← Equity
Equity
Inside Ode with Anthropic, the private equity-backed startup betting AI services are the future of enterprise
00:31:21
Can a handful of engineers really do the work of an army of consultants? That’s the bet behind Ode with Anthropic — the joint venture dedicated to embedding forward-deployed engineers in enterprise firms, backed by Anthropic, Blackstone, Hellman & Friedman, Goldman Sachs and others.
Loading summary
Transcript44 lines
- [00:00]
A
Hello, and welcome Back to Equity, TechCrunch's flagship podcast about the business of startups. I'm Rebecca Balon, and this is the episode where we bring on industry experts to help us explore a trend in the tech world and dive deep. Today we're joined by Chris Taylor and Eddie Siegel, former co founders of Fractional AI. And I say former because they're now working on a new venture. It's still in the nascent AI services industry, but this time it's backed by heavy hitters like Anthropic and Blackstone. Chris, Eddie Welch, welcome to the show.
- [00:33]
B
Thanks for having us.
- [00:34]
C
Thank you very much. Great to be here.
- [00:35]
A
Fractional is the company that is on your email still, but you were recently acquired by a JV between Anthropic, Blackstone and some others. And you've got some news to share about that JV today. Well, can you give us the news?
- [00:50]
B
Yes. So we are launching under our official name, which is ode, with Anthropic. Very excited to put the Fractional AI chapter behind us, clear up the confusion that has been created by the lack of having a name for this JV and put a real name to it.
- [01:07]
A
Okay, so tell us about the company and what the mission is and why Fractional is at its center.
- [01:14]
B
The basic founding belief of ODE is that non AI companies are going to be among the big winners of this whole AI moment if they adopt the technology the right way. But adopting AI is very hard, and companies need a lot of help to do it right. You know, you're essentially taking this magic hallucinating ingredient and trying to rewire your core business operations or your core customer experiences with it. That requires top caliber applied AI talent, which is not something that most companies have. And so ODE is being founded to close that gap and help companies with that adoption. I think this is an emerging market that you're hearing more and more about that's being referred to as AI native services. And I think with ODE and with the close relationship that we have with Anthropic, we're aiming to build the defining company of this AI services market.
- [02:04]
A
Okay, so what is the breakdown of your companies or your customers, I should say, like, who are you working with? What kinds of companies are you helping? Are they all Blackstone portfolio companies?
- [02:14]
B
No. So it's a big mix of clients that we're working with. We are working with many Blackstone portfolio companies as well as other portfolio companies from across the consortium of investors that we have, so Hellman and Friedman, and we're now exploring engagements with Goldman Sachs portfolio companies and many others. We're also working with many anthropic clients. And so there's a close collaboration between our team and the anthropic commercial team. And they're bringing us into some of these really exciting opportunities where there's an opportunity to use AI to innovate and drive outsized business impact.
- [02:50]
A
Okay, so then what is ODE actually building and delivering? Right. Eddie, I'm curious, like from a CTO's perspective, what are you doing? How is it different than this is just some API calls and we're helping set up an agent or something like that?
- [03:06]
C
Yeah, it's a great question. I think to Chris's point, part of what sort of got us into this market to begin with is we just saw the really early days of this gap starting to grow between these frontier models. The API calls you can make and the answers you can get back are getting increasingly incredible. Like really smart models that are getting smarter and smarter and smarter. And yet the ability to sort of achieve real business results with those models is staying really hard. And so what we're doing is we're going embedding with these companies and like really studying their problems and then building these custom bespoke pieces of software that fit into their problem, that fit into their infrastructure, and can use the new technology to sort of solve these problems. And so there's a mix of really challenging engineering and really complex systems. It's also a bunch of really complex understanding of bespoke processes. Like, one example of a project we were working on recently is for this company that makes software in the governance, risk and compliance space. So picture like complicated B2B software that in order to sell it to their clients, the very first thing they have to do is a bunch of customization and configuration to make their software useful to their clients. It's like very, very customized software. And so they have this kind of team of people internally that has to go understand their clients needs, apply a bunch of configuration and customization. It takes months and it really delays, you know, how long it takes for those customers to be able to get value. And so we went in there, we spent as much of time in there trying to help them, you know, make this whole process better. With AI, where we now are now, you know, building this, this system that can take as input a bunch of, you know, sort of things that happen in their sales process. Right? It's like calls with customers and PDF documents describing processes and email exchanges, things like that, and help them do a bunch of their work. Much, much faster. Where like it can sort of pre plan how to configure the software for them. It can make, it can, it can apply that configuration internally. And so going into that is like a lot of ingredients, right? We have to like build it pretty hard to build system. We have to stand up a bunch of evals to help us measure how well is our system performing versus what their team would be doing without AI. Are we performing as well or are we not performing as well against a bunch of real complicated use cases? It has to integrate with all their internal systems. They have complicated software. If you want to automate the configuration, you have to be making lots of internal API calls into their systems. You have to tune for accuracy, you have to tune for cost.
- [05:35]
B
So this is an example that we see across a handful of, of clients, both software companies and tech enabled services companies where implementation of their software or the configuration of their software is the bottleneck for their business. So the particular example that I believe Eddie was referencing here was logicgate. The CEO of logicgate called us up and basically came to us with the idea that hey, I think there's an opportunity here to automate many of the tasks involved in implementing our software. This implementation process is the key bottleneck of our business. Revenue is six months out while we're implementing software for our clients. How can we pull that revenue in, help clients get more value out of this software and unbottleneck the growth of our business from the growth of this implementation team? And the conversation was essentially like, you are, you are correct, this is a massive opportunity. Many of these tasks are automatable, but it's a lot more complex than simply rolling out cloud license to your team. And so we went in and designed a custom piece of software working hand in hand with their implementation team that was aimed at automating many of these tasks and creating a lot of leverage for their team.
- [06:58]
A
How long does this take? Like if I'm a company that you know is trying to boost my whole workflow with AI, I know there's a problem, right? I know, I know there's ways that AI can speed things up or automate certain processes, but I don't know how to do it right. So I call you guys. How long will it take for you to get like, how long like end to end does it take? How involved is the process? Do you like I'm imagining an army of FDEs for deploy engineers just like hovering behind people as they do their jobs to see how it works. Like what does this look like the
- [07:30]
C
Process is very hands on and it changes throughout the life cycle of an engagement. Right. It usually starts with someone coming to us with some set of hypotheses, but that are not totally well understood. Right. They're not that they think they know where sort of AI can help, but they don't necessarily have every detail in place. And so we spend a bunch of time with them sort of first just solidifying the business case with them. Right. We want to be building things that actually have business impact, that are actually going to see the light of day, that are going to move the needle. And so we spend time with them refining their business case, understanding things like, is this possible? Can I really do this? Is your data available? Will your teams be able to accomplish this kind of thing? And then from there, when we move into implementation, then there's a lot of building. And that looks like what you were describing. Sitting next to their teams multiple times a week, talking to the folks that really know their processes really well and kind of collaborating on the right things to do. The how long does it take? Question is hard because this, often this exists on some long term sort of roadmap and the endpoint isn't necessarily clear. So in the software configuration example, the first thing we wanted to build was something that their team could use to speed up a subset of implementations within a few months. We wanted something in production, useful to them, speeding them up in short order. But the long term goal is to build something that can become so robust they can actually make it a product feature and give it to their customers to do some of their own configuration work. And that, that might take a year or more to actually get to see the light of day, but in the interim, we want to be shipping something of value. One rule of thumb we have is it may be part of a long term big vision, but we want to make sure that something is in production, adding value, measurable business value, within three to six months. Otherwise I think there's risk that it can become this perpetual endeavor that never shows roi. We want to make sure we're actually moving the needle on something.
- [09:20]
A
Yeah. Well, how do you, like, what is a metric that you use to tell you that this is succeeding? Right? Is it adoption? Do you, is it ebitda? Is it headcount reduction? You know, is there revenue growth? Like, how do you know that there's an ROI for this? Like, do you, are you matching like a, like a software developer's salary spend? Like, what are we talking about here?
- [09:39]
C
Maybe Chris, you can like share more about the Kinds of metrics we see. But I think it is like the range of stuff you're thinking about there is the range of stuff. Right. We want it to be revenue and not we successfully ship to chatbot. Right. We want it to be like something that is a business case that actually impacts the business more often than not it takes the form of revenue or growth or some sort of efficiency. Like in the case of the GRC software, they are measuring their time to value for new customers. Right. How quickly can a new customer get onboarded and actually see value? And that time should shrink and they want to see how many new customers can be served by their stretched, way too thin implementation team. And they should see that that same team should be able to handle more and more customers or that they should be able to engage in more and more sort of like high leverage work. So different, different projects are different. I think a lot of times there's sort of like AI enthusiasm that is driving this sort of like investment in the technology. That enthusiasm is usually rooted in like a real opportunity. Like they see that like AI can really help with this part of our business that's driving us crazy every day. Like this. This thing used to be like an immovable and now AI has changed that equation. But sometimes then the measurement of it is immature or different people at the company will disagree about how to measure the results. And so we try and kind of get involved in shaping it into like, okay, if we accomplish this thing, that'll show up in the share price of this company or that'll show up in their ability to grow revenue.
- [11:08]
B
Yeah, more often than not it's revenue growth with some efficiency gains. It's removing that primary bottleneck of the business to unloc more revenue growth. And then if you grow that revenue, you don't have to grow expenses at the same pace. And so ultimately you end up with a business that's making more money. Both top line and ebitda.
- [11:30]
A
Yeah, I mean, well, it's still such early days, right? So I imagine you have to do this on like a quarter by quarter basis, these kinds of check ins to make sure that a lot of the tools that you're, that you're creating are working. And I imagine that because you have so much experience working with different enterprises, like you have a really good bird's eye view into why so many enterprise pilots are failing and not making it to production. What do you think is the biggest bottleneck?
- [11:54]
C
Yeah, we're not checking in quarter by quarter, we're checking in every two or Three days. So it is a very, very kind of, I don't know, we're very hand in glove with these clients where it's very iterative. It's our kind of AI expertise meets their domain expertise that actually makes these things successful. For every one of our engagements, we're very religious internally about tracking a North Star. For every project having once a week we check in on that North Star, we make sure it's correct, we think about the blockers to that North Star we design with all these AI systems, EVALs are absolutely critical. So the fun thing about these AI projects is when you think about how AI can be used to solve these problems, usually everybody's got a million ideas for things you can do. You can use this model, you can use that technique, you can build the system this way. It's like very. There's a lot of enthusiasm. The way to sort of rein that in is to measure it. You start with something really simple. You build the simplest possible system and then you build some system for measurement where you can say, okay, today my system does this thing. You feed it input, here's the output it gives you, and then you compare that to known good output for input. And if you can build that, you can then measure consistently over time. You're sort of up into the right trajectory for how well you're sort of performing at these tasks. And so we are like measuring every couple of days we're at some threshold. We use these evals to spot some edge cases. We bring those edge cases to the team doing the work at the customer and we talk to them about it. We're like, hey, GRC software company, we found this edge case. It's a really hard implementation thing. This is what our system did. How would you have done it? And we use that to check in on the project. And usually some part of those evals is either directly measuring or it's like highly correlated with the business impact you're measuring. Well, if you can drive this up, it is one for one with revenue you're going to drive or it's one for one with time you're going to save. And so you try and measure things that are not just accuracy, but that are very tied to the roi.
- [13:54]
B
And if I zoom back to something he said at the top there, it's the combination of the applied AI experience and expertise to build these systems with the deep subject matter expertise of the client and working in close collaboration that makes these projects successful. Ultimately, you also are needing the senior level buy in of the company because you're trying to change how a company does business. And so you need, you need all of those things to be successful. A lot of the work that we're doing is the top one or two priority for the CEO of the company. It's the most important product feature that the company is going to build over the course of the next two years, or it's reworking the most important business process they have. And so it's really exciting work. The team that we have gets to go work on the most exciting project at every one of our clients with that senior level support, with the deep subject matter experts on the other side as thought partners in building these systems,
- [14:50]
C
I mean, we're in like a super fortunate position, right? Like ODE kind of fundamentally exists in this market where there is a lot more demand than there is supply. And so it means that we get to be somewhat choosy about which projects we take on. You know, one of the primary lenses we're applying is what is the size of the impact we think we can have here. We're trying to take on these top couple priorities of a CEO of a big company where we think AI could provide huge impact if you can get it right. And then usually getting it right is bottlenecked by having the best applied AI team, super talented engineers, super talented forward deployed product managers, folks that actually have experience doing these projects. That's like the sort of the magic set of ingredients that we see work for us.
- [15:33]
A
Yeah. So then how much of enterprise AI success today comes from choosing the right model versus redesigning these workflows? Right. Like, because we know we're always, everyone's so excited about the next model that comes out, this one's going to be a game changer, et cetera. But it sounds like maybe that's not the most important thing. Or is it a combination?
- [15:50]
C
I mean, I think that the improvement of frontier models and seeing them get smarter is a rising tide that sort of like helps with all of this stuff. And so it is very exciting when new models come out that make new things possible. Specifically for the category that we're talking about here of bespoke implementations, where you're building a custom thing to solve a unique business problem. I think model selection matters, but it's not where the majority of calories are spent. You spend a lot more time. It's one ingredient in a system that has to be engineered. It's the choice of programming language. When you build a piece of software, you don't want to make a crazy choice in programming language. And like for some projects it matters more than others. But I would not define an enterprise transformation in terms of like, whether they chose like Python or Java for something
- [16:38]
A
so owed by Anthropic is Claude first, right? You're always going to choose Claude first if you can, otherwise you're model agnostic. Or how does that work exactly? Like, what happens if you know in a few months, like GPT 6 just way outperforms anything that Claude does? Like, will you always pick Claude first if it works well enough?
- [16:59]
C
I think there's more to think about here than just the model. When you think about Anthropic, an example here is like we recently started becoming very heavy users of Claude Tag, which is not a model, it's an application. If you're not familiar with it, it's sort of the ability for Claude's models to sort of live inside of your Slack and independently Cloud Tag. Yes. Run a bunch of workflows for you. It acts as like your coworker that's sitting inside of Slack with you and it is using Anthropic models under the hood. By the way, this has been a really cool experience for us. It's taken a bunch of work for us to get it into a place where it's running these workflows. But there's all these business processes we run internally, where we have our ops channel, where people come with operational requests and it is the primary individual responsible for these requests. Now, internally, sometimes it's got to delegate to a human, but it's delegating to us. It's not the other way around. And so that sort of thing, I think, is what happens when you have this sort of magical kind of close collaboration between us and Anthropic, close collaboration between the application layer, the model layer, there's sort of this coming together of all these capabilities that yields a really good outcome for the business. And so those are the things we're thinking about every day.
- [18:12]
A
So we're seeing forward deployed engineer model is suddenly everywhere. Right. So Anthropic's doing this as like a separate unit. OpenAI, AWS, Microsoft, everyone's doing this. I mean, Palantir is the first one that popularized it but never really talked about it. Why do you think it's worth like having a new company for this? Like, what gap exists to justify it? Or, you know, what did Deloitte and Accenture get wrong? Right. About deploying generative AI?
- [18:39]
B
So I would say the need in the market for this and the reason you're seeing this category explode, there's just overwhelming demand for AI native services and for the types of teams that we are deploying, what it takes, the skill sets that are on these teams that are ultimately successful. It's a combination of an entrepreneurial spirit, an ability to be a flexible thinker, understand how a system works today, rethink it with the AI fluency to rethink it, the flexible thinking to rethink it, and then the builder mindset to go actually bring that, you know, bring that into, into real life despite whatever technical challenges you encounter and whatever organizational challenges you, you encounter. And, and so that, that type of skill set is not really the, the thing that the traditional services businesses were, were built for. And so the, the, that that is why this AI native services category is, is exploding at the moment and why you're seeing, you know, all of these companies pop up. It's because companies, enterprises out there are reaching for exactly this type of solution. They're not getting it from the traditional players in the space. And so there's just a overwhelming demand for these types of FTE teams.
- [20:03]
C
The engineering degree of difficulty also tends to be very high in a way that it is not in many other types of services. Like our team is predominantly, or at least our engineering team is predominantly very experienced engineers. They tend to be generalist elite software engineers. Our headquarters are in San Francisco and in New York. These are top tier, experienced generalist engineers. Over half of them are former founders. And so they're the kinds of folks that can sort of juggle both a really challenging technical problem but also own something end to end.
- [20:36]
A
So not necessarily like a whiz kid that just knows how to play with the tools. Right. They're people who have experience hands on with this kind of thing.
- [20:43]
C
Correct. Especially in this environment. I think what is key is keeping the teams that you're deploying pretty small. And so these folks are tasked with owning in one brain the engineering challenges, the deployment challenges, the cultural challenges, the business challenges, the UX challenge. They're thinking about all of this stuff. And part of that is because the trajectory we're on is one where you can take. And I think this is true in some sense today and it's getting more and more true every day. You can take a team of four of the right people, put the right models and technology in their hands, put them into a company that needs some transformation and has the right appetite for it. They're sort of CEO or board level. Buy into this and that team of four might be able to have a hundred million dollar impact on that business, that's a pretty unique world that I think didn't other businesses weren't designed for. And so we're trying to design for that.
- [21:39]
A
It's really interesting because we were talking about this before, about this new trend of engineers. Talent is scarce, but engineers are increasingly not just looking to the top AI labs and working on the frontier, but they're interested in deployment. So I'm kind of curious if you could talk a little bit about that.
- [21:57]
C
What I hear from engineers all day is what they see is the world just changed. Like some very impressive transformative technology was just invented. And as a result, like the two things that make sense to them for their career are like one, you can go work on that technology. You can be building the models and the applications right above them. You can go work for, you know, go work for Anthropic or you can own, you know, full business challenges end to end. You can own outcomes and go use this technology to create outcomes that weren't possible before. For it's like those are increasingly the two categories that are exciting to engineers. And they look at this sort of middle layer of like, you know, like it used to be really, really exciting to join a SaaS company. It's now really, really scary. Yeah, like it, it makes less sense in this context. Like you wonder if those businesses are going to be okay versus like, you know, will the model sort of subsume some of this stuff?
- [22:49]
A
Yeah, the old death of SaaS. So then how do you guys, I mean, you're not, you're a startup, but you're not a startup. Right. You've got the backing of Anthropic and Blackstone, et cetera. Like are you finding it difficult to acquire the, you know, elite engineers during these poaching wars with the crazy insane comp packages you're seeing from the top labs?
- [23:10]
B
So it's a different talent that we're going after than some of the talent wars you're referring to. You know, the talent wars have generally been about the top researchers and we're going after the top applied AI talent. The key challenge that we have with that is that there isn't enough experienced applied AI talent out there to meet the needs of the market. Right. And in response to that, our strategy here is to hire the best generalists. You don't have to have AI experience and then we try to create the best environment to learn applied AI here on the job. You get to work on AI projects, get hands on experience, you're surrounded by top caliber AI engineers and you can Learn from all of your teammates. We try to build systems and knowledge sharing rituals that allow every team member to learn from every single project project that we're doing and really lean into this as a way of enabling us to hire generalists and help them very quickly become the world's leading applied AI experts to really meet the need because there's no other way to do it. There just simply isn't enough applied AI expertise out there.
- [24:15]
A
And there's also probably increasingly there'll be less generalist expertise. Right. Like when we think about the next generation, they haven't had their hands in enough stuff to be generalists. Right. And they're coming in at a time when they're not going to be learning necessarily the skills you need to like build the blocks up to that. So are there any moves from you guys or anything that you've thought about or anything you've seen to help increase that supply? Like with helping with upscaling? Like for example, a couple weeks ago we had on Pim dewitt from General Intuition and he talked a lot about how his company is, you know, trying to create job opportunities for people who are, who are being displaced by the AI revolution. So I'm kind of curious how you guys think about that.
- [24:56]
C
I'm not sure if I agree with the assertion that there'll be fewer generalists in the world. I think it has never been an easier time to become an entrepreneur. Like AI is making it really easy and increasingly appealing as like a first career path to go. You can go start a company and go try and have an impact on the world. And I think you learn so much by like trying to own problems end to end, like going to try and get product market fit. Move the needle on a business. You learn a lot of stuff there that you don't learn from just solving sort of a narrow problem. And like that's, that's the skill set that fits really well within ODE is these, you know, over half of our team or former founders, like that's not to say that only founders fit in here, but I think it's like a sign that like it's a sign of the types of, you know, sort of like attributes that work well, right? Like broad experiences, broad end to end ownership, sort of hunger to move the needle on something real and see an actual outcome in the world. So I'm actually excited about the sort of supply of entrepreneurship that's going to come into the world. To some of your point though, yes, it is challenging in this environment to build a super elite team of engineers like this. But I don't think actually it's the environment that makes it hard. I think it's always been hard to build really, really talented teams, especially in the face of overwhelming demand. Right. Like we see so much opportunity to grow and scale and like keeping culturally the like our obsession is quality and like keeping that obsession as you have pressure to kind of serve all this demand is really challenging. And it'll be interesting to see like how all the players in the market sort of where they land, you know, in the sort of quality spectrum as we all like kind of face this unlimited demand sitting in front of us
- [26:40]
A
when we think about scaling this like consulting doesn't scale like software. Right. Do you imagine this right now as sort of like a boutique services firm? Like are you, how, how are you thinking about scaling and avoiding becoming just like another labor intensive consulting firm?
- [26:54]
B
I mean the ambition here is very much to become a scaled provider of AI native services that can meet the overwhelming demand of the market. And you know we've, we've, we've always been aimed at taking a very ambitious swing. We've long Talked about building $100 billion plus business in this category. I actually think that in the grand scheme of things that's maybe not even ambitious enough. I think we could potentially. It's pretty easy to imagine this is a trillion dollar company someday if we execute well. The key challenge of the business is how do you go through that phase of hyper growth without losing the emphasis on quality that Eddie was just talking about. We have a key cultural value here of over delivering for our clients. It's something we talk about all the time internally. And I think the key tens that we face every single day is trying to over deliver on that next project is hard. And where, where did the resources for that next over delivery come from? If you just add 300 new people tomorrow and you deploy them all on projects, those projects aren't going to go well. That's not going to be a great experience for those 300 new new people on the team. Not going to be, not going to be a great experience for anybody. And so the, the kind of, the things that we invest in every day are the things that are going to allow us to absorb new folks faster, allow them to become applied AI experts faster, to over deliver on that next project faster. That's a combination of investing in technology, learning and development and other things. But ultimately the ambition here is large. And I think navigating that tension isn't going to be easy. But I think we're up for the challenge.
- [28:32]
C
I think AI really does change some of the laws of physics about what it means to be a services business. First of all, we get tons of efficiency internally by leveraging AI to help us do a lot of our work. It can help us with the oversight of our projects, it can help us with consistency, it can help us learn across projects, it can help us do check ins. And there's lots of operational pieces of our business that get easier with AI. There's lots of reusable building blocks and componentry that can be built to make these engagements go faster. But I also think the biggest Uber is the thing that I was talking about earlier where like the impact you can have in a, in, in inside of a company with a really small team is just crazy high with AI. Like if you look at the really successful previous generation services businesses, I mean there are a number of them, but a lot of them are like the size of a small country and they're doing like deployments that have like hundreds of people on them. I think that the opportunity to at scale be doing four person engagements where those four people can transform a whole business is an AI phenomenon. And so you still need to scale headcount to become large the same way every kind of like every software company has to do that too. But not size of a small country large, just like normal software company large.
- [29:52]
A
Right. So a lot of the process of FTEs is being automated already. Do you imagine that you could create an FDE agent eventually? Like learning from the way that all of your employees help other companies implement AI workflows?
- [30:12]
C
I mean I think we see like sort of the Sidekick version of that all the time internally. You know I talked about Claude Tag before but like you know, Claude lives right, right alongside us inside of our Slack discussions and is very aware of how we do these engagements and it's helping us all the time. I think those capabilities are going to get better and better and better. I think that's just going to create room for even larger outcomes and make us value sort of judgment even more. And so I think it's an accelerant to all this stuff.
- [30:39]
A
Okay, well we are probably just about out of time, but this was really interesting to learn about. Where can our listeners connect with you online if they want to learn more or get in touch or apply to be an FDE?
- [30:50]
B
So our site is o.com, you can go there and apply or you can find us on LinkedIn.
- [30:56]
A
Okay, well thank you again so much for joining us on the show to our listeners. You can find me on LinkedIn as well, or X, and you can find Equity on XM threads at Equity Pod. Talk to you next time. Equity is hosted by TechCrunch senior reporters and produced by Teresa Loconsolo with editing by kel. Subscribe on YouTube or wherever you get your podcasts and find out what's next@techcrunch.com events. Thanks so much for listening and we'll talk to you next time.