Scaling Europe
00:06:56 · Nicolas Sharp, Sebastian Johnson
[00:06:56] S1: X number of agents. Sorry, I was going to just add, I think you're completely right there. But if you think about how limited the throughput of these systems was previously by especially the work that's very, that requires lot of, has a lot of unstructured data or what's historically been unstructured data, very high context work. It was really bottlenecked. And we all know that everyone, you know, there's the old kind of... In in the old world, you would have a calendar event in your diary to update the CRM. And that was something that you might do once a week if your manager put a ton of pressure, you know, put a ton of pressure on you. But now, Attio is being updated just continually in a loop. You know, every little signal that comes in is being... We're running inference on it. We're making updates. We're sending out webhooks. You know, we've got external agents acting in the system. There's just the... There's just orders and orders of magnitude more stuff happening in these systems now. [00:07:55] S0: And what does that mean for the traditional go to marketing? You know, I remember for for one of my projects, my first startup, I was I was, like, the sole revenue ops person Yeah. Right, in a small startup. And that was so much of it was, chasing down sales team, filling the CRM. What kind of things or activities are kind of go to market teams now able to do that are not having to do so many of those tasks that can be handed over to agents?
00:08:18 · Nicolas Sharp
[00:08:18] S1: Yeah, mean, is a great question because I think a lot of the first versions of these systems were essentially very skeuomorphic in the philosophy, which was, let's do exactly what we were doing before, but let's make it an agent. So you had this very direct translation. You went from an SDR to an SDR agent. But actually, the idea here isn't just to replace work. That's not what we're trying to do here. We're trying to give a massive amount of leverage. I think that people are always going to be extremely, extremely important in these customer facing roles. It's a fundamental desire that humans have to interact with other humans. And what we're trying to do is give teams leverage so that the work that they do is just a really important work. And it's not all of the busy work of filling in a CRM and doing all of that sort of manual work. Also allowing them to become system architects and build things that just wouldn't be possible because of the constraints of human organization. So let me give you an example of this. You're a company like Attio, let's say, or we could take any of our other customers like Mark or Granola or Modal, etcetera. But you have thousands of sign ups a day. Now, it wouldn't make sense to have every single one of those signups go to a person to go and do a bunch of research, to go and work out whether it's a good fit for the company, whether it's growth potential, etcetera. So you might not do that, Or you might rely on some very narrow data set to look for certain signals. But what you can now do is actually say, no, a new agent abundant mindset. Let's go and do a bunch of work for every single customer that signs up, and let's give them an extraordinary experience regardless of their size and free of constraints. And so you'd now go and build agents that went to go and do research on that company, check if it was a good fit, and and all that kind of stuff. And then, of course, at the right time, get them in touch with with someone on your team. But those are sort of things where you're not just replacing human work, you're augmenting it and doing things that just wouldn't have previously made sense.
00:10:42 · Sebastian Johnson
[00:10:42] S0: It's just like a massive enhancement in, I guess, the end user experience. Yeah. And I I wanna talk about what that user experience kind of looks like in the future because you've got... You've you've very intentionally pushed Attio into places like Claude and Notion so that people are sort of interacting with the architecture inside of applications. Yeah. Do you see a future of Attio where it becomes more infrastructure architecture that is embedded in the applications that people use every day? And do you think that's potentially symbolic of all software? Will we just interact with it in other systems?
00:11:17 · Nicolas Sharp
[00:11:17] S1: I think to a large extent, yes. And so I think that, you know, those see that I see this in my own usage, of course, that there is definitely massive utility to being able to use tools from a single place. Actually, especially when you want to do work across tools, that's really helpful. Now, it's difficult for those unified tools to sometimes get the results that you get from a dedicated tool. So as an example, we've done some benchmarking for Ask Attio. And there are significant advantages to using Ask Attio over MCP. Now, those advantages mostly come down to our ability to use the right model and sub agents for very specific tasks that we know are optimized for those sub agents and those prompts and for us to know exactly how to interact with the vector store and those sorts of things. And so you definitely see that specific harnesses can give you great utility over generic ones. And where we really see the difference is in speed and cost. So the benchmarks that we're releasing will show essentially that Ask Attio versus just using ChatGPT with the same model or the same orchestrating model, I should say, will return roughly similar results because fundamentally it is the same model underneath. But the harness will basically mean about a 6x time reduction and therefore cost reduction. And that's pretty significant. But equally, we want to meet the customer where the customer is. And so if the customer wants to do specific work in Claude or ChatGPT, that's great. We'll continue to make sure that our MCP is as good as it can be. And at the same time, we'll make sure that our own agents are as good as they can be too.
00:13:17 · Sebastian Johnson
[00:13:17] S0: We're living in this world where people talk about moats a lot and software can be replicated easily. You're obviously making this big bet on infrastructure architecture. You've also spoken a lot about universal context. Can you explain what that is and why that is so important?
00:13:33 · Nicolas Sharp
[00:13:33] S1: Yeah, absolutely. So I think I'll answer universal context first, then we can talk about modes. Universal context is essentially our is our is our is our vector store. And what it does is it gives a... You know, it gives the the sort of dream of CRM that we've always had for CRM, which is a true kind of understanding of a customer, a three sixty view of the customer. It essentially is... It it ties all of the activity that you have, all of the calls, all of the emails, all of those things, customer support tickets, etcetera, onto entities, and then makes that queryable by agents. Now that could be you using your own... It could be using MCP, it could be an agent that's customer facing, etcetera. But just having all of that context with vector search is pretty awesome. Now, in regards to moats, I think that tools, software that provides, that just access a tool is going be a pretty difficult thing to defend even in the short term. But I think that where the real value will lie is in these systems that facilitate large scale data governance and and all of the policies around how you want... What what do you want your customer experience to be? How do you want these things to interact with your customer? And how do you make sure that it's a dangerous thing? You really don't want to mess this stuff up. And that's where I think systems of record and these large workflow tools will continue to do really well. And actually, because of the scale that we're going to be operating at and because the amount of agents people are going have running, I think that they're going to become more and more important. So I wouldn't want to be a tool builder, but a platform builder that's delivering end results and doing the work, I think, is a good spot to be.