Founders in Arms
00:50:04 · Immad Akhund, Rajat Suri
[00:50:04] S0: people don't apply for one... You know, 20 jobs. They apply for 10,000 jobs. Mhmm. So, yeah, there's definitely a filter, but I think it's more done on, like, here is the job profile, and this is the candidate we would be willing to look at. So it's like, you're... You know, if we're expecting a tech or banking background for, like, a compliance role, we will filter on that. I don't I don't think the filters are really making, like, like, case judgments. I should check that. [00:50:34] S1: There's definitely an element of selecting talent that is taste based, but it's very hard to do in the resume stage. You know? It's more like when you know somebody's, like, you know, are they the right thing for you? [00:50:44] S0: Yeah. Only 60% of us is... Only about 30% of our job offers are to people who are inbound. Yeah. Most of it is referral and outbound. Yeah.
00:50:44 · Immad Akhund, Jean-Denis Greze, Rajat Suri, S4
[00:50:44] S0: Yeah. Only 60% of us is... Only about 30% of our job offers are to people who are inbound. Yeah. Most of it is referral and outbound. Yeah. [00:50:57] S2: So just one more. People with Taste usually have built things with Taste before, even if it's just their personal website and whatnot. And so I like to, when I meet somebody, I like to ask, find something they've built and ask them why they built it, what choices they made. And sometimes people have really good answers about things that at first don't look like they have taste, but you realize they made some trade offs so that it is the way that it is, and there are good trade offs, and you're like, that person has a different kind of taste than just it looks good or whatnot. But I think asking people about their thought process behind it. And to your point about AI, it's somewhat AI just thought their way to something that looks really good. I mean, looking really good is not really taste, but I think some people sometimes are blind to that. You ask them questions about it, and they don't have any good answers, that you know they don't have the kind of thoughtfulness that you might want at your company. So I think it's, again, it's resume staff. But, you know, history is always a good indicator, I think, of future [00:52:00] S1: behavior. Yeah. [00:52:04] S4: Couple. Been Mercury customer for a few years. [00:52:26] S2: Yeah. So
00:52:26 · Jean-Denis Greze, Immad Akhund, Rajat Suri
[00:52:26] S2: Yeah. So [00:52:30] S0: the question is, like, how do I think AI and agents will use cards to make payments? You know, I guess, like, one interesting thing, like, a newish behavior is everyone is using kind of Stripe link and doing payments with news and things like that where it's not even creating a virtual card. At least it does. [00:52:53] S1: I don't [00:52:53] S0: think it's creating a virtual card, but it is requiring someone to click a button. I think most people are still at the stage where they wanna click the button to verify a payment. And I think, yeah, [00:53:08] S2: most of
00:53:08 · Jean-Denis Greze, Immad Akhund, Rajat Suri
[00:53:08] S2: most of [00:53:09] S0: the way people are thinking about building these agents is don't have, like, human in the loop on clicking the button. We have an agent account thing, which is a little bit more like you give the you give the card to an agent, but you set limits on, you know, where and how much you can spend, which is another mode to do it. I I really do think people want to spend with their agents. And so, like, I think there'll be... Both of those modes will exist, like, limited use, like, wallet cards or whatever, and, like, click to spend and buy. So probably one of those of either. I'm a little bit skeptical of stablecoin as, like, the way agents pay for things. I just think cards are pretty good. Like, that'll be the most big one. You know, there's, like, some world where, like, agents are spinning up lots of other agents that they need to then pay for, and maybe, like, the agent to agent payment and layer layer is stablecoin. I haven't seen too much of that. And even then, you know, both both those agents have, like, companies behind them, so they could probably just agree to each other with money instead of stablecoins. Yeah. I don't know. [00:54:24] S1: Do you need cards? Can you do, like, little bank transfers? [00:54:29] S0: These cards have, like, a lot of protection, though. Right? Like, the reason people use cards is you can [00:54:34] S2: you can dispute them. You can change card numbers. Like, I kind
00:54:34 · Jean-Denis Greze, Immad Akhund
[00:54:34] S2: you can dispute them. You can change card numbers. Like, I kind [00:54:38] S0: of know... Yeah. Especially in The US. Right? I kind of know this and reading numbers are, like, secure things that you're not supposed to give people. So it's just like a messy process to use them. Maybe in the future.
00:54:52 · Jean-Denis Greze
[00:54:52] S2: I actually think even the problem statement is not really accurate. Like, I think there's, like, fear that people have of agents taking cards and just spending money left and right is mostly nonsensical for for, like, a variety of reasons. Like, first of all, like, just the velocity on the cards will be detected by the fraud of, like, any of the processors. So, like, number one. Number two, it's actually really hard to get the agent to do truly stupid things with money, and it's only gonna get more difficult because all the labs are post training for commerce scenarios. And so it, like, it... You know, it... You give some vague instruction where it thinks it's supposed to buy, like, a 100 Ferraris on your black Amex special, like, you know, it's gonna be like, hey. Hold on a second. You show you a 100 Ferraris? Maybe just one start? Like... And so I I just don't believe that this world where, like, the the AI is going totally crazy, at least with cards. Actually, would say, like, with stablecoin or or ACH, I would be much more worried about it because I think there's much less... There's much... Many fewer signals in the system. I think the one element that is a little worrisome is fraudsters actually doing things with cards because I think there are are less signals around agents than there are around humans. Meaning, like, when a human does something with a stolen card, it's... There are ways across websites and whatnot for it to be known that it might be a human who's both... Who's doing things that are bad. I think with the agents look all kinda the same, and so I think there's some fingerprinting that will need to be figured out and maybe some identity connection. But I think it's, again, it's the fraud problem and bad person problem and not the AI is doing weird thing problem.