00:22:56 · Matthew Fitzpatrick
[00:22:56] S0: Yeah. It's an it's an interesting question. So I think what you're asking in some ways is everything is called an agent at the moment, but a lot of it looks still like traditional software. I I think that is accurate, by the way. I think there's... I've seen a bunch of interesting data on what percentage of, quote, unquote, agents are actually agents, and you see that a huge chunk, like, third share just basic, like, scripts, for example. So so so I think that's been one of the other reasons that adoption has been slower than estimated is actually the the prices of building, like, a task based agent that is actually autonomous is is not something that a lot of folks, like, what I would call, like, a multi agent system where you have an agent that orchestrates numerous single task based agents. A lot of people talk about that. Very few companies I've seen have been able to implement that effectively. And so I think that is is a hard thing to do, and I think that's been one part of it. But I think when this is all... Wait. Let's take five years from now for further along the curve. You will realize that the answer for everything... There there are numerous tools that you might consider. Deterministic software, machine learning, agents, a probabilistic LM, any of those. And and, actually, a lot of solutions are gonna be a mix of those four things, not just one thing. And so you've gotta be very directive about, like, which of those best solves a particular problem you're looking at. There are many cases where an agent might be worse than a software product. And so in that case, you might just wanna build software faster that solves a need. You wanna be purely deterministic. There are many other cases. Like, I think contact centers is probably the cleanest one where an agent is absolutely where you wanna go because actually deterministic flow of, like, an IVR, which is how you do routing in a contact center, is just very bad. And so you're much better off with an agent that can make judgments and respond. But there are other cases where if you want, like, much more, like, deterministic routing, then traditional software is better. And Yeah. I like there is the reason SaaS has come down...
00:24:57 · Matthew Fitzpatrick, Guillaume Moubeche
I I think software is gonna be more and more popular in the next ten years, but it won't be out of the box, like, kind of one one function software like much of the SaaS world has been. It will be... You know, it will allow for a lot more customization. That is the the benefit of many of the cool... New coding tools. [00:25:15] S1: And do you believe, like, customization can be done by AI down the line? Meaning that you would have, like, software that kind of, like, fork themselves into a unique version for each person using it? Or do you feel like this is a... This... There must be, like, a human working and a human in the loop to actually create what fits more of a group of people rather than an individual?
00:25:15 · Guillaume Moubeche, Matthew Fitzpatrick
[00:25:15] S1: And do you believe, like, customization can be done by AI down the line? Meaning that you would have, like, software that kind of, like, fork themselves into a unique version for each person using it? Or do you feel like this is a... This... There must be, like, a human working and a human in the loop to actually create what fits more of a group of people rather than an individual? [00:25:42] S0: I do think there will be variants of that that will that will be AI enabled for sure. Like, I think that the reality is if I'm using a piece of software every day and it's able to see, like, my usage patterns and my flows, then it should adapt to what I'm doing. I think the the challenge is... And people have asked, can that happen today? You have a bit of a kind of a threshold that has to be crossed first, which is you need to get people using the technology first, and and you need to get adoption. And so what I'd say to you, you take an example, let's say there are a 100 people at a company, and you wanna develop a piece of software for them. Even if AI could observe a lot of it, the the process of bringing those people along and getting their feedback and the kind of process of of building something in the way they want it is is incredibly important for adoption. And so I think there still has to be a very material human element to get people excited about using much of this new tech. But I do think down the line, more of it, you will be able to have AI kind of evolve evolve the software dynamically.
00:26:41 · Guillaume Moubeche
[00:26:41] S1: And right now in the in the way you work, because I think at some point you mentioned that if you let your IT department, you know, like, have the the control over AI in in building, you know, like this prototype, Basically, everything gets stuck in pilots. So how, you know, like, would you go about building like an AI project within a company? Like, who should be involved? Like, how how should that be structured? Like, and how do you, like, build everything you've done, you know, so far?
00:27:16 · Matthew Fitzpatrick
[00:27:16] S0: Yeah. So the the MIT study I mentioned earlier that talks about, you know, 5% of project making... AI projects making into production, they they cite in there that the... They look at the companies that make... That bring AI all the way to production less consistently. And one of the main characteristics they cite is line ownership of AI issues. And alongside that, very clearly defined KPIs of performance tied to that. And so, you know, what I what I say about that is it makes a lot of sense. Right? Like, if if you took any... Like, take commercial credit. Let's say you wanted to change the way you do credit underwriting in a bank. You do need the person who's responsible for the underwriting process. You need the teams to actually say, okay. I would move forward with this technology. And so the process by which you get a team mobilized around using technology defines the success criteria, decide... Define the measurement criteria, you do need line ownership of that. And so I think that the organizations I've seen that are doing this really well have a mix... Have a kind of a cross functional Agile pod. Very different... Very similar to actually how, like, Agile worked, you know, eight, ten years ago in the early days of digital, but you want a technology representative, kind of a line owner, usually somebody from, like, the legal department. You want, like, three to four different cross functional owners who can tell you what are things you can do, you can't do. Legal is important to have in there, right, to understand, like, you don't wanna build something that has a ton of regulatory risk or compliance risk. And so having those folks all involved from the beginning, but most importantly in that is a line owner who's gonna lead it day to day as well as somebody from the operations team to make sure that the adoption is is clearly measured.
00:29:02 · Guillaume Moubeche
[00:29:02] S1: Okay. Really interesting. And, you know, like, whenever you were talking about the implementation for the the the e commerce websites of whether or not you wanted to... I mean, how you could predict actually, like, stocks. And and I was I was wondering, like, you leverage their existing data to actually build predictive model. But to do this, do you take, like, your own model or do you also leverage, you know, existing public model or open source model that are basically, like, on the shelf?