00:00Ben, what is the problem that Raindrop is trying to solve?
00:05Look, very simply, I mean, you mentioned it even in the case of Robinhood.
00:08But what we're seeing is that agents are being deployed into increasingly high stake environments.
00:16So we're talking about like medical, finance, defense.
00:19And at the same time, they're kind of being given increasing capabilities, right?
00:26The ability to do more and more autonomously.
00:31And what we're seeing already is that these mistakes are going to be, in the best case, very, very expensive.
00:39And in some cases, either soon, if not already, these cases, these mistakes are going to be a lot more
00:45than just expensive.
00:47And so that's the problem that we solve.
00:49And we think of this a lot.
00:50We use this phrase, you know, humanity's last problem.
00:52But we really do kind of consider this to be humanity's last big problem to solve.
00:57You know, part of me also wonders, Ben, it's like, so we have AI, we've got, you know, LLMs, we've
01:04got agents, and then we've got things to fix the mistakes and agents.
01:07But like, is there an assumption at some point that there will not be any mistakes, that the data is
01:14so deep and wide that things are smart?
01:17Are there always going to be, like, we can't totally trust everything out of AI broadly?
01:25Well, this is really interesting, right?
01:26Right. So if you think about, you know, humans, if you think about, you know, you're managing employees, employees do
01:32still make mistakes, right?
01:35But I think that there's sort of two separate things going on.
01:39And I think one thing that sort of gets lost in a lot of the dialogue, if you're not kind
01:42of in it every single day.
01:43So when you hear labs talk about improving models, like Anthropic or OpenAI, what they're actually mostly talking about is,
01:51you know, increasing the capability ceiling.
01:54So having models solve, you know, new mathematical proofs or, you know, kind of like work autonomously longer, like these
02:01sort of things.
02:02What we really focus on is working with, you know, Fortune 100 companies on raising the floor.
02:08So what is the dumbest thing their agent can do?
02:10And again, this really matters when you're talking about, you know, writing prescriptions, which there are now AI agents that
02:15are authorized to write prescriptions in some states, right?
02:19So raising the floor, what is the dumbest thing they can do?
02:23And the problem is that there's often no objective answer.
02:26So it's not the kind of thing you can just gather a ton of data for.
02:29It's like different people have different preferences, different states, different, you know, there's a lot of kind of a lot
02:35of criteria that defines right or wrong.
02:38How do you use agents in your life?
02:42Well, that's a very good question.
02:44So, you know, engineer by trade, right?
02:46So I think, like most engineers, have shifted to the point where I barely write code by hand, right?
02:53And every step of that software engineering process, there are agents involved, whether that's the writing it, planning it, communicating
03:00to other people, reviewing it, right?
03:02Every single step, there's agents involved.
03:04And then, obviously, as, you know, a co-founder of the company, you know, we touch all sorts of things
03:09from marketing to sales, even legal.
03:12And, you know, there are agents being, you know, involved in every single one of those pillars.
03:16What about outside of work?
03:18And the reason I ask is because a colleague of ours stopped by our desk just before our program.
03:24I will not name him.
03:25You all know him.
03:25But he is talking about using AI agents for everything outside of work and talking about, in fact, actually recovering
03:33some money from tax authorities in a certain jurisdiction right now.
03:37Something the accountant wasn't able to do, but his wife was able to do using Claude Code.
03:41And it's really remarkable because that fight, you know, a human being didn't do that, but the AI agent wasn't,
03:48was able to do it.
03:49How do you use it outside of work?
03:52100%.
03:52So, I think that, first of all, I think that we have not really seen the mainstream adoption yet.
04:01And I think a lot of it is because of these sort of failures, to be very clear.
04:05Because in order for agents to be kind of maximally beneficial, you have to give it access to everything.
04:11It needs access to your email.
04:12It needs access to your calendar.
04:13It needs access, you know, to buy things on your behalf.
04:16Right?
04:17And so, there's very few people, and we saw this sort of with the open claw wave, where people sort
04:21of just give it access to everything.
04:22You know, there were some kind of very catastrophic failures.
04:25As far as how I use it, I'll often, you know, if I'm buying something, whether that's furniture, whether that's,
04:31you know, whatever it is, I will often have, or for example, planning a trip.
04:36I'll actually have, I use this agent called Devin, which is made by this company, Cognition.
04:40Oh, yeah.
04:41And it'll actually, yeah, if you've heard of it, it's really, really fantastic.
04:44Again, not used as much personally yet, but I will have it spin up, you know, dig, do a bunch
04:51of research, and then actually generate a website.
04:54So, I did this for a trip recently.
04:55It generated an entire website I could add as a little home icon to my screen.
04:59This was maybe five minutes total, is the thing.
05:01I'm not talking about, you know, vibe coding for days or something.
05:04Five minutes of a map of all these different locations.
05:07Some I gave it, some I told it to find, you know, where my hotel was, what my schedule should
05:12be.
05:13And so, that's the stuff that really, really excites me.
05:15I should note, our Bloomberg News team reporting in February, that the AI coder Devin has aimed to modernize government
05:21systems.
05:22And just a couple of weeks ago, Cognition did raise $26 billion, sorry, $1 billion at a $26 billion valuation.
05:29So, that's the company that Ben's talking about.
05:31We're speaking with Ben Heilak, the co-founder and CTO of Raindrops.
05:35So, is it a case, Ben, in terms of agents, that we're just going to have a whole library of
05:39different agents that we're using in our life, some at work, some in, like, or how does this play out?
05:44Or is it too early to know, kind of?
05:47It's one of the really big and very, very important questions.
05:50So, for example, how do you separate that personal and work context is going to be a very, very big
05:57question, right?
05:57Because your employer is going to want visibility into how you're using agents, right?
06:04How you're spending their money, especially as, you know, again, one of the trends we're seeing, increasingly so, is that
06:10companies are, they want to understand what they're paying for, right?
06:13Like, this sort of, you know, the last year, you know, companies have been in this phase of just spending
06:20tens of thousands of dollars per month per employee on AI consumption.
06:25And, you know, there's roles like, you know, engineering where maybe that justifies it.
06:29There's other roles where it doesn't.
06:30So, anyway, I don't know, is the answer.
06:36My own personal vision is I think we'll see, we'll definitely start to see some centralization around, you know, like,
06:44specific agents that are the ones you want to trust with more and more of your context.
06:47And there will sort of be task-specific agents that still exist.
06:51That's what I believe.
06:52All right.
06:53Fair enough.
06:53We already know we want you to come back soon because we want to talk more about this.
06:56But we would be remiss, especially in a week where we've got the SpaceX IPO on Friday, not asking you
07:03how you were thinking about this IPO and maybe what you can share with us about your experience there.
07:09Are you going to sell if you still have equity?
07:11Do you have equity?
07:13No, I don't have any equity from back then.
07:18I think SpaceX is one of the most mission-driven companies I've ever experienced.
07:23And I think that that's one thing that really gets lost, I think, when people talk about it.
07:27Like, the people, at least when, you know, I think it's a very different company than when I was there.
07:30When I was there, probably 3,000 employees.
07:33So, it was like 2016.
07:36But one of the most mission-driven companies.
07:38What I mean is that when people go to work every single day, they think about, you know, at that
07:42time going to Mars.
07:43And I know that sounds crazy probably to a lot of people that are listening, but that's really what people
07:46thought about every single day.
07:49So, you know, I know that I'm not, you know, qualified to talk about valuations and all that sort of
07:54things.
07:54But, Ben, I just, we only have a minute left.
07:57Again, we want you to come back.
07:58But the thing is, is that company in 2016 that was mission-driven to go to Mars is very different
08:03than the SpaceX of 2026 that sees a $27 trillion total addressable market where a lot of that has to
08:09do with its XAI business.
08:12Yes.
08:14Yeah, yeah, that's the thing that I wouldn't be able to answer.
08:18That's the thing I wouldn't be able to answer.
08:19I think that what I will say is, like, again, I'm not qualified to speak on any of this sort
08:24of stuff.
08:24But the cool thing about SpaceX is that it kind of has this infinite market cap, right, when you think
08:29of it as a space company, at least.
08:31And, yeah, from there, you know, I have no idea what comes next.
08:34Yeah, it's just kind of fascinating.
08:36Is there anything you can share in terms of Elon?
08:38Like, we cover him so much.
08:41And as many have said, you know, never count Elon out.
08:44Any thoughts there?
08:45Just real quickly.
08:46I would not count Elon out.
08:48You know, I think he's a hard guy to bet against.
08:50I would never bet against him.
08:53You know, there's not that much I can say that hasn't already been said by someone, you know, someone else.
08:58What I will say is that, you know, I think people talk about it all the time, how involved he
09:02is in really not just a figure face, you know, figurehead of the company.
09:07He is involved in engineering decisions.
09:09And, again, at least at that time, it's so long ago, but he really was in meetings and critiquing things.
09:14And that was something that was always very inspiring to me when I thought about founding our company.
09:20So, let's go.
Comments