they're okay, they're going to be successful, some of them if not all. On this biopharma, or these types of companies, or in manufacturing or construction, I could see that their architecture may look different, because they're actually trying to solve a fundamental workflow or process problem, which was not possible before. But with AI now, they can. It makes it very, very easy for them to solve that.
AI is a big word. What you're talking about, these process, these machine learning problems have been in the industry for 30 years. People have been using big data and data science around this kind of stuff. The difference is now the LLMs give you the ability to ask and query that information and retrieve it differently. Is that the big difference here that you guys have found? Is it simply that you're able to architect a system that people can access in a new or different way?
So I think even with AI, the core basics of engineering don't go away. Anyone can say, hey, we can bring in all these different data sets and put it in a cloud, and why can't we get the same results? You can't, because how about you ask the same question after four years? Would your cloud have the four-year-old memory? Would it have the context? The times have changed, the data has changed. That's where the engineering comes in, because you need to build an end-to-end engineering system where you're thinking about context management, your long-term memory, short-term memory, your caching layer, your hallucination-fixing layer, your audit logs, creating a feedback loop. Everything is an engineering end-to-end. And that's where the real beauty of these companies is going to come. Whoever does it really, really well is the one that's going to be successful, because that's when you can show the impact. Even after four years, you're like, hey, I'm asking this particular thing and I need this analysis. And if I refer it back to four years, I think we had a similar thing you asked about, and this was the thing, but now this has changed, and this is the outcome.
Are we at the point now where you can build a system that is sustainable for four years? Things seem to still be changing so rapidly that architecture evolves.
It's a great question. I think no one knows, but at the same time, I also feel like the basics of engineering, which were founded even 50 years ago, are not going to go away. What's going to happen is AI is going to get better at everything it does, but you would still need a caching layer. You would still need a very scalable database. You would still need a vector search. You would still need some sort of machine learning training and inference. So that is still going to exist, and you need to optimize it. You need people to manage it. It will probably do it faster, it will probably do some things out of the box, but you'll still need people to do that.
You've been in the industry for a while. You've seen tools come and improve. What's worrying you right now?
I think the biggest worry is that every company, with this revolution of AI, thinks that they have the power to do a lot, which they do, but it's loose right now. So what's happening is every team is creating 10 more things in a week. Who's going to manage it after two years? No one knows. How are you going to make sure what you're creating is actually valuable? People are creating a lot of bots, a lot of random tooling, applications and everything. Everyone is like, oh, this is really good, but then what happens after it? How does it turn into a return on investment? How does it turn into real impact, business growth? How does anyone benefit from it? I think companies need to figure out, and that's the biggest challenge for everyone, constraining the environments where they're developing this. Then they can really use the AI in a very good way instead of going loose with it. It's like you have a billion dollars and suddenly you're like, hey, what am I going to do with this? I'm going to spend a lot of money everywhere. But it's more about carefully planning that.
Well, it's also that building is easy, maintenance is hard. And to be more crass, making a baby is fun, raising a baby is exhausting. Not everybody is cut out for actually raising the baby and doing what it takes to turn it into another productive member of society. We're kind of in the AI making-baby phase and haven't really considered what's going to happen when they start growing up.
Yeah, that's a great analogy, and I'm going to use it for different things. You have fun with that one. Yes, but I think you're absolutely right. Everyone now has a lot of babies, and they just need to figure out how to make sure that they drive them in the right direction for the right thing. Otherwise, we are going to create a huge tech debt for ourselves. We are going to create a huge amount of work, and then that would kill the companies eventually.
Or I've heard plenty of people arguing that tech debt isn't a problem now, we'll just throw AI at it.
AI doesn't solve its own problem. Probably it will come up with another problem. So it's going to go in a spiral loop. It's basically, oh, I'm fixing this tech debt, now I'm going to be at another tech debt by fixing this tech debt. Then how do you solve that? Because it's just AI fighting with each other. It's hard.
AI is real good at solving problems AI made.
Yes.
Where's your bottleneck? You're running a team right now. The coding obviously isn't the bottleneck anymore. Where is it for you?
I think the talent that is good at building systems is the bottleneck. You will always find people who can write code very fast and make 10 PRs a day when they were making maybe one PR in a week. But finding those people who can actually build an end-to-end system with that product mindset is the challenge. There are not many people who actually can do that.
Herein lies the question. Teams used to be built with a couple of superstar ICs, and then, well, this guy will just get this part done because somebody needs to do that. But if we have to optimize everything for the 10X, the 100X guy, how do we make those?
I think that's where this forward-deployed engineer term is probably going to be heard a lot in this AI world. Now there are so many problems in this world yet to be solved, but the companies who are dealing with those problems don't know how to solve them because they're stuck in that environment for a long time. Then there are these tech companies which have all the power and know how to solve it, but they don't know how to bridge it, because someone needs to go in there, really understand it, and then connect it to solve it. I think that's where a lot of it is going. Microsoft recently just did a $2.5 billion investment in forward-deployed engineering. It's like hiring so many engineers to go into those particular problem sets and connect it through their own internal AI systems. That's where the revolution is going to happen next.
In product, ultimately. People that are just walking out there and saying, I need to understand everything about this, and then we'll figure out how to fix it.
Exactly.
Not every engineer can be a product person.
And that's where the challenge is.
What do you do with those engineers that aren't going to make that transition?
I think there's still a lot of internal work, and there's also a lot of innovation. Even if you're not solving a product problem, for example, I was talking to one of the engineers on my team recently, and we talked about how we can do a complete end-to-end agentic workflow even for internal development. Can we have a system where someone just says, hey, this could be a bug, and tags a bot, and that bot spins up a new environment, tests it, validates it, reproduces the bug, fixes it, and then opens up a PR? But building that whole system with all the rail guards and making sure that everything works properly and doesn't do anything stupid is going to take some time, a lot of time from a couple of engineers to build it and manage it for a long period of time. So having that will also add a lot of value. Or it could be automating your internal processes, which is going to be a huge amount of work. Or it could be converting a sales process into something else. So there's a lot of mundane work going to turn around with those types of engineering.
Raising the baby, I think, is actually applicable here too, because there are a lot of people that are parents. Some people are just there, and some people are present even for the boring parts. And you need somebody that is trying to guide, trying to evaluate, like, is it helicopter parenting? I'm trying to find the right metaphor here, but ultimately it is about somebody who isn't just going to set it and forget it, or just show up around dinner. They actually have to be ever-present in the thing they're building.
Yeah, I always use the analogy of a machine, a very complicated machine. Not every part plays the same role. There could be a particular part which is very, very important and does this. But there's also a screw and a bolt, and there is some panel which also plays a part. The machine won't work without anything. So in this whole system, everyone will play their part, even if they don't have a product mindset. It's just like the same baby analogy. Someone is going to get a great GPA and become something in their career. But then someone is going to do something else which is equally important. So it's all about, when I'm thinking about building a team, for me the most important part is that I can create a really good setup with the superstars, the rock stars, the people who actually make things happen, the people who can solve customer issues, the people who can manage things. So everyone will play their part to basically make the great part.
I guess, can you identify, there is something that you are the best at. I just need to know what that thing is.
Exactly. I always say you work with people's strengths and you work around their weaknesses. So you just work with them in that way.
Thanks so much.
Be part of the
conversation.
Whether you're a CTO who wants to be featured, a company looking to sponsor, or an engineering leader wanting a seat in the room — there's a place for you here.