Coveo VP Sébastien Paquet on MCP, Agent Pitfalls, and Real-World Enterprise AI Applications

Hey everyone. Alan Shimmel back here on Techstrong tv. My next guest is Sebastian Packet, SE packet.

Yes. I hope I got that right, Sebastian. Yes.

Uh, Sebastian is VP of AI strategy at Coveo. Welcome to Tech Drunk tv. Sebastian, it's great to have you on here.

Thanks, Ireland. Great to be here with you. Yep.

So, Sebastian, even as big as AI is today, and it's all the rage, I don't meet a lot of VP of AI strategies, and I'm not sure many people in our audience do either. Give us an idea what, what, what exactly does it mean to be that you're a VP of AI strategy and how did you get this job? Hey, uh, it started, I had the beginning of, uh, of this 2000, uh, so I got my PhD in AI actually in 2006.

Uh, then I worked for like eight years in a more consultant role. Um, and I joined Coveo in 2014, um, where I, we started to deploy to the cloud. We had our, uh, usage analytics and all of all, uh, all this new data.

So I started a machine learning group at Coveo to train on, uh, on the data we were gathering. And I started the team at Coveo, built it from scratch, started developing some ML models, uh, that were there to improve search relevancy, um, and, and from there grow, uh, from an applied scientist to a director to a v vice president at Coveo. Um, and now my, now my role is really to define the AI vision for Coveo and more importantly, recently where we position ourselves because this market is now really crowded as, as you know.

Yes, it is. It is. That's a great, that's a, a, a great story.

And, um, you know, we didn't mention, let, let's dig into Coveo a little bit more. If it's okay with you, Sebastian. First of all, let's start with the website.

What's the URL? com. com.

VEO. Just the way it looks on the screen here, just it's spelled on the screen. Um, give us, if you can, for people who haven't heard of Coveo, I mean, you gave us a little bit here, but give us a little bit more depth, if you will, about, you know, what, what Coveo does, solutions problem solving, how to engage, stuff like that.

Yeah, sure. Um, so, uh, for the last 20 years, so a lot, uh, uh, way before the ai, uh, hype, uh, was, uh, already a market and industry leader into enterprise search. So we've been doing enterprise search for two decades now.

Um, and what we provide is a relevance platform where we can index the, uh, unstructured data that an enterprise as within all their enterprise silos. So most enterprises have hundreds of applications they're using. Their data is scattered all around, um, the place, so COO can index it.

What's really, uh, different, uh, differentiator for COO is that at the same time that we index the content, we also index the access rights so that we can provide a unified, uh, secured search for our customers. Um, we have big customers in, uh, commerce where, uh, we provide the product search and discovery. So when you go shopping on some of our customer's websites, it'll be cover providing the search results.

Um, we also have, uh, a knowledge line of business where, uh, for, uh, self service and, um, the, uh, workplace environment, we can provide a search and question answering and, and case deflection services. So this is where people go on a support website, they ask a question because something is broken, something is not working as expected, or they just need more information about the product data spot, and we can provide direct answers to them. Um, and if they go into submitting a case, we even follow what they are asking for and we can, uh, answer them directly so they get the, the end user get their answer faster.

And for our customers, they, they don't have to pick up their phone. So it's, uh, it's easier for them. It's what we call the case deflection.

So the user is happy because he found the information by himself more rapidly, and the company is happy also because, uh, it reduces their support cost. Um, so in terms of generative AI and, and, uh, what we provide is a fully managed solution for question answering. So something that can be a service that can be deployed in less than than a day.

Uh, you, you can have a generative answering just like what you see right now on Google. When you ask a question, you have the answer right below the, the search box. This is exactly what Kvi offers, but not for the internet, for each enterprises.

So on our customer's website, when people search on their websites, they can get this slide, this same, um, generative answering, uh, capability. And for agents, uh, or any generative AI applications, we provide APIs so that, uh, our customers or SI partners can build solution on top of Coveo. So if I summarize, I think Coveo is really the information access layer for the enterprise knowledge, enterprise information, so that other applications like search, like, uh, generative solutions, like agents can have access to the enterprise knowledge in a secure manner.

Uh, and, and it's also really scalable and can scale to, to any number of users or agents. Love it. Thank you for that.

I appreciate it. All right. So let's start to some recent news.

Cove Coveo recently announced you guys launched an MCP server, and that this MCP server is available in the, uh, agent force from Salesforce, right? Um, this is where they're building all of their agent functionality, and it's not just Salesforce's agents. It's, it's kind of like a marketplace for a lot of people's agent technology.

Um, why Agent Force Sebastian? And, you know, you know, I, I think, I think our audience knows what an MCP server is. Let's say some people out here don't know what an MCP server is.

Why don't we start there and then tell us why Agent Force? Okay. So, uh, sure.

Um, so MCP is for model context protocol. So it's an open standard, uh, that was released actually less than a year ago, and so late 2004, 2024, uh, by, uh, entropic Andro and yeah. Yeah.

And it's becoming a standard, but it's really nascent. Uh, so it's started to really pick up in, uh, the beginning of the year. Uh, and the, uh, what it allows is to, uh, give a standard protocol for agents to get access to external tools, data or context.

So it allows agents to communicate with external systems, um, and CVO being one, uh, providing search services, uh, for the agent. So an agent can call a CO OMCP server to, uh, access the search capability and get back the information from the enterprise. Um, what Salesforce announced last week was that Agent Force was now supporting MCP, uh, servers, or will support it really soon.

Um, and CO is one of the first, uh, initial partners that, uh, Salesforce announced with. Um, we already provide, uh, what they called an agent force action, which was the, the previous way to connect an agent force agent to caveo. So it's already possible to use an agent force action to connect an agent force agent to the overall enterprise knowledge.

So things that are in Salesforce or outside Salesforce, um, and provide this information within the agent force agent. What they announce is the MCP support, and kvi will work with Salesforce to provide it through their, their, um, agent marketplace. Um, excellent.

And what's, what's really interesting for Coveo is being this, uh, um, information access layer and, and why Agent Force, because Salesforce is a big ecosystem where it's a big partner of coveo. We have already a lot of, uh, common customer, um, and, uh, we were already providing for, uh, for Salesforce agent or Salesforce self service portals, um, search capabilities and question answering capabilities. Um, and, um, but Coveo also, uh, is providing connectivity to Salesforce, but it's also providing connectivity to other agent tech platforms.

So like Microsoft copilot, like Amazon Bedrock, like, so all the big players out there add their agent tech platforms. And, uh, Coro really wants to position itself as the, in, as the agnostic information access layer for an enterprise so that all these agents can have access to the same information, uh, really easily. Um, so that, uh, for an enterprise, most enterprises actually use many of these platforms.

So all big enterprises will have a Salesforce, uh, implementation, Microsoft one and others, uh, and SAP one. So, and we will connect you to these different, uh, ecosystems and provide, uh, the enterprise knowledge to all of them. Sebastian, this I think is very helpful to our audience.

I've got one more line of questioning for you. So you're right. Uh, MCP very quickly has become a defacto standard.

I, I did see recently the Linux Foundation came out with like this a to a project. The agent to agent project Google, uh, contributed some open source, uh, or open source and contributed some, some software to it. Some of the other big players involved, you know, some of the biggest names in tech, but certainly MCP has quickly become the standard, as you said, come, comes from philanthropic.

Seems like there's a lot of MCP servers out there. There's probably multiple MCP servers available on Agent Force even. What, what's special about the Coveo one versus other MCP servers?

Um, I think right now, uh, every enterprises that have an APIs will provide MCP servers to, to wrap their APIs and make them accessible to agents. So I think what makes the CO MCP server different is not necessarily the MCP server itself, but it's the capabilities that COO can provide to our customers. And this is everything I thought about before serving, uh, secured enterprise search, and where you can have access to all your enterprise knowledge, not only the knowledge within a specific ecosystem like Salesforce or Microsoft.

Um, so I think this is this, uh, this broad access to, or, uh, broad access to the information within enterprise is really the strength of coveo. Um, so we do have connectors to connect to all the enterprise systems. We provide one access point, which is, which is any unified index to access all this information.

And this is way better than what we've seen out there. So, as you said, there's many multiple, there's many, uh, MCP servers and all systems have theirs MCP server, but if you want your agent to actually have access to, I don't know, 10 data sources, it's way easier to contact one MCP server at Coveo than to contact the 10 different sources and ask then your agent to figure out which information is the right one for, uh, answering any, uh, customer or any user question. So, um, I see this as the same trend that was, uh, 15 to 20 years ago about federated search be, uh, and unified search.

So if you contact 10 different sources, it's really hard to come up with, uh, a good relevance, uh, for all. There's these different sources. Having all of this within a unified index is, is really valuable, and you get really, uh, more relevant results out of it That's available on Agent Force itself.

com though, you can probably grab it there as well, right? Uh, yes. We'll, of course, uh, ize, uh, um, where to get access to it.

And MCP is really, uh, interesting. It's really new. Um, they just released, uh, in June 18, uh, the specification for, uh, identification and secure connectivity.

So, uh, this is really a, a stepping stone, making it, uh, available, uh, making it, uh, usable within an enterprise setup. Um, so you were saying, when, when is it gonna get the ground running? Uh, I think we will get there.

Uh, currently it's, it's still a lot of experimentations company trying it out. Um, uh, so we provide a code, they, people can, enterprises can deploy them, can try it out. Uh, but, uh, now that we have, uh, uh, uh, identification and security protocol, uh, and now it's gonna be, uh, more enterprise ready, so, uh, I think that absolutely this was a missing pieces.

Absolutely. Sebastian, thank you so much for coming here on Text Drunk tv. Good luck to you and Coveo.

You know, this, obviously, this whole NPC and AG agentic AI thing is, you know, evolving day by day, minute by minute, hour by hour. So we'll be watching come back again and us posted. Okay.

Thanks, Alan. Thank you. Sebastian Pak, VP of AI strategy at Coveo here on Tech Trunk tv.

We're gonna take a break. We'll be back with some more.