WEBVTT
00:00:00.160 --> 00:00:08.800
There's a lot of like nuance around like if agents are not humans and they're not machines, then what is that identity standard look like in the future.
00:00:15.439 --> 00:00:22.399
Welcome to Partner Path, a podcast that unpacks the venture capital and growth equity ecosystem from a junior perspective.
00:00:22.559 --> 00:00:28.960
Young entrepreneurs and investors have already had a massive impact on the industry, having started unicorns and launched billion-dollar funds.
00:00:29.280 --> 00:00:35.679
We discuss these success stories and more by sharing perspectives and advice from some of the industry's most prominent role models.
00:00:35.840 --> 00:00:42.240
This podcast expresses our views as of the date published and does not represent the views of and is not endorsed by any company for which we work.
00:00:44.719 --> 00:00:49.759
Before we get to the episode, Laura and I are excited to announce a partnership with Overlap.
00:00:50.240 --> 00:00:59.759
Fresh Out of Y Combinator's Summer Badge, Overlap is an AI-driven app that uses large language models to curate the best moments from podcast episodes.
00:01:00.079 --> 00:01:11.920
Imagine having a smart assistant who reads through every podcast transcript, finds the best parts or parts most relevant to your search, and strings them together to form a new curated stream of content.
00:01:12.239 --> 00:01:14.079
That is what Overlap does.
00:01:14.560 --> 00:01:22.319
I personally use Overlap to explore content on future podcast guests and highly recommend it as another diligence tool for investors.
00:01:22.879 --> 00:01:24.959
Check out Overlap on the App Store today.
00:01:25.359 --> 00:01:28.719
Today we're chatting with Daniel Mason, co-founder and CEO at Anon.
00:01:28.959 --> 00:01:37.040
Daniel previously founded Spring Labs, a platform that builds data privacy technology to power use cases across financial services, consumer identity, and crypto.
00:01:37.200 --> 00:01:40.719
And now Daniel's focus on building the identity layer for agents.
00:01:40.879 --> 00:01:41.359
Welcome.
00:01:42.000 --> 00:01:43.120
Yeah, it's good to be here.
00:01:43.359 --> 00:01:44.400
Excited to chat today.
00:01:44.719 --> 00:01:50.000
After graduation, you worked in a variety of product manager roles before starting sabbatical co-working.
00:01:50.239 --> 00:01:58.799
What exactly sparked your interest in becoming an entrepreneur and going into the coworking space after seeing some of the hardships that WeWork has faced?
00:01:59.280 --> 00:02:04.000
I feel like my journey to entrepreneurship was like a pretty easy transition, actually.
00:02:04.159 --> 00:02:07.040
I actually didn't work anywhere for very long.
00:02:07.200 --> 00:02:14.080
Like I had a job right out of a college at a startup, which is pretty rare when I was graduating college, like much more common now.
00:02:14.240 --> 00:02:20.560
But I happened to have a friend who started a company that I joined, taught myself to code, was an engineer there in the very early days.
00:02:20.879 --> 00:02:25.759
And we built marketing software and we worked with companies that were doing like advertising campaigns.
00:02:25.840 --> 00:02:27.520
It was kind of like an ad tech company.
00:02:27.759 --> 00:02:32.479
And occasionally companies would come to us and ask if we could help them with professional services.
00:02:32.639 --> 00:02:36.080
And we would do that for engagements that were above a certain size.
00:02:36.400 --> 00:02:40.719
And I was also a BD guy in addition to an engineer and source deals.
00:02:40.800 --> 00:02:45.759
And sometimes the company would say no and I would say yes to those deals and do them on the nights and weekends.
00:02:46.000 --> 00:02:52.800
And within a couple years, and I had another job during this time too at Red Hat as a product manager.
00:02:52.960 --> 00:03:01.599
But during this time, at some point I looked up and was making more money from my night and weekend side job than I was from my main W-2 paycheck.
00:03:01.759 --> 00:03:05.039
So I decided to start working for myself exclusively.
00:03:05.120 --> 00:03:07.360
And that was kind of the start of the entrepreneurial journey.
00:03:07.599 --> 00:03:09.039
And that was Spring Labs?
00:03:09.360 --> 00:03:11.199
No, this was way before Spring Labs.
00:03:11.439 --> 00:03:13.120
This was when I was just out of college.
00:03:13.360 --> 00:03:15.919
And then, yeah, you mentioned sabbatical co-working.
00:03:16.000 --> 00:03:19.199
I mean, my that was actually a business I started with my brother.
00:03:19.439 --> 00:03:28.879
He, my brother, worked in real estate private equity, and we both had this vision of remote work as kind of the future of work or at least some sort of hybrid work, which was really not the case then.
00:03:28.960 --> 00:03:31.199
This was like 2013 or something.
00:03:31.520 --> 00:03:37.280
And so we decided we wanted to build a space for digital nomads to go travel and work.
00:03:37.520 --> 00:03:39.360
So we moved to Puerto Rico together.
00:03:39.520 --> 00:03:45.120
We master leased a hotel, tried to get a couple more hotels, raised some money for it.
00:03:45.360 --> 00:03:48.159
And yeah, it was a catastrophic disaster.
00:03:48.479 --> 00:03:54.240
We can go more into if you want, but it was a fun foray into a totally different type of entrepreneurship.
00:03:54.400 --> 00:04:08.240
And through those first two experiences, I think I realized I don't want to build large-scale services businesses, and I don't want to build non-tech businesses, especially in other countries where it's a lot harder to be an entrepreneur.
00:04:08.400 --> 00:04:12.479
So yeah, I was kind of set on building tech venture-back tech companies after that.
00:04:12.719 --> 00:04:13.520
Okay, that's fair.
00:04:13.759 --> 00:04:16.480
Do you think the bottleneck was more on the former or the latter?
00:04:16.800 --> 00:04:17.920
Of those two types of company.
00:04:18.319 --> 00:04:23.759
With the services business, like with my consulting business I ran, it actually got decently sized over time.
00:04:23.920 --> 00:04:27.199
It was just always boring, is like my only complaint about it.
00:04:27.279 --> 00:04:31.120
It was like good cash flow, a consistent customer base.
00:04:31.279 --> 00:04:33.360
We had like a service that people wanted.
00:04:33.519 --> 00:04:38.480
We did a lot of marketing and data science consulting work was like what we found our specialty.
00:04:38.720 --> 00:04:42.639
And yeah, it was just like, I don't know, I woke up every day and wasn't excited by it.
00:04:42.720 --> 00:04:48.480
So after a few years of doing it, even though it was like a great economic opportunity, I decided I wanted to go do something else.
00:04:48.720 --> 00:04:50.560
And that's when you went to Spring Labs.
00:04:50.720 --> 00:04:51.439
Yes, exactly.
00:04:51.680 --> 00:04:51.920
Okay.
00:04:52.079 --> 00:04:53.839
And maybe break down how you got the idea.
00:04:54.000 --> 00:04:58.000
I mean, you raised a lot of venture money, which is very impressive and hard to do.
00:04:58.240 --> 00:05:01.120
But talk about like the original thesis and how it evolved.
00:05:01.439 --> 00:05:06.879
So Spring Labs was a company I started in 2017 with a few co-founders.
00:05:07.040 --> 00:05:15.199
I was, it was a very lucky break for me in some ways because my co-founders were very seasoned entrepreneurs who had built businesses together previously.
00:05:15.360 --> 00:05:26.959
I was kind of like the last guy to join in, like the other, there were four of us, the other three businesses together in the past, including some like very large businesses that are now once now publicly traded and some really serious companies.
00:05:27.120 --> 00:05:37.279
And the original vision for the business was that we wanted to build a new credit bureau, which is kind of one of those tar pit ideas you hear a lot of startups pitch and then eventually pivot away from.
00:05:37.439 --> 00:05:40.000
That was kind of our trajectory on it, too.
00:05:40.160 --> 00:05:46.639
But I think we thought we had a lot of connections in the financial services space, and a lot of our early customers were fintech businesses.
00:05:46.800 --> 00:05:50.399
And the business, it's a multi-product business now that does a bunch of different things.
00:05:50.560 --> 00:05:59.360
But the first product that we built was really like a data storage tokenized vault type product, which was kind of what a credit bureau would do anyway.
00:05:59.439 --> 00:06:03.519
And we ended up working with credit bureaus instead of selling, instead of becoming a credit bureau.
00:06:03.600 --> 00:06:06.240
Transunion eventually became a big customer of ours.
00:06:06.399 --> 00:06:10.800
They also were one of our largest investors and led our Series B round.
00:06:10.959 --> 00:06:15.680
So we ended up working in the credit bureau space, just not in the exact original version of the idea.
00:06:16.000 --> 00:06:20.399
Did you take any lessons from your first in entrepreneurship into Spring Labs?
00:06:20.639 --> 00:06:23.279
I honestly was kind of in the passenger seat for that one.
00:06:23.439 --> 00:06:26.800
I was the one learning from my other co-founders in that business.
00:06:27.040 --> 00:06:29.360
I think that I learned from my earlier businesses.
00:06:29.439 --> 00:06:31.759
When you build a services business, you learn how to sell.
00:06:31.920 --> 00:06:33.680
So I did a lot of our selling.
00:06:33.759 --> 00:06:41.600
I ran product in the early days of the company, but for most of my time there, I ran our sales and marketing and growth operations.
00:06:42.000 --> 00:06:47.360
And I think that I learned to sell and selling my own services for years, kind of in that side business.
00:06:47.519 --> 00:06:51.279
And that's made me a better salesperson and kind of builder of sales teams.
00:06:51.439 --> 00:06:57.360
But other than that, on fundraising, on product, on engineering, I learned so much from my co-owners in that business.
00:06:57.600 --> 00:07:01.040
I think one of the best skill sets is someone who's technical that can sell.
00:07:01.199 --> 00:07:01.360
Yeah.
00:07:01.839 --> 00:07:02.560
It's hard to come by.
00:07:02.800 --> 00:07:03.360
I think so.
00:07:03.519 --> 00:07:07.680
And and I'm by by like proper technical standards, I'm not technical.
00:07:07.759 --> 00:07:08.800
Like I'm code.
00:07:09.120 --> 00:07:10.319
Yeah, I code from time to time.
00:07:10.480 --> 00:07:13.040
I code a lot more these days with AI tools, actually.
00:07:13.279 --> 00:07:18.560
I feel like my GitHub finally has some green on it after a few years of being pretty, pretty dark.
00:07:18.879 --> 00:07:20.560
Are you so you're using a cursor?
00:07:20.879 --> 00:07:21.120
Yeah.
00:07:21.360 --> 00:07:26.079
Do you think you have to have a base skill set nowadays, technical base to build a company?
00:07:26.480 --> 00:07:28.560
I think it depends on the type of company.
00:07:28.800 --> 00:07:34.959
I feel that every company has some things that are its differentiator.
00:07:35.040 --> 00:07:37.600
It's like, what do you have to get right to build this company?
00:07:37.759 --> 00:07:44.319
And I think in the in like a post-AI world, a lot of times you just have to get distribution or something to build the company.
00:07:44.480 --> 00:07:48.000
The actual product is somewhat simple as, you know, I don't want to say a wrapper.
00:07:48.240 --> 00:07:56.480
I think that term's like way overused, but there's not as much technical, like some companies, and on, I would say like we have a lot that we have to build technically.
00:07:56.560 --> 00:08:00.800
Like we have like risk and building on the technical side as well as the distribution side.
00:08:00.959 --> 00:08:12.160
But I think there's a lot of businesses it's like a better product for, you know, accounting firms or for like some like large-scale business where what you really have to get is distribution on top of the product.
00:08:12.480 --> 00:08:16.319
Maybe we should touch on Anon and how you got to Anon and what is Anon.
00:08:16.879 --> 00:08:18.079
Yeah, absolutely.
00:08:18.319 --> 00:08:20.879
So we started Anon about two years ago.
00:08:21.120 --> 00:08:24.160
So I left Spring Labs a while after our series B round.
00:08:24.240 --> 00:08:26.240
I'd sort of like worked my way out of a job.
00:08:26.399 --> 00:08:36.159
I was a the chief strategy officer, I think was my last role there after we hired a chief revenue officer who had sold a lot larger contracts and stuff than me in like a past life.
00:08:36.399 --> 00:08:37.919
So eventually left.
00:08:38.080 --> 00:08:42.639
I actually worked in venture for a year at a fund called Framework Ventures that I knew the founders of.
00:08:42.879 --> 00:08:45.679
Did to give you a different perspective, being on the other side of the table?
00:08:46.080 --> 00:08:46.879
Yeah, it did.
00:08:46.960 --> 00:08:47.360
It did.
00:08:47.519 --> 00:09:01.360
I think it made me understand fundraising more and it made me think more about markets and sizing and a lot of the meta aspects of companies, I think that when you're in the trenches day-to-day, you just don't think as much about.
00:09:01.679 --> 00:09:08.720
But yeah, and I kind of thought I would do that for a while, investing and was honestly pretty burnt out on startups after Spring Labs.
00:09:08.879 --> 00:09:10.240
Just like it's a it's a journey.
00:09:10.320 --> 00:09:16.240
I think anyone who's ever started a company, like by year five or whatever that I was in, like you're pretty tired.
00:09:16.399 --> 00:09:16.799
I don't know.
00:09:16.879 --> 00:09:21.279
I just was really fascinated by some of the things going on in the AI space.
00:09:21.440 --> 00:09:43.519
And I think at this intersection of building companies in identity and security and this advent of new AI technology, we got really obsessed with this, like what we'll call like identity layer of agents or how agents use human identities, use like machine identities, how they get access to accounts, because a lot of the things that you want your agent to do, they need to use your account for.
00:09:43.600 --> 00:09:46.639
So I think it's a really important kind of primitive in the broader stack.
00:09:46.960 --> 00:09:50.399
But were you not seeing Okta, Ottsier, their product?
00:09:50.559 --> 00:09:52.559
Were they not doing it well today?
00:09:52.960 --> 00:09:59.440
I think that it's it's actually a very nuanced concept that's still kind of playing out over time.
00:09:59.679 --> 00:10:01.519
I actually get a lot of outreach.
00:10:01.679 --> 00:10:10.879
It's funny from a lot of people on different, especially investors actually, that are like, oh, I have a thesis on agent identity, and I'm always like, okay, what's your thesis?
00:10:11.120 --> 00:10:16.720
And I just find that people haven't really thought it through that much in a lot of instances.
00:10:16.960 --> 00:10:23.840
What I would say is that you almost today have this bipolar idea of what identity is.
00:10:24.000 --> 00:10:29.919
Like identity is either human identity, which is an email password, a magic lake, pass keys.
00:10:30.000 --> 00:10:33.279
It's like something that is an authentication mechanism for humans.
00:10:33.519 --> 00:10:37.759
And on the other side, you have machine identity, right?
00:10:38.000 --> 00:10:45.840
You have API keys, you have OAuth, you have ways of authorizing an application to basically act on behalf of you.
00:10:46.159 --> 00:10:48.720
And agents have to be one of those two things.
00:10:48.879 --> 00:11:07.840
So when you look at the rollout of different types of technologies, you see that agents are either using human identities and that the agents collecting an email and password, authenticating the user, and then letting it take actions on your behalf, which is usually using computer use or sometimes it's using reverse engineering.
00:11:07.919 --> 00:11:10.000
It's like usually when there's not an API route.
00:11:10.159 --> 00:11:14.399
And when there is an API route, OAuth is still the dominant authentication mechanism.
00:11:14.559 --> 00:11:21.120
And OAuth is designed for delegation of human identity to a machine, which actually fits an agent pretty well.
00:11:21.279 --> 00:11:29.919
But there's a lot of like nuance around like if agents are not humans and they're not machines, then what does that identity standard look like in the future?
00:11:30.000 --> 00:11:40.159
And I think you are seeing people like Okta did a release actually very recently that something like they called like non-human identity or something, which is like kind of a blend of the two identities.
00:11:40.399 --> 00:11:41.440
I need to dig more into it.
00:11:41.679 --> 00:11:43.519
Or invested in a non-human identity company.
00:11:43.679 --> 00:11:44.000
Yeah.
00:11:44.159 --> 00:11:44.399
Yeah.
00:11:44.639 --> 00:11:49.279
I think they're going to start doing well because Agent is because you could consider it a non-human.
00:11:49.440 --> 00:11:57.919
So perhaps if there's more breaches, then folks are going to be turning to not the traditional identity players, which is a very solved place, but the non-human on the security side.
00:11:58.240 --> 00:11:59.919
But you're more on the developer side.
00:12:00.159 --> 00:12:00.399
Yeah.
00:12:00.559 --> 00:12:00.720
Right.
00:12:00.960 --> 00:12:02.720
How are you approaching the problem differently?
00:12:02.879 --> 00:12:05.279
What does it look like to work or to use a non?
00:12:05.600 --> 00:12:12.960
I would say that like our product lives, I almost would put authentication in in two camps.
00:12:13.039 --> 00:12:16.879
There's like defensive authentication and more like offensive authentication.
00:12:17.039 --> 00:12:21.200
I don't know if that's I've never heard anyone else use that, but that's kind of how I think about it.
00:12:21.519 --> 00:12:26.799
And we're specifically in the business of giving agents access to tools.
00:12:27.120 --> 00:12:41.919
So, you know, like a better comp for us would be businesses like plaid that have been built in the past that were building automation on top of like some sort of workflow where there weren't APIs or there weren't easy ways of building that automation.
00:12:42.240 --> 00:12:51.360
We're not in the business of like we work for the agent companies, we don't work for the services that are furnishing access to the agent companies, if that makes any sense.
00:12:51.440 --> 00:12:54.559
Like AUK is objectively on the other side, I would say, of this.
00:12:55.679 --> 00:12:57.840
Now, a quick word from our sponsor.
00:12:58.480 --> 00:13:04.720
Laura and I are thrilled to announce a partnership with Grata, an intuitive deal sourcing and diligence platform.
00:13:05.120 --> 00:13:06.960
Keep missing out on private market deals.
00:13:07.279 --> 00:13:09.120
Ditch the Rolodex and get Grata.
00:13:09.519 --> 00:13:17.759
Grata's AI-powered platform provides accurate data, deep insights, and time-saving workflows that help get more deals done.
00:13:18.000 --> 00:13:23.120
And it's trusted by over 500 firms and 3,000 deal makers, including myself.
00:13:23.440 --> 00:13:33.039
I've personally used Grata in both my deal sourcing and diligence workflows and would recommend their tool to any investor across private equity, growth equity, corpdev, or investment banking.
00:13:33.360 --> 00:13:36.080
We've negotiated a special offer for our audience.
00:13:45.919 --> 00:13:47.440
Now back to the episode.
00:13:48.000 --> 00:13:53.279
How do you handle the services or the applications where you can't integrate with today?
00:13:53.600 --> 00:13:55.039
But they're asking you for that.
00:13:55.360 --> 00:13:55.600
Yeah.
00:13:55.840 --> 00:13:57.759
Yeah, it's actually really interesting.
00:13:58.000 --> 00:14:16.720
We chose this specific type of business to work with because I think if you divide it into the offensive and defensive authentication, and then things where there are good routes with APIs and things where there aren't good routes with APIs, I think that anything on the API side, I'm curious to learn more.
00:14:16.879 --> 00:14:21.120
I actually don't know a ton about startups building in this like non-human identity space.
00:14:21.440 --> 00:14:30.720
But you know, there's this actually this interesting blog that I've sent to a bunch of people that says like agent auth is just OAuth or something is the title of the article.
00:14:30.879 --> 00:14:40.480
And it's it's basically making the case that for things where APIs exist is 96% good for actually being agent identity.
00:14:40.559 --> 00:14:41.759
Like it is missing a few things.
00:14:41.919 --> 00:14:49.360
I think there's been some like interesting posts about it missing uh rate limits and time, like you can't time bound OAuth and things.
00:14:49.440 --> 00:14:58.480
But the way, I don't know, the same way you let Calendly use your Google Calendar to book times, it's actually really easy to let an agent use your Google Calendar to send emails.
00:14:58.720 --> 00:15:04.960
And it, you know, there could be more protections put in place for the non-determinism of LLMs, but it's actually pretty good.
00:15:05.120 --> 00:15:12.320
We like the space we're in because we think the internet is like a big, vast place, and there's a lot of automations that you can't do today.
00:15:12.399 --> 00:15:27.759
And we think a lot of companies struggle with that, and a huge amount of the services economy and a lot of the messy things, the rooms full of people that are all doing the same clicking and repetitive actions, or the, you know, follow these 16 steps to integrate this new tool.
00:15:27.919 --> 00:15:32.240
We think that's where there's a massive amount of opportunity, and there aren't APIs there.
00:15:32.320 --> 00:15:40.639
And in the world of APIs not existing, there's actually a massive amount of room for how identity is solved in that space, and that's the space that we're really focused on.
00:15:40.960 --> 00:15:41.600
That's interesting.
00:15:41.759 --> 00:15:48.159
Have you seen any like unique automation use cases today where a non has become a critical part of the workflow?
00:15:48.480 --> 00:15:49.200
Totally.
00:15:49.440 --> 00:15:55.039
We have sort of two products, I would say, that people use us for in in different ways today.
00:15:55.360 --> 00:16:00.720
One is kind of our original product that we launched, which is almost just the identity layer.
00:16:00.799 --> 00:16:11.039
It's like we collect credentials, we handle authentication, and we authenticate users, a lot of startups, some early agent companies, a lot of things you saw on Twitter probably last year were using that product.
00:16:11.279 --> 00:16:26.559
We have recently, though, we've gotten much more interested in building much deeper into the stack and actually building the agents, building the authentication, building more of the infrastructure and the actual capturing more of the business value ourselves and building more for specific use cases.
00:16:26.799 --> 00:16:30.399
And these are the types of use cases that are more repeatable.
00:16:30.480 --> 00:16:33.519
They're on the enterprise side versus the consumer side.
00:16:33.759 --> 00:16:35.679
And yeah, I think there's a lot of them.
00:16:35.759 --> 00:16:37.360
And that's where we've been really focused.
00:16:37.519 --> 00:16:45.360
Like we actually were going to announce some big logos of companies we're working with in the next uh couple months and the specific use cases we're working on with them.
00:16:45.519 --> 00:16:47.840
So I may save that for that announcement.
00:16:48.480 --> 00:16:49.519
We'll put it in the LinkedIn post.
00:16:50.559 --> 00:16:57.919
I mean, it sounds like we're pretty early to the agent market, but it sounds like for anyone to stand and deploy an agent in production, you need authentication.
00:16:58.080 --> 00:17:06.720
So that's probably why your areas had more traction than maybe some of the other agent infrastructure areas, because this is one of the first things that your customers are turning to.
00:17:07.039 --> 00:17:09.279
I mean, I think the word agent is just so broad.
00:17:09.359 --> 00:17:12.480
It's like software with an LLM, is how people use it.
00:17:12.559 --> 00:17:15.200
So I don't know, like it's called so many different things.
00:17:15.440 --> 00:17:15.599
Yeah.
00:17:15.839 --> 00:17:23.119
We're specifically focused on, you know, agents that that drive browsers or use computer use models to like handle things.
00:17:23.279 --> 00:17:25.279
So it's a different era of agents.
00:17:25.440 --> 00:17:33.519
I would say the first agents that really work would be, I don't know if people call chat applications agents, or you know, that's obviously the thing that worked first.
00:17:33.839 --> 00:17:39.839
I think that agents that use APIs are working now.
00:17:40.079 --> 00:17:42.160
I mean, I use it on a daily basis.
00:17:42.319 --> 00:17:51.680
Like I use an MCP server within a client like Claude and have it like draft emails, write emails, do research, all that kind of stuff.
00:17:51.839 --> 00:17:57.519
Like I would call that objectively an agent, even though it's more in like a co-pilot versus like an autopilot function.
00:17:57.759 --> 00:17:58.640
And I think it's easier.
00:17:58.799 --> 00:18:02.319
I really do think the auth piece is easier when there are APIs.
00:18:02.480 --> 00:18:06.960
And like I just all the MCP servers right now just use OAuth as an auth mechanism.
00:18:07.119 --> 00:18:08.720
But I think those are agents that are working.
00:18:08.880 --> 00:18:16.640
I think the more different from how software works today you get, the more new tooling that needs to be built in support of the space.
00:18:16.880 --> 00:18:17.119
Yeah.
00:18:17.200 --> 00:18:20.640
So how does a non evolve with the space or without APIs?
00:18:20.960 --> 00:18:29.119
So in bringing this all into a single product instead of serving as a thinner horizontal layer in it, we control more about how it works.
00:18:29.359 --> 00:18:33.200
So we do take credentials from end users.
00:18:33.279 --> 00:18:35.440
I mean, our product looks a lot like plaid.
00:18:35.839 --> 00:18:46.400
If you've seen, like I like to mention that back, it's a modal that can be used within a dashboard we build, or it can be embedded within a dashboard within our customers' products.
00:18:46.559 --> 00:19:00.880
We collect credentials, we authenticate users, we securely store those credentials, we do some other things to make make it to where we can do persistent authentications or syncs over time to where it's not always just one-time authentication.
00:19:01.039 --> 00:19:06.960
So we actually are building a little bit into what I'll call the defensive side of authentication for some of those use cases.
00:19:07.200 --> 00:19:18.880
And then, yeah, we we're deciding, I think, from like a UX perspective, how much of tooling to expose to end users versus how much of the tooling we build ourselves.
00:19:19.119 --> 00:19:27.759
Our product, as we're working with a lot of these larger logos right now, is less of a developer product and it more feels like an API.
00:19:27.839 --> 00:19:38.000
It sort of feels like a magic API where there isn't an API otherwise, where your API used to be 30 people sitting in a room and doing some repetitive task, and now all of a sudden that's sort of solved.
00:19:38.240 --> 00:19:41.279
And we're building that in a scalable way right now.
00:19:41.359 --> 00:19:52.720
And then over time, we're certainly gonna expose more and more of that, but it lets us build a more reliable, resilient product when we're a bigger piece of the stack for a smaller number of use cases.
00:19:53.039 --> 00:19:57.839
And so you're handling the authentication and the integrations today with what your customers are requesting.
00:19:58.079 --> 00:20:04.640
Authentication, integrations, also building the agents and building the automations, and we're doing more on that side.
00:20:04.880 --> 00:20:08.240
But you're not being the browser use of the browser-based side, an agent.
00:20:08.559 --> 00:20:09.920
No, we use a lot of tooling.
00:20:10.000 --> 00:20:13.599
I mean, we use a lot of any infrastructures in building that space, we adopt.
00:20:13.759 --> 00:20:26.240
We could we'll use people to build proxies, we'll use companies like Browser Base to build headless browsers, we'll use these tools as part of our stack, but we are the ones that are actually building the agents for many of these use cases.
00:20:27.039 --> 00:20:27.599
Interesting.
00:20:27.839 --> 00:20:32.160
How do you decide as the solo founder and CEO which direction to take?
00:20:32.400 --> 00:20:37.359
Do you rely heavily on your CTO, or now that you've done this a few times, you rely on your own intuition?
00:20:37.680 --> 00:20:39.599
From a perspective of like company direction.
00:20:39.759 --> 00:20:39.920
Yeah.
00:20:40.240 --> 00:20:42.000
I think we just learn a lot in market.
00:20:42.079 --> 00:20:56.400
I mean, I would say that one of my hot takes on the computer use space overall is that there are not as many use cases as people think there will be for it, but the use cases that there are will be like deep, deep use cases.
00:20:56.559 --> 00:21:02.319
Like the number of conversations I've gone into recently with people where I say, hey, what are you excited about in the agent space?
00:21:02.400 --> 00:21:03.440
So like computer use models.
00:21:03.519 --> 00:21:07.519
I'm like, great, what is your favorite commercial use case for computer use models?
00:21:07.680 --> 00:21:11.359
And they're like, uh I mean, how would an enterprise use it today?
00:21:11.599 --> 00:21:13.200
What is actually working with computer use?
00:21:13.759 --> 00:21:19.200
Enterprises have huge services teams that support certain functions within the company.
00:21:19.359 --> 00:21:32.079
There are also very big businesses in the RPA space, and there are a lot of people that do a lot of things with antiquated methods, computer use models that are just automated script writing and cron jobs and things like that.
00:21:32.240 --> 00:21:35.680
Like there is a lot of surface area to replace on that side.
00:21:35.839 --> 00:21:45.359
I think what I mean more is that if you go to the operator page within ChatGPT and you look at the use cases, it has all these boilerplate use cases you can scroll through.
00:21:45.440 --> 00:21:50.079
And it's like go order burrito on DoorDash and go book my reservation.
00:21:50.160 --> 00:22:03.599
And I don't know, last year we were getting, I mean, we still get so many ins for our product that people are like, oh, the thing I'm building is a consumer agent, the thing I'm building is a travel agent, the thing I'm Building is a food ordering agent or a date planning agent.
00:22:03.839 --> 00:22:08.240
And I don't know that those will be commercial use cases for computer use.
00:22:08.400 --> 00:22:12.240
And I don't think it has anything to do with the models and where they are.
00:22:12.480 --> 00:22:32.880
When you think about things that are worth automating from a financial perspective, from a commercial perspective, and where it's actually better for a human, you look at big companies like Apple with Siri and Amazon with Alexa that have tried building something that was in this human consumer-proseumer automation space for a long time and have really struggled with it.
00:22:33.119 --> 00:22:35.759
I think it's not all technology related.
00:22:35.920 --> 00:22:43.200
There are some use cases that will be really killer, but I don't think people are going to stop using DoorDash and use DoorDash via chatbot.
00:22:43.440 --> 00:22:46.720
You have to think about your day-to-day and what takes up the most time.
00:22:46.880 --> 00:22:49.200
For me, what takes up a lot of time is writing a memo.
00:22:49.440 --> 00:22:49.599
Yeah.
00:22:50.000 --> 00:22:50.640
That can take hours.
00:22:50.880 --> 00:22:54.240
So therefore, if but that's not, you wouldn't use computer use for that.
00:22:54.319 --> 00:22:58.880
You would likely use GPT based on the research and summarizing the notes you already have.
00:22:59.200 --> 00:22:59.680
Totally.
00:22:59.839 --> 00:23:11.599
But I think the research, I think using computer use for research is the browser web functionality that's been built into Chat GPT and Cloud and all perplexity and all these products is actually really killer functionality.
00:23:11.839 --> 00:23:14.400
But I think, yeah, for taking actions, I I don't know.
00:23:14.480 --> 00:23:24.240
I had somebody who I consider really, really intelligent in this space categorize it like AI is going to be really good where it's replacing jobs, not tasks.
00:23:24.400 --> 00:23:30.720
And if you think about the average person, they don't do that many jobs on a daily basis outside of their work life.
00:23:31.039 --> 00:23:36.559
I don't know how many vacations do people really plan a year, how many times do people really book restaurant reservations?
00:23:36.799 --> 00:23:46.559
I think people get a little mixed up because the type of people that are early adopters of agents actually are taking a lot of trips and booking a lot of things and all these, but the average person doesn't.
00:23:46.640 --> 00:23:49.200
Like I don't think these are mass market consumer products.
00:23:49.440 --> 00:23:52.640
Yeah, and I think what we're seeing work today is some of the tedious, repetitive work.
00:23:52.720 --> 00:23:54.720
Like we're seeing a lot of contacts center automation as well.
00:23:55.359 --> 00:24:01.759
It goes more into the voice AI space, a lot of go-to-market tools, a lot of customer support, and then coding, obviously.
00:24:01.839 --> 00:24:07.759
But these are use cases where you have what good looks like, so you're able to hopefully mimic that into an agentic or AI product.
00:24:07.920 --> 00:24:13.279
But I don't know what happens in some of these more ambiguous spaces where what good looks like isn't necessarily defined.
00:24:13.599 --> 00:24:14.400
Yeah, 100%.
00:24:14.880 --> 00:24:19.519
I think that finding use cases where you know what good looks like is actually a really good way to put it.
00:24:19.839 --> 00:24:26.960
I think that deterministic workflows where you're able to objectively evaluate did the agent do what it was supposed to.
00:24:27.039 --> 00:24:38.720
And there's intuition in the how, but not a lot of intuition in the what is actually the type of use cases that we're personally going after, and that I think are going to be the best ones in the space, at least in the early days.
00:24:38.960 --> 00:24:49.839
So when you think about if you have a market map in your head right now, you have frameworks, hosting, routing, gosh, observability, evals, authentication, all these layers of the agent infrastructure stack, some are converging.
00:24:50.000 --> 00:24:56.960
Do you think there's any layer that's just not going to be needed or that maybe one of the hyperscalers or foundation model companies will just own?
00:24:57.200 --> 00:25:10.160
I think the big question that people are running into is how much is the infrastructure that worked for computers in the past and for building software in the past the same as for building software in the future?
00:25:10.400 --> 00:25:13.279
I think that there are instances where the how is better.
00:25:13.440 --> 00:25:23.039
I think cursors objectively taken over the valley and seemingly the world soon because it's like the how you write software that is better, but you're still writing software.
00:25:23.200 --> 00:25:24.319
Agents are still code.
00:25:24.960 --> 00:25:31.519
Like I mentioned before, we have seen companies that are building what I'll call defensive authentication for agents.
00:25:31.839 --> 00:25:34.000
And I don't actually understand that space.
00:25:34.160 --> 00:25:43.200
I don't understand what people are really building because I actually think that implement the slightly different implementations of OAuth, which is like an existing standard, are sufficient.
00:25:43.440 --> 00:25:47.599
I think that for hosting, a lot of the same tools are sufficient.
00:25:47.759 --> 00:25:55.920
I think for observability, there's maybe some differences, but I think a lot of the original observability tools actually are still pretty useful.
00:25:56.079 --> 00:25:59.839
I mean, I think there's more on that side, depending on like how deep you go.
00:25:59.920 --> 00:26:04.400
And I think there are things that they're very unique about agents in some aspects and not in others.
00:26:04.640 --> 00:26:05.680
But I don't know.
00:26:05.759 --> 00:26:11.759
I don't think the enemy of agents isn't like necessarily people launching new agent products.
00:26:11.839 --> 00:26:17.519
It's just like, are you sufficiently differentiated from traditional infrastructure in building agents?
00:26:17.599 --> 00:26:19.279
Because agents are still just software.
00:26:19.519 --> 00:26:24.720
I think that the difference between machine learning and gen AI, you started to see a new wave of tools coming up.
00:26:24.799 --> 00:26:28.640
But I think a lot of the Gen AI tools you will be able to use for agents.
00:26:28.799 --> 00:26:37.599
What you said with observability evals, I'm seeing that space converge and people just started in one package monitoring logs, testing the agent in production or testing your application in production.
00:26:37.759 --> 00:26:42.960
But I think generally speaking, there'll only be a few nuanced areas to spend something up.
00:26:43.119 --> 00:26:45.200
Do you price based on outcomes today?
00:26:45.680 --> 00:26:49.920
We price based on usage today, but usage is heavily tied to outcomes.
00:26:50.240 --> 00:26:50.480
Okay.
00:26:50.720 --> 00:26:52.000
Can you unpack that a little bit?
00:26:52.240 --> 00:26:52.880
What that means?
00:26:53.359 --> 00:26:58.079
We're facilitating specific workflows for and we charge like per workflow.
00:26:58.319 --> 00:27:01.680
So we charge a monthly base rate as a platform fee.
00:27:01.839 --> 00:27:05.440
And then the more you use the product, the more you pay for it.
00:27:05.680 --> 00:27:09.039
Similar to like I mentioned Plat earlier, similar pricing to that.
00:27:09.200 --> 00:27:15.440
The more data you pull, the more people are paying us to basically go and retrieve data and complex workflows for them.
00:27:15.920 --> 00:27:25.680
Do you think with some of these more of the application layer of agents, do you think this pricing debate that's going on, there actually will be a pricing evolution?
00:27:25.920 --> 00:27:31.119
Or do you think it'll still be what we've been seeing with SaaS for the last decade?
00:27:31.519 --> 00:27:35.440
I think that we'll price agents like a human, not software, is silly.
00:27:36.240 --> 00:27:50.559
Actually, I actually want to get your take on, I haven't unpacked this with anyone yet, but like Cognition's recent pricing change, where they went from like 200 a month, or no, was it 500 per seat, I guess, to now I think they're trying to match cursor's model closely.
00:27:50.799 --> 00:28:01.039
Do you think it's just because the products that are on the market today, they're just evolving so quickly and advancing and they're offering a lower price point for the same capabilities as your autonomous AI engineer.
00:28:01.200 --> 00:28:08.000
So therefore the price basically converges down to where the standard of what good looks like, which is cursor right now, might change.
00:28:08.480 --> 00:28:10.319
I don't know that it has anything to do with cursor.
00:28:10.400 --> 00:28:15.599
I would just say pricing dynamics are driven by the same market dynamics that have always driven markets.
00:28:15.759 --> 00:28:21.279
The product costs a certain amount to build, it's worth a certain amount to the end customer.
00:28:21.599 --> 00:28:29.039
If someone tries to charge$500 a month and their competitor is charging$20 a month, everyone's going to choose their competitor.
00:28:29.200 --> 00:28:36.079
And the product costs, you know, a few dollars a month to deliver, probably in terms of infrastructure costs.
00:28:36.319 --> 00:28:49.039
So it's just like if you were the only guy mowing lawns in town and you're the only lawn mower and you charged$1,000 a yard, and then someone could bought a second lawn mower and charging$50 a yard, your pricing has to come down to match that.
00:28:49.279 --> 00:28:54.799
Pricing has nothing to do with the presentation of it or whether you give your agent a name, like, hey, this is Doug, my agent.
00:28:54.880 --> 00:28:55.680
He's$100,000.
00:28:55.920 --> 00:29:03.119
Like it just has to do with the value you're delivering and what the cost relatively to other competitors into the market is.
00:29:03.440 --> 00:29:08.079
It sounds like something you're alluding to is it also just comes down to execution and go to market speed.
00:29:08.640 --> 00:29:09.359
Yeah, yeah.
00:29:09.519 --> 00:29:12.559
I think markets behave like markets behave in a lot of ways.
00:29:12.720 --> 00:29:19.519
It's like a very just like baseline economics, having your agents seem more like a human versus not, or having it be.
00:29:19.759 --> 00:29:30.480
If somebody comes up with something that adds insane value and it can't be replicated by any competitors for whatever reason, then they're going to be able to charge an ungodly amount of money for it.
00:29:30.640 --> 00:29:33.839
But as soon as you have market competition, it's going to drive prices down.
00:29:34.079 --> 00:29:36.480
Do you have a product you admire that charges a lot?
00:29:36.720 --> 00:29:41.680
A product that has that's like unparalleled in terms of it works well, you would pay double for it today.
00:29:42.079 --> 00:29:43.920
I think this space is moving so fast.
00:29:44.160 --> 00:29:50.160
I've talked before, I was a big, I was a big early adopter of Claude and the anthropic products.
00:29:50.559 --> 00:30:00.880
Like I, you know, there was a point in time I was probably using it like eight hours a day in my work to write everything I wrote, to send memos, send messages, plans, write PRDs, like everything we do.
00:30:01.119 --> 00:30:07.759
It's still good, but I think that ChatGBT has launched a lot of improvements recently and has come to parity with it, maybe even better.
00:30:07.839 --> 00:30:12.319
I think these products, like again, they actually drive so much outsized value.
00:30:12.480 --> 00:30:21.759
If I objectively were how much productivity do I get by using these products on a daily basis, then I would pay them tens of thousands of dollars a year for a personal license.
00:30:21.920 --> 00:30:27.519
But luckily, there's competition in the market, so they drive each other's prices down and you don't have to worry about it.
00:30:27.599 --> 00:30:29.839
So do you use perplexity?
00:30:30.079 --> 00:30:30.640
Yeah.
00:30:30.960 --> 00:30:31.839
Over Google?
00:30:32.079 --> 00:30:32.559
Yeah.
00:30:32.799 --> 00:30:34.160
I really like it.
00:30:34.480 --> 00:30:35.279
Why is that?
00:30:35.680 --> 00:30:39.839
I really like it because I think it gives the answer you're looking for.
00:30:40.160 --> 00:30:46.559
I think Google's product, I mean, I think Google's done a lot of amazing work, but objectively, Google's selling.
00:30:46.799 --> 00:30:49.839
Google's so heavily gamified at this point.
00:30:50.000 --> 00:30:58.240
Yeah, with ads and with SEO, if you're like, you know, what's the best, you know, give me five pizza places in South Park to go to lunch?
00:30:58.319 --> 00:31:00.880
It's like who's bidding the most, who hired the agency.
00:31:01.039 --> 00:31:05.200
I mean, my consulting business I built years ago, we did a lot of SEO work, game Google.
00:31:05.279 --> 00:31:06.480
There's a whole thing around it.
00:31:06.640 --> 00:31:08.480
I think Perplexity's results are cleaner.
00:31:08.559 --> 00:31:11.119
It gives me the thing I'm looking for, it has citations.
00:31:11.359 --> 00:31:14.640
I don't know how they monetize, but beyond that, I think it's great.
00:31:14.880 --> 00:31:23.839
Do you think people will start to game Perplexity, GPT, all of the foundation model consumer products with this new wave of AI SEO companies?
00:31:24.160 --> 00:31:26.000
I think that is the question, right?
00:31:26.160 --> 00:31:30.960
Because Google, it's honestly this trade-off between experience and monetization.
00:31:31.200 --> 00:31:36.400
Google is the best monetization machine that's ever been built, and it's heavily gamed.
00:31:36.559 --> 00:31:51.440
I think actually Google in hindsight probably should have spent more time thinking about UX and less about monetization, maybe over the last like, and they might not be in as precarious, not that they're in a precarious position, but they have they have people sieging them for the first time, right?
00:31:51.599 --> 00:31:53.279
Their business is under attack.
00:31:53.519 --> 00:31:59.359
And right now, perplexity is pristine in that there are no ads and it gives you objectively the right results.
00:31:59.599 --> 00:32:02.400
But I think there's a real question of monetization on that.
00:32:02.559 --> 00:32:10.480
And maybe they meet more in the middle at some point, and Google has to retreat from their, you know, money printing machine to deliver a better experience.
00:32:10.640 --> 00:32:15.920
And probably perplexity has to build some ads and some bias and Google bias perplexity.
00:32:16.160 --> 00:32:17.519
Yeah, I mean, totally.
00:32:17.680 --> 00:32:18.480
I think, yeah.
00:32:18.720 --> 00:32:18.960
Okay.
00:32:19.039 --> 00:32:20.559
Do you want to end with a quick fire round?
00:32:20.720 --> 00:32:22.480
We went a little off script, but that's okay.
00:32:22.720 --> 00:32:23.039
Totally.
00:32:23.200 --> 00:32:25.920
Actually, I don't think I asked any questions off script.
00:32:26.480 --> 00:32:29.440
What's the one word to describe your company culture today?
00:32:29.680 --> 00:32:30.720
I would say fun.
00:32:31.279 --> 00:32:34.480
I think that we really are trying to build a culture.
00:32:34.720 --> 00:32:39.680
A big learning I had from my last company is that it's a journey.
00:32:39.839 --> 00:32:42.559
Things are gonna go well, things are not gonna go well.
00:32:42.880 --> 00:32:48.400
And over time, the thing that is consistent is the people that are on the table through those good times and bad.
00:32:48.480 --> 00:32:52.880
So we really try to make time to have a good time with each other, bond with each other.
00:32:53.039 --> 00:32:55.759
We work really hard, we spend a lot of hours in the office.
00:32:56.000 --> 00:32:58.720
But last night we had an amazing customer demo.
00:32:58.880 --> 00:33:02.799
We killed it and we did a celebratory game night and hung out at the office.
00:33:02.960 --> 00:33:05.200
And I think it's really important to find that balance.
00:33:05.599 --> 00:33:11.359
What's one or two attributes of a new candidate that you're looking at that you really admire?
00:33:11.680 --> 00:33:16.880
So, again, based on I'd say experience, we hire people that have done the thing before.
00:33:17.039 --> 00:33:20.240
And what I mean by that is we're hiring a lot of engineers right now.
00:33:20.400 --> 00:33:22.799
So, any great engineers watching, reach out to me.
00:33:22.960 --> 00:33:26.240
We want to hire startup engineers that have built startups before.
00:33:26.400 --> 00:33:30.160
And we want to hire people to lead teams that have led teams before.
00:33:30.400 --> 00:33:38.720
We want to hire people at this stage we're at right now where, you know, we're early but growing really quickly that have been early and grown really quickly before.
00:33:38.880 --> 00:33:48.480
I think you get a lot of interviews where someone's like, Yeah, I've worked at, you know, meta for the last eight years, but I really want to go work at a startup because I want more ownership.
00:33:48.640 --> 00:33:53.759
And, you know, it's not that I don't believe them or I think they're being disingenuous, but I don't think they know what that means.
00:33:53.920 --> 00:33:59.440
I think until you've sat in that exact seat, you actually don't really know what you're talking about.
00:33:59.680 --> 00:34:03.519
So I want people that have sat in that seat before and done that exact thing before.
00:34:03.599 --> 00:34:06.880
And it usually means you're finding more like diamond in the rough type of candidates.
00:34:07.119 --> 00:34:10.400
They don't have the big logos, they probably don't have the schools.
00:34:10.559 --> 00:34:13.679
You know, we're not averse to that, and we will hire people like that.
00:34:13.840 --> 00:34:26.079
But the our big criteria is that you're hardworking, you're humble, you've done the thing you're doing before, and you're motivated by the mission, the team, the journey, not the equity price or the comp or things like that.
00:34:26.400 --> 00:34:29.199
And what's an AI tool that people aren't talking enough about?
00:34:29.440 --> 00:34:32.320
An AI tool that people aren't talking enough about, anon.
00:34:32.800 --> 00:34:34.639
Will passwords ever fully die?
00:34:34.960 --> 00:34:36.880
Will passwords ever fully die?
00:34:37.199 --> 00:34:38.880
You know, maybe.
00:34:39.280 --> 00:34:56.159
I you know what's a funny thing is that we had all this move to certain types of passwordless auth, pass keys and stuff, which I think are okay, and CAPTCHAs, which you know now are getting solved at a lower rate with humans and AI, like AI is actually better at solving.
00:34:56.320 --> 00:35:08.960
The funny thing is the thing that is actually really effective in handling did this user actually really want this other service to use the thing that they're using is old school two-factor authentication.
00:35:09.199 --> 00:35:19.199
It's sending a text message to the person and having them enter that code in because it establishes a physical presence of a user holding a phone and access to the device.
00:35:19.440 --> 00:35:22.079
People don't change phone numbers that often, it's hard to hack.
00:35:22.239 --> 00:35:30.239
So it's funny, it's like some things never change, and that auth does, but I actually think passwords will be around for a pretty long period of time.
00:35:30.559 --> 00:35:31.039
Awesome.
00:35:31.119 --> 00:35:31.679
Well, thank you.
00:35:31.840 --> 00:35:33.039
Yeah, that was fun.
00:00:00.160 --> 00:00:08.800
There's a lot of like nuance around like if agents are not humans and they're not machines, then what is that identity standard look like in the future.
00:00:15.439 --> 00:00:22.399
Welcome to Partner Path, a podcast that unpacks the venture capital and growth equity ecosystem from a junior perspective.
00:00:22.559 --> 00:00:28.960
Young entrepreneurs and investors have already had a massive impact on the industry, having started unicorns and launched billion-dollar funds.
00:00:29.280 --> 00:00:35.679
We discuss these success stories and more by sharing perspectives and advice from some of the industry's most prominent role models.
00:00:35.840 --> 00:00:42.240
This podcast expresses our views as of the date published and does not represent the views of and is not endorsed by any company for which we work.
00:00:44.719 --> 00:00:49.759
Before we get to the episode, Laura and I are excited to announce a partnership with Overlap.
00:00:50.240 --> 00:00:59.759
Fresh Out of Y Combinator's Summer Badge, Overlap is an AI-driven app that uses large language models to curate the best moments from podcast episodes.
00:01:00.079 --> 00:01:11.920
Imagine having a smart assistant who reads through every podcast transcript, finds the best parts or parts most relevant to your search, and strings them together to form a new curated stream of content.
00:01:12.239 --> 00:01:14.079
That is what Overlap does.
00:01:14.560 --> 00:01:22.319
I personally use Overlap to explore content on future podcast guests and highly recommend it as another diligence tool for investors.
00:01:22.879 --> 00:01:24.959
Check out Overlap on the App Store today.
00:01:25.359 --> 00:01:28.719
Today we're chatting with Daniel Mason, co-founder and CEO at Anon.
00:01:28.959 --> 00:01:37.040
Daniel previously founded Spring Labs, a platform that builds data privacy technology to power use cases across financial services, consumer identity, and crypto.
00:01:37.200 --> 00:01:40.719
And now Daniel's focus on building the identity layer for agents.
00:01:40.879 --> 00:01:41.359
Welcome.
00:01:42.000 --> 00:01:43.120
Yeah, it's good to be here.
00:01:43.359 --> 00:01:44.400
Excited to chat today.
00:01:44.719 --> 00:01:50.000
After graduation, you worked in a variety of product manager roles before starting sabbatical co-working.
00:01:50.239 --> 00:01:58.799
What exactly sparked your interest in becoming an entrepreneur and going into the coworking space after seeing some of the hardships that WeWork has faced?
00:01:59.280 --> 00:02:04.000
I feel like my journey to entrepreneurship was like a pretty easy transition, actually.
00:02:04.159 --> 00:02:07.040
I actually didn't work anywhere for very long.
00:02:07.200 --> 00:02:14.080
Like I had a job right out of a college at a startup, which is pretty rare when I was graduating college, like much more common now.
00:02:14.240 --> 00:02:20.560
But I happened to have a friend who started a company that I joined, taught myself to code, was an engineer there in the very early days.
00:02:20.879 --> 00:02:25.759
And we built marketing software and we worked with companies that were doing like advertising campaigns.
00:02:25.840 --> 00:02:27.520
It was kind of like an ad tech company.
00:02:27.759 --> 00:02:32.479
And occasionally companies would come to us and ask if we could help them with professional services.
00:02:32.639 --> 00:02:36.080
And we would do that for engagements that were above a certain size.
00:02:36.400 --> 00:02:40.719
And I was also a BD guy in addition to an engineer and source deals.
00:02:40.800 --> 00:02:45.759
And sometimes the company would say no and I would say yes to those deals and do them on the nights and weekends.
00:02:46.000 --> 00:02:52.800
And within a couple years, and I had another job during this time too at Red Hat as a product manager.
00:02:52.960 --> 00:03:01.599
But during this time, at some point I looked up and was making more money from my night and weekend side job than I was from my main W-2 paycheck.
00:03:01.759 --> 00:03:05.039
So I decided to start working for myself exclusively.
00:03:05.120 --> 00:03:07.360
And that was kind of the start of the entrepreneurial journey.
00:03:07.599 --> 00:03:09.039
And that was Spring Labs?
00:03:09.360 --> 00:03:11.199
No, this was way before Spring Labs.
00:03:11.439 --> 00:03:13.120
This was when I was just out of college.
00:03:13.360 --> 00:03:15.919
And then, yeah, you mentioned sabbatical co-working.
00:03:16.000 --> 00:03:19.199
I mean, my that was actually a business I started with my brother.
00:03:19.439 --> 00:03:28.879
He, my brother, worked in real estate private equity, and we both had this vision of remote work as kind of the future of work or at least some sort of hybrid work, which was really not the case then.
00:03:28.960 --> 00:03:31.199
This was like 2013 or something.
00:03:31.520 --> 00:03:37.280
And so we decided we wanted to build a space for digital nomads to go travel and work.
00:03:37.520 --> 00:03:39.360
So we moved to Puerto Rico together.
00:03:39.520 --> 00:03:45.120
We master leased a hotel, tried to get a couple more hotels, raised some money for it.
00:03:45.360 --> 00:03:48.159
And yeah, it was a catastrophic disaster.
00:03:48.479 --> 00:03:54.240
We can go more into if you want, but it was a fun foray into a totally different type of entrepreneurship.
00:03:54.400 --> 00:04:08.240
And through those first two experiences, I think I realized I don't want to build large-scale services businesses, and I don't want to build non-tech businesses, especially in other countries where it's a lot harder to be an entrepreneur.
00:04:08.400 --> 00:04:12.479
So yeah, I was kind of set on building tech venture-back tech companies after that.
00:04:12.719 --> 00:04:13.520
Okay, that's fair.
00:04:13.759 --> 00:04:16.480
Do you think the bottleneck was more on the former or the latter?
00:04:16.800 --> 00:04:17.920
Of those two types of company.
00:04:18.319 --> 00:04:23.759
With the services business, like with my consulting business I ran, it actually got decently sized over time.
00:04:23.920 --> 00:04:27.199
It was just always boring, is like my only complaint about it.
00:04:27.279 --> 00:04:31.120
It was like good cash flow, a consistent customer base.
00:04:31.279 --> 00:04:33.360
We had like a service that people wanted.
00:04:33.519 --> 00:04:38.480
We did a lot of marketing and data science consulting work was like what we found our specialty.
00:04:38.720 --> 00:04:42.639
And yeah, it was just like, I don't know, I woke up every day and wasn't excited by it.
00:04:42.720 --> 00:04:48.480
So after a few years of doing it, even though it was like a great economic opportunity, I decided I wanted to go do something else.
00:04:48.720 --> 00:04:50.560
And that's when you went to Spring Labs.
00:04:50.720 --> 00:04:51.439
Yes, exactly.
00:04:51.680 --> 00:04:51.920
Okay.
00:04:52.079 --> 00:04:53.839
And maybe break down how you got the idea.
00:04:54.000 --> 00:04:58.000
I mean, you raised a lot of venture money, which is very impressive and hard to do.
00:04:58.240 --> 00:05:01.120
But talk about like the original thesis and how it evolved.
00:05:01.439 --> 00:05:06.879
So Spring Labs was a company I started in 2017 with a few co-founders.
00:05:07.040 --> 00:05:15.199
I was, it was a very lucky break for me in some ways because my co-founders were very seasoned entrepreneurs who had built businesses together previously.
00:05:15.360 --> 00:05:26.959
I was kind of like the last guy to join in, like the other, there were four of us, the other three businesses together in the past, including some like very large businesses that are now once now publicly traded and some really serious companies.
00:05:27.120 --> 00:05:37.279
And the original vision for the business was that we wanted to build a new credit bureau, which is kind of one of those tar pit ideas you hear a lot of startups pitch and then eventually pivot away from.
00:05:37.439 --> 00:05:40.000
That was kind of our trajectory on it, too.
00:05:40.160 --> 00:05:46.639
But I think we thought we had a lot of connections in the financial services space, and a lot of our early customers were fintech businesses.
00:05:46.800 --> 00:05:50.399
And the business, it's a multi-product business now that does a bunch of different things.
00:05:50.560 --> 00:05:59.360
But the first product that we built was really like a data storage tokenized vault type product, which was kind of what a credit bureau would do anyway.
00:05:59.439 --> 00:06:03.519
And we ended up working with credit bureaus instead of selling, instead of becoming a credit bureau.
00:06:03.600 --> 00:06:06.240
Transunion eventually became a big customer of ours.
00:06:06.399 --> 00:06:10.800
They also were one of our largest investors and led our Series B round.
00:06:10.959 --> 00:06:15.680
So we ended up working in the credit bureau space, just not in the exact original version of the idea.
00:06:16.000 --> 00:06:20.399
Did you take any lessons from your first in entrepreneurship into Spring Labs?
00:06:20.639 --> 00:06:23.279
I honestly was kind of in the passenger seat for that one.
00:06:23.439 --> 00:06:26.800
I was the one learning from my other co-founders in that business.
00:06:27.040 --> 00:06:29.360
I think that I learned from my earlier businesses.
00:06:29.439 --> 00:06:31.759
When you build a services business, you learn how to sell.
00:06:31.920 --> 00:06:33.680
So I did a lot of our selling.
00:06:33.759 --> 00:06:41.600
I ran product in the early days of the company, but for most of my time there, I ran our sales and marketing and growth operations.
00:06:42.000 --> 00:06:47.360
And I think that I learned to sell and selling my own services for years, kind of in that side business.
00:06:47.519 --> 00:06:51.279
And that's made me a better salesperson and kind of builder of sales teams.
00:06:51.439 --> 00:06:57.360
But other than that, on fundraising, on product, on engineering, I learned so much from my co-owners in that business.
00:06:57.600 --> 00:07:01.040
I think one of the best skill sets is someone who's technical that can sell.
00:07:01.199 --> 00:07:01.360
Yeah.
00:07:01.839 --> 00:07:02.560
It's hard to come by.
00:07:02.800 --> 00:07:03.360
I think so.
00:07:03.519 --> 00:07:07.680
And and I'm by by like proper technical standards, I'm not technical.
00:07:07.759 --> 00:07:08.800
Like I'm code.
00:07:09.120 --> 00:07:10.319
Yeah, I code from time to time.
00:07:10.480 --> 00:07:13.040
I code a lot more these days with AI tools, actually.
00:07:13.279 --> 00:07:18.560
I feel like my GitHub finally has some green on it after a few years of being pretty, pretty dark.
00:07:18.879 --> 00:07:20.560
Are you so you're using a cursor?
00:07:20.879 --> 00:07:21.120
Yeah.
00:07:21.360 --> 00:07:26.079
Do you think you have to have a base skill set nowadays, technical base to build a company?
00:07:26.480 --> 00:07:28.560
I think it depends on the type of company.
00:07:28.800 --> 00:07:34.959
I feel that every company has some things that are its differentiator.
00:07:35.040 --> 00:07:37.600
It's like, what do you have to get right to build this company?
00:07:37.759 --> 00:07:44.319
And I think in the in like a post-AI world, a lot of times you just have to get distribution or something to build the company.
00:07:44.480 --> 00:07:48.000
The actual product is somewhat simple as, you know, I don't want to say a wrapper.
00:07:48.240 --> 00:07:56.480
I think that term's like way overused, but there's not as much technical, like some companies, and on, I would say like we have a lot that we have to build technically.
00:07:56.560 --> 00:08:00.800
Like we have like risk and building on the technical side as well as the distribution side.
00:08:00.959 --> 00:08:12.160
But I think there's a lot of businesses it's like a better product for, you know, accounting firms or for like some like large-scale business where what you really have to get is distribution on top of the product.
00:08:12.480 --> 00:08:16.319
Maybe we should touch on Anon and how you got to Anon and what is Anon.
00:08:16.879 --> 00:08:18.079
Yeah, absolutely.
00:08:18.319 --> 00:08:20.879
So we started Anon about two years ago.
00:08:21.120 --> 00:08:24.160
So I left Spring Labs a while after our series B round.
00:08:24.240 --> 00:08:26.240
I'd sort of like worked my way out of a job.
00:08:26.399 --> 00:08:36.159
I was a the chief strategy officer, I think was my last role there after we hired a chief revenue officer who had sold a lot larger contracts and stuff than me in like a past life.
00:08:36.399 --> 00:08:37.919
So eventually left.
00:08:38.080 --> 00:08:42.639
I actually worked in venture for a year at a fund called Framework Ventures that I knew the founders of.
00:08:42.879 --> 00:08:45.679
Did to give you a different perspective, being on the other side of the table?
00:08:46.080 --> 00:08:46.879
Yeah, it did.
00:08:46.960 --> 00:08:47.360
It did.
00:08:47.519 --> 00:09:01.360
I think it made me understand fundraising more and it made me think more about markets and sizing and a lot of the meta aspects of companies, I think that when you're in the trenches day-to-day, you just don't think as much about.
00:09:01.679 --> 00:09:08.720
But yeah, and I kind of thought I would do that for a while, investing and was honestly pretty burnt out on startups after Spring Labs.
00:09:08.879 --> 00:09:10.240
Just like it's a it's a journey.
00:09:10.320 --> 00:09:16.240
I think anyone who's ever started a company, like by year five or whatever that I was in, like you're pretty tired.
00:09:16.399 --> 00:09:16.799
I don't know.
00:09:16.879 --> 00:09:21.279
I just was really fascinated by some of the things going on in the AI space.
00:09:21.440 --> 00:09:43.519
And I think at this intersection of building companies in identity and security and this advent of new AI technology, we got really obsessed with this, like what we'll call like identity layer of agents or how agents use human identities, use like machine identities, how they get access to accounts, because a lot of the things that you want your agent to do, they need to use your account for.
00:09:43.600 --> 00:09:46.639
So I think it's a really important kind of primitive in the broader stack.
00:09:46.960 --> 00:09:50.399
But were you not seeing Okta, Ottsier, their product?
00:09:50.559 --> 00:09:52.559
Were they not doing it well today?
00:09:52.960 --> 00:09:59.440
I think that it's it's actually a very nuanced concept that's still kind of playing out over time.
00:09:59.679 --> 00:10:01.519
I actually get a lot of outreach.
00:10:01.679 --> 00:10:10.879
It's funny from a lot of people on different, especially investors actually, that are like, oh, I have a thesis on agent identity, and I'm always like, okay, what's your thesis?
00:10:11.120 --> 00:10:16.720
And I just find that people haven't really thought it through that much in a lot of instances.
00:10:16.960 --> 00:10:23.840
What I would say is that you almost today have this bipolar idea of what identity is.
00:10:24.000 --> 00:10:29.919
Like identity is either human identity, which is an email password, a magic lake, pass keys.
00:10:30.000 --> 00:10:33.279
It's like something that is an authentication mechanism for humans.
00:10:33.519 --> 00:10:37.759
And on the other side, you have machine identity, right?
00:10:38.000 --> 00:10:45.840
You have API keys, you have OAuth, you have ways of authorizing an application to basically act on behalf of you.
00:10:46.159 --> 00:10:48.720
And agents have to be one of those two things.
00:10:48.879 --> 00:11:07.840
So when you look at the rollout of different types of technologies, you see that agents are either using human identities and that the agents collecting an email and password, authenticating the user, and then letting it take actions on your behalf, which is usually using computer use or sometimes it's using reverse engineering.
00:11:07.919 --> 00:11:10.000
It's like usually when there's not an API route.
00:11:10.159 --> 00:11:14.399
And when there is an API route, OAuth is still the dominant authentication mechanism.
00:11:14.559 --> 00:11:21.120
And OAuth is designed for delegation of human identity to a machine, which actually fits an agent pretty well.
00:11:21.279 --> 00:11:29.919
But there's a lot of like nuance around like if agents are not humans and they're not machines, then what does that identity standard look like in the future?
00:11:30.000 --> 00:11:40.159
And I think you are seeing people like Okta did a release actually very recently that something like they called like non-human identity or something, which is like kind of a blend of the two identities.
00:11:40.399 --> 00:11:41.440
I need to dig more into it.
00:11:41.679 --> 00:11:43.519
Or invested in a non-human identity company.
00:11:43.679 --> 00:11:44.000
Yeah.
00:11:44.159 --> 00:11:44.399
Yeah.
00:11:44.639 --> 00:11:49.279
I think they're going to start doing well because Agent is because you could consider it a non-human.
00:11:49.440 --> 00:11:57.919
So perhaps if there's more breaches, then folks are going to be turning to not the traditional identity players, which is a very solved place, but the non-human on the security side.
00:11:58.240 --> 00:11:59.919
But you're more on the developer side.
00:12:00.159 --> 00:12:00.399
Yeah.
00:12:00.559 --> 00:12:00.720
Right.
00:12:00.960 --> 00:12:02.720
How are you approaching the problem differently?
00:12:02.879 --> 00:12:05.279
What does it look like to work or to use a non?
00:12:05.600 --> 00:12:12.960
I would say that like our product lives, I almost would put authentication in in two camps.
00:12:13.039 --> 00:12:16.879
There's like defensive authentication and more like offensive authentication.
00:12:17.039 --> 00:12:21.200
I don't know if that's I've never heard anyone else use that, but that's kind of how I think about it.
00:12:21.519 --> 00:12:26.799
And we're specifically in the business of giving agents access to tools.
00:12:27.120 --> 00:12:41.919
So, you know, like a better comp for us would be businesses like plaid that have been built in the past that were building automation on top of like some sort of workflow where there weren't APIs or there weren't easy ways of building that automation.
00:12:42.240 --> 00:12:51.360
We're not in the business of like we work for the agent companies, we don't work for the services that are furnishing access to the agent companies, if that makes any sense.
00:12:51.440 --> 00:12:54.559
Like AUK is objectively on the other side, I would say, of this.
00:12:55.679 --> 00:12:57.840
Now, a quick word from our sponsor.
00:12:58.480 --> 00:13:04.720
Laura and I are thrilled to announce a partnership with Grata, an intuitive deal sourcing and diligence platform.
00:13:05.120 --> 00:13:06.960
Keep missing out on private market deals.
00:13:07.279 --> 00:13:09.120
Ditch the Rolodex and get Grata.
00:13:09.519 --> 00:13:17.759
Grata's AI-powered platform provides accurate data, deep insights, and time-saving workflows that help get more deals done.
00:13:18.000 --> 00:13:23.120
And it's trusted by over 500 firms and 3,000 deal makers, including myself.
00:13:23.440 --> 00:13:33.039
I've personally used Grata in both my deal sourcing and diligence workflows and would recommend their tool to any investor across private equity, growth equity, corpdev, or investment banking.
00:13:33.360 --> 00:13:36.080
We've negotiated a special offer for our audience.
00:13:45.919 --> 00:13:47.440
Now back to the episode.
00:13:48.000 --> 00:13:53.279
How do you handle the services or the applications where you can't integrate with today?
00:13:53.600 --> 00:13:55.039
But they're asking you for that.
00:13:55.360 --> 00:13:55.600
Yeah.
00:13:55.840 --> 00:13:57.759
Yeah, it's actually really interesting.
00:13:58.000 --> 00:14:16.720
We chose this specific type of business to work with because I think if you divide it into the offensive and defensive authentication, and then things where there are good routes with APIs and things where there aren't good routes with APIs, I think that anything on the API side, I'm curious to learn more.
00:14:16.879 --> 00:14:21.120
I actually don't know a ton about startups building in this like non-human identity space.
00:14:21.440 --> 00:14:30.720
But you know, there's this actually this interesting blog that I've sent to a bunch of people that says like agent auth is just OAuth or something is the title of the article.
00:14:30.879 --> 00:14:40.480
And it's it's basically making the case that for things where APIs exist is 96% good for actually being agent identity.
00:14:40.559 --> 00:14:41.759
Like it is missing a few things.
00:14:41.919 --> 00:14:49.360
I think there's been some like interesting posts about it missing uh rate limits and time, like you can't time bound OAuth and things.
00:14:49.440 --> 00:14:58.480
But the way, I don't know, the same way you let Calendly use your Google Calendar to book times, it's actually really easy to let an agent use your Google Calendar to send emails.
00:14:58.720 --> 00:15:04.960
And it, you know, there could be more protections put in place for the non-determinism of LLMs, but it's actually pretty good.
00:15:05.120 --> 00:15:12.320
We like the space we're in because we think the internet is like a big, vast place, and there's a lot of automations that you can't do today.
00:15:12.399 --> 00:15:27.759
And we think a lot of companies struggle with that, and a huge amount of the services economy and a lot of the messy things, the rooms full of people that are all doing the same clicking and repetitive actions, or the, you know, follow these 16 steps to integrate this new tool.
00:15:27.919 --> 00:15:32.240
We think that's where there's a massive amount of opportunity, and there aren't APIs there.
00:15:32.320 --> 00:15:40.639
And in the world of APIs not existing, there's actually a massive amount of room for how identity is solved in that space, and that's the space that we're really focused on.
00:15:40.960 --> 00:15:41.600
That's interesting.
00:15:41.759 --> 00:15:48.159
Have you seen any like unique automation use cases today where a non has become a critical part of the workflow?
00:15:48.480 --> 00:15:49.200
Totally.
00:15:49.440 --> 00:15:55.039
We have sort of two products, I would say, that people use us for in in different ways today.
00:15:55.360 --> 00:16:00.720
One is kind of our original product that we launched, which is almost just the identity layer.
00:16:00.799 --> 00:16:11.039
It's like we collect credentials, we handle authentication, and we authenticate users, a lot of startups, some early agent companies, a lot of things you saw on Twitter probably last year were using that product.
00:16:11.279 --> 00:16:26.559
We have recently, though, we've gotten much more interested in building much deeper into the stack and actually building the agents, building the authentication, building more of the infrastructure and the actual capturing more of the business value ourselves and building more for specific use cases.
00:16:26.799 --> 00:16:30.399
And these are the types of use cases that are more repeatable.
00:16:30.480 --> 00:16:33.519
They're on the enterprise side versus the consumer side.
00:16:33.759 --> 00:16:35.679
And yeah, I think there's a lot of them.
00:16:35.759 --> 00:16:37.360
And that's where we've been really focused.
00:16:37.519 --> 00:16:45.360
Like we actually were going to announce some big logos of companies we're working with in the next uh couple months and the specific use cases we're working on with them.
00:16:45.519 --> 00:16:47.840
So I may save that for that announcement.
00:16:48.480 --> 00:16:49.519
We'll put it in the LinkedIn post.
00:16:50.559 --> 00:16:57.919
I mean, it sounds like we're pretty early to the agent market, but it sounds like for anyone to stand and deploy an agent in production, you need authentication.
00:16:58.080 --> 00:17:06.720
So that's probably why your areas had more traction than maybe some of the other agent infrastructure areas, because this is one of the first things that your customers are turning to.
00:17:07.039 --> 00:17:09.279
I mean, I think the word agent is just so broad.
00:17:09.359 --> 00:17:12.480
It's like software with an LLM, is how people use it.
00:17:12.559 --> 00:17:15.200
So I don't know, like it's called so many different things.
00:17:15.440 --> 00:17:15.599
Yeah.
00:17:15.839 --> 00:17:23.119
We're specifically focused on, you know, agents that that drive browsers or use computer use models to like handle things.
00:17:23.279 --> 00:17:25.279
So it's a different era of agents.
00:17:25.440 --> 00:17:33.519
I would say the first agents that really work would be, I don't know if people call chat applications agents, or you know, that's obviously the thing that worked first.
00:17:33.839 --> 00:17:39.839
I think that agents that use APIs are working now.
00:17:40.079 --> 00:17:42.160
I mean, I use it on a daily basis.
00:17:42.319 --> 00:17:51.680
Like I use an MCP server within a client like Claude and have it like draft emails, write emails, do research, all that kind of stuff.
00:17:51.839 --> 00:17:57.519
Like I would call that objectively an agent, even though it's more in like a co-pilot versus like an autopilot function.
00:17:57.759 --> 00:17:58.640
And I think it's easier.
00:17:58.799 --> 00:18:02.319
I really do think the auth piece is easier when there are APIs.
00:18:02.480 --> 00:18:06.960
And like I just all the MCP servers right now just use OAuth as an auth mechanism.
00:18:07.119 --> 00:18:08.720
But I think those are agents that are working.
00:18:08.880 --> 00:18:16.640
I think the more different from how software works today you get, the more new tooling that needs to be built in support of the space.
00:18:16.880 --> 00:18:17.119
Yeah.
00:18:17.200 --> 00:18:20.640
So how does a non evolve with the space or without APIs?
00:18:20.960 --> 00:18:29.119
So in bringing this all into a single product instead of serving as a thinner horizontal layer in it, we control more about how it works.
00:18:29.359 --> 00:18:33.200
So we do take credentials from end users.
00:18:33.279 --> 00:18:35.440
I mean, our product looks a lot like plaid.
00:18:35.839 --> 00:18:46.400
If you've seen, like I like to mention that back, it's a modal that can be used within a dashboard we build, or it can be embedded within a dashboard within our customers' products.
00:18:46.559 --> 00:19:00.880
We collect credentials, we authenticate users, we securely store those credentials, we do some other things to make make it to where we can do persistent authentications or syncs over time to where it's not always just one-time authentication.
00:19:01.039 --> 00:19:06.960
So we actually are building a little bit into what I'll call the defensive side of authentication for some of those use cases.
00:19:07.200 --> 00:19:18.880
And then, yeah, we we're deciding, I think, from like a UX perspective, how much of tooling to expose to end users versus how much of the tooling we build ourselves.
00:19:19.119 --> 00:19:27.759
Our product, as we're working with a lot of these larger logos right now, is less of a developer product and it more feels like an API.
00:19:27.839 --> 00:19:38.000
It sort of feels like a magic API where there isn't an API otherwise, where your API used to be 30 people sitting in a room and doing some repetitive task, and now all of a sudden that's sort of solved.
00:19:38.240 --> 00:19:41.279
And we're building that in a scalable way right now.
00:19:41.359 --> 00:19:52.720
And then over time, we're certainly gonna expose more and more of that, but it lets us build a more reliable, resilient product when we're a bigger piece of the stack for a smaller number of use cases.
00:19:53.039 --> 00:19:57.839
And so you're handling the authentication and the integrations today with what your customers are requesting.
00:19:58.079 --> 00:20:04.640
Authentication, integrations, also building the agents and building the automations, and we're doing more on that side.
00:20:04.880 --> 00:20:08.240
But you're not being the browser use of the browser-based side, an agent.
00:20:08.559 --> 00:20:09.920
No, we use a lot of tooling.
00:20:10.000 --> 00:20:13.599
I mean, we use a lot of any infrastructures in building that space, we adopt.
00:20:13.759 --> 00:20:26.240
We could we'll use people to build proxies, we'll use companies like Browser Base to build headless browsers, we'll use these tools as part of our stack, but we are the ones that are actually building the agents for many of these use cases.
00:20:27.039 --> 00:20:27.599
Interesting.
00:20:27.839 --> 00:20:32.160
How do you decide as the solo founder and CEO which direction to take?
00:20:32.400 --> 00:20:37.359
Do you rely heavily on your CTO, or now that you've done this a few times, you rely on your own intuition?
00:20:37.680 --> 00:20:39.599
From a perspective of like company direction.
00:20:39.759 --> 00:20:39.920
Yeah.
00:20:40.240 --> 00:20:42.000
I think we just learn a lot in market.
00:20:42.079 --> 00:20:56.400
I mean, I would say that one of my hot takes on the computer use space overall is that there are not as many use cases as people think there will be for it, but the use cases that there are will be like deep, deep use cases.
00:20:56.559 --> 00:21:02.319
Like the number of conversations I've gone into recently with people where I say, hey, what are you excited about in the agent space?
00:21:02.400 --> 00:21:03.440
So like computer use models.
00:21:03.519 --> 00:21:07.519
I'm like, great, what is your favorite commercial use case for computer use models?
00:21:07.680 --> 00:21:11.359
And they're like, uh I mean, how would an enterprise use it today?
00:21:11.599 --> 00:21:13.200
What is actually working with computer use?
00:21:13.759 --> 00:21:19.200
Enterprises have huge services teams that support certain functions within the company.
00:21:19.359 --> 00:21:32.079
There are also very big businesses in the RPA space, and there are a lot of people that do a lot of things with antiquated methods, computer use models that are just automated script writing and cron jobs and things like that.
00:21:32.240 --> 00:21:35.680
Like there is a lot of surface area to replace on that side.
00:21:35.839 --> 00:21:45.359
I think what I mean more is that if you go to the operator page within ChatGPT and you look at the use cases, it has all these boilerplate use cases you can scroll through.
00:21:45.440 --> 00:21:50.079
And it's like go order burrito on DoorDash and go book my reservation.
00:21:50.160 --> 00:22:03.599
And I don't know, last year we were getting, I mean, we still get so many ins for our product that people are like, oh, the thing I'm building is a consumer agent, the thing I'm building is a travel agent, the thing I'm Building is a food ordering agent or a date planning agent.
00:22:03.839 --> 00:22:08.240
And I don't know that those will be commercial use cases for computer use.
00:22:08.400 --> 00:22:12.240
And I don't think it has anything to do with the models and where they are.
00:22:12.480 --> 00:22:32.880
When you think about things that are worth automating from a financial perspective, from a commercial perspective, and where it's actually better for a human, you look at big companies like Apple with Siri and Amazon with Alexa that have tried building something that was in this human consumer-proseumer automation space for a long time and have really struggled with it.
00:22:33.119 --> 00:22:35.759
I think it's not all technology related.
00:22:35.920 --> 00:22:43.200
There are some use cases that will be really killer, but I don't think people are going to stop using DoorDash and use DoorDash via chatbot.
00:22:43.440 --> 00:22:46.720
You have to think about your day-to-day and what takes up the most time.
00:22:46.880 --> 00:22:49.200
For me, what takes up a lot of time is writing a memo.
00:22:49.440 --> 00:22:49.599
Yeah.
00:22:50.000 --> 00:22:50.640
That can take hours.
00:22:50.880 --> 00:22:54.240
So therefore, if but that's not, you wouldn't use computer use for that.
00:22:54.319 --> 00:22:58.880
You would likely use GPT based on the research and summarizing the notes you already have.
00:22:59.200 --> 00:22:59.680
Totally.
00:22:59.839 --> 00:23:11.599
But I think the research, I think using computer use for research is the browser web functionality that's been built into Chat GPT and Cloud and all perplexity and all these products is actually really killer functionality.
00:23:11.839 --> 00:23:14.400
But I think, yeah, for taking actions, I I don't know.
00:23:14.480 --> 00:23:24.240
I had somebody who I consider really, really intelligent in this space categorize it like AI is going to be really good where it's replacing jobs, not tasks.
00:23:24.400 --> 00:23:30.720
And if you think about the average person, they don't do that many jobs on a daily basis outside of their work life.
00:23:31.039 --> 00:23:36.559
I don't know how many vacations do people really plan a year, how many times do people really book restaurant reservations?
00:23:36.799 --> 00:23:46.559
I think people get a little mixed up because the type of people that are early adopters of agents actually are taking a lot of trips and booking a lot of things and all these, but the average person doesn't.
00:23:46.640 --> 00:23:49.200
Like I don't think these are mass market consumer products.
00:23:49.440 --> 00:23:52.640
Yeah, and I think what we're seeing work today is some of the tedious, repetitive work.
00:23:52.720 --> 00:23:54.720
Like we're seeing a lot of contacts center automation as well.
00:23:55.359 --> 00:24:01.759
It goes more into the voice AI space, a lot of go-to-market tools, a lot of customer support, and then coding, obviously.
00:24:01.839 --> 00:24:07.759
But these are use cases where you have what good looks like, so you're able to hopefully mimic that into an agentic or AI product.
00:24:07.920 --> 00:24:13.279
But I don't know what happens in some of these more ambiguous spaces where what good looks like isn't necessarily defined.
00:24:13.599 --> 00:24:14.400
Yeah, 100%.
00:24:14.880 --> 00:24:19.519
I think that finding use cases where you know what good looks like is actually a really good way to put it.
00:24:19.839 --> 00:24:26.960
I think that deterministic workflows where you're able to objectively evaluate did the agent do what it was supposed to.
00:24:27.039 --> 00:24:38.720
And there's intuition in the how, but not a lot of intuition in the what is actually the type of use cases that we're personally going after, and that I think are going to be the best ones in the space, at least in the early days.
00:24:38.960 --> 00:24:49.839
So when you think about if you have a market map in your head right now, you have frameworks, hosting, routing, gosh, observability, evals, authentication, all these layers of the agent infrastructure stack, some are converging.
00:24:50.000 --> 00:24:56.960
Do you think there's any layer that's just not going to be needed or that maybe one of the hyperscalers or foundation model companies will just own?
00:24:57.200 --> 00:25:10.160
I think the big question that people are running into is how much is the infrastructure that worked for computers in the past and for building software in the past the same as for building software in the future?
00:25:10.400 --> 00:25:13.279
I think that there are instances where the how is better.
00:25:13.440 --> 00:25:23.039
I think cursors objectively taken over the valley and seemingly the world soon because it's like the how you write software that is better, but you're still writing software.
00:25:23.200 --> 00:25:24.319
Agents are still code.
00:25:24.960 --> 00:25:31.519
Like I mentioned before, we have seen companies that are building what I'll call defensive authentication for agents.
00:25:31.839 --> 00:25:34.000
And I don't actually understand that space.
00:25:34.160 --> 00:25:43.200
I don't understand what people are really building because I actually think that implement the slightly different implementations of OAuth, which is like an existing standard, are sufficient.
00:25:43.440 --> 00:25:47.599
I think that for hosting, a lot of the same tools are sufficient.
00:25:47.759 --> 00:25:55.920
I think for observability, there's maybe some differences, but I think a lot of the original observability tools actually are still pretty useful.
00:25:56.079 --> 00:25:59.839
I mean, I think there's more on that side, depending on like how deep you go.
00:25:59.920 --> 00:26:04.400
And I think there are things that they're very unique about agents in some aspects and not in others.
00:26:04.640 --> 00:26:05.680
But I don't know.
00:26:05.759 --> 00:26:11.759
I don't think the enemy of agents isn't like necessarily people launching new agent products.
00:26:11.839 --> 00:26:17.519
It's just like, are you sufficiently differentiated from traditional infrastructure in building agents?
00:26:17.599 --> 00:26:19.279
Because agents are still just software.
00:26:19.519 --> 00:26:24.720
I think that the difference between machine learning and gen AI, you started to see a new wave of tools coming up.
00:26:24.799 --> 00:26:28.640
But I think a lot of the Gen AI tools you will be able to use for agents.
00:26:28.799 --> 00:26:37.599
What you said with observability evals, I'm seeing that space converge and people just started in one package monitoring logs, testing the agent in production or testing your application in production.
00:26:37.759 --> 00:26:42.960
But I think generally speaking, there'll only be a few nuanced areas to spend something up.
00:26:43.119 --> 00:26:45.200
Do you price based on outcomes today?
00:26:45.680 --> 00:26:49.920
We price based on usage today, but usage is heavily tied to outcomes.
00:26:50.240 --> 00:26:50.480
Okay.
00:26:50.720 --> 00:26:52.000
Can you unpack that a little bit?
00:26:52.240 --> 00:26:52.880
What that means?
00:26:53.359 --> 00:26:58.079
We're facilitating specific workflows for and we charge like per workflow.
00:26:58.319 --> 00:27:01.680
So we charge a monthly base rate as a platform fee.
00:27:01.839 --> 00:27:05.440
And then the more you use the product, the more you pay for it.
00:27:05.680 --> 00:27:09.039
Similar to like I mentioned Plat earlier, similar pricing to that.
00:27:09.200 --> 00:27:15.440
The more data you pull, the more people are paying us to basically go and retrieve data and complex workflows for them.
00:27:15.920 --> 00:27:25.680
Do you think with some of these more of the application layer of agents, do you think this pricing debate that's going on, there actually will be a pricing evolution?
00:27:25.920 --> 00:27:31.119
Or do you think it'll still be what we've been seeing with SaaS for the last decade?
00:27:31.519 --> 00:27:35.440
I think that we'll price agents like a human, not software, is silly.
00:27:36.240 --> 00:27:50.559
Actually, I actually want to get your take on, I haven't unpacked this with anyone yet, but like Cognition's recent pricing change, where they went from like 200 a month, or no, was it 500 per seat, I guess, to now I think they're trying to match cursor's model closely.
00:27:50.799 --> 00:28:01.039
Do you think it's just because the products that are on the market today, they're just evolving so quickly and advancing and they're offering a lower price point for the same capabilities as your autonomous AI engineer.
00:28:01.200 --> 00:28:08.000
So therefore the price basically converges down to where the standard of what good looks like, which is cursor right now, might change.
00:28:08.480 --> 00:28:10.319
I don't know that it has anything to do with cursor.
00:28:10.400 --> 00:28:15.599
I would just say pricing dynamics are driven by the same market dynamics that have always driven markets.
00:28:15.759 --> 00:28:21.279
The product costs a certain amount to build, it's worth a certain amount to the end customer.
00:28:21.599 --> 00:28:29.039
If someone tries to charge$500 a month and their competitor is charging$20 a month, everyone's going to choose their competitor.
00:28:29.200 --> 00:28:36.079
And the product costs, you know, a few dollars a month to deliver, probably in terms of infrastructure costs.
00:28:36.319 --> 00:28:49.039
So it's just like if you were the only guy mowing lawns in town and you're the only lawn mower and you charged$1,000 a yard, and then someone could bought a second lawn mower and charging$50 a yard, your pricing has to come down to match that.
00:28:49.279 --> 00:28:54.799
Pricing has nothing to do with the presentation of it or whether you give your agent a name, like, hey, this is Doug, my agent.
00:28:54.880 --> 00:28:55.680
He's$100,000.
00:28:55.920 --> 00:29:03.119
Like it just has to do with the value you're delivering and what the cost relatively to other competitors into the market is.
00:29:03.440 --> 00:29:08.079
It sounds like something you're alluding to is it also just comes down to execution and go to market speed.
00:29:08.640 --> 00:29:09.359
Yeah, yeah.
00:29:09.519 --> 00:29:12.559
I think markets behave like markets behave in a lot of ways.
00:29:12.720 --> 00:29:19.519
It's like a very just like baseline economics, having your agents seem more like a human versus not, or having it be.
00:29:19.759 --> 00:29:30.480
If somebody comes up with something that adds insane value and it can't be replicated by any competitors for whatever reason, then they're going to be able to charge an ungodly amount of money for it.
00:29:30.640 --> 00:29:33.839
But as soon as you have market competition, it's going to drive prices down.
00:29:34.079 --> 00:29:36.480
Do you have a product you admire that charges a lot?
00:29:36.720 --> 00:29:41.680
A product that has that's like unparalleled in terms of it works well, you would pay double for it today.
00:29:42.079 --> 00:29:43.920
I think this space is moving so fast.
00:29:44.160 --> 00:29:50.160
I've talked before, I was a big, I was a big early adopter of Claude and the anthropic products.
00:29:50.559 --> 00:30:00.880
Like I, you know, there was a point in time I was probably using it like eight hours a day in my work to write everything I wrote, to send memos, send messages, plans, write PRDs, like everything we do.
00:30:01.119 --> 00:30:07.759
It's still good, but I think that ChatGBT has launched a lot of improvements recently and has come to parity with it, maybe even better.
00:30:07.839 --> 00:30:12.319
I think these products, like again, they actually drive so much outsized value.
00:30:12.480 --> 00:30:21.759
If I objectively were how much productivity do I get by using these products on a daily basis, then I would pay them tens of thousands of dollars a year for a personal license.
00:30:21.920 --> 00:30:27.519
But luckily, there's competition in the market, so they drive each other's prices down and you don't have to worry about it.
00:30:27.599 --> 00:30:29.839
So do you use perplexity?
00:30:30.079 --> 00:30:30.640
Yeah.
00:30:30.960 --> 00:30:31.839
Over Google?
00:30:32.079 --> 00:30:32.559
Yeah.
00:30:32.799 --> 00:30:34.160
I really like it.
00:30:34.480 --> 00:30:35.279
Why is that?
00:30:35.680 --> 00:30:39.839
I really like it because I think it gives the answer you're looking for.
00:30:40.160 --> 00:30:46.559
I think Google's product, I mean, I think Google's done a lot of amazing work, but objectively, Google's selling.
00:30:46.799 --> 00:30:49.839
Google's so heavily gamified at this point.
00:30:50.000 --> 00:30:58.240
Yeah, with ads and with SEO, if you're like, you know, what's the best, you know, give me five pizza places in South Park to go to lunch?
00:30:58.319 --> 00:31:00.880
It's like who's bidding the most, who hired the agency.
00:31:01.039 --> 00:31:05.200
I mean, my consulting business I built years ago, we did a lot of SEO work, game Google.
00:31:05.279 --> 00:31:06.480
There's a whole thing around it.
00:31:06.640 --> 00:31:08.480
I think Perplexity's results are cleaner.
00:31:08.559 --> 00:31:11.119
It gives me the thing I'm looking for, it has citations.
00:31:11.359 --> 00:31:14.640
I don't know how they monetize, but beyond that, I think it's great.
00:31:14.880 --> 00:31:23.839
Do you think people will start to game Perplexity, GPT, all of the foundation model consumer products with this new wave of AI SEO companies?
00:31:24.160 --> 00:31:26.000
I think that is the question, right?
00:31:26.160 --> 00:31:30.960
Because Google, it's honestly this trade-off between experience and monetization.
00:31:31.200 --> 00:31:36.400
Google is the best monetization machine that's ever been built, and it's heavily gamed.
00:31:36.559 --> 00:31:51.440
I think actually Google in hindsight probably should have spent more time thinking about UX and less about monetization, maybe over the last like, and they might not be in as precarious, not that they're in a precarious position, but they have they have people sieging them for the first time, right?
00:31:51.599 --> 00:31:53.279
Their business is under attack.
00:31:53.519 --> 00:31:59.359
And right now, perplexity is pristine in that there are no ads and it gives you objectively the right results.
00:31:59.599 --> 00:32:02.400
But I think there's a real question of monetization on that.
00:32:02.559 --> 00:32:10.480
And maybe they meet more in the middle at some point, and Google has to retreat from their, you know, money printing machine to deliver a better experience.
00:32:10.640 --> 00:32:15.920
And probably perplexity has to build some ads and some bias and Google bias perplexity.
00:32:16.160 --> 00:32:17.519
Yeah, I mean, totally.
00:32:17.680 --> 00:32:18.480
I think, yeah.
00:32:18.720 --> 00:32:18.960
Okay.
00:32:19.039 --> 00:32:20.559
Do you want to end with a quick fire round?
00:32:20.720 --> 00:32:22.480
We went a little off script, but that's okay.
00:32:22.720 --> 00:32:23.039
Totally.
00:32:23.200 --> 00:32:25.920
Actually, I don't think I asked any questions off script.
00:32:26.480 --> 00:32:29.440
What's the one word to describe your company culture today?
00:32:29.680 --> 00:32:30.720
I would say fun.
00:32:31.279 --> 00:32:34.480
I think that we really are trying to build a culture.
00:32:34.720 --> 00:32:39.680
A big learning I had from my last company is that it's a journey.
00:32:39.839 --> 00:32:42.559
Things are gonna go well, things are not gonna go well.
00:32:42.880 --> 00:32:48.400
And over time, the thing that is consistent is the people that are on the table through those good times and bad.
00:32:48.480 --> 00:32:52.880
So we really try to make time to have a good time with each other, bond with each other.
00:32:53.039 --> 00:32:55.759
We work really hard, we spend a lot of hours in the office.
00:32:56.000 --> 00:32:58.720
But last night we had an amazing customer demo.
00:32:58.880 --> 00:33:02.799
We killed it and we did a celebratory game night and hung out at the office.
00:33:02.960 --> 00:33:05.200
And I think it's really important to find that balance.
00:33:05.599 --> 00:33:11.359
What's one or two attributes of a new candidate that you're looking at that you really admire?
00:33:11.680 --> 00:33:16.880
So, again, based on I'd say experience, we hire people that have done the thing before.
00:33:17.039 --> 00:33:20.240
And what I mean by that is we're hiring a lot of engineers right now.
00:33:20.400 --> 00:33:22.799
So, any great engineers watching, reach out to me.
00:33:22.960 --> 00:33:26.240
We want to hire startup engineers that have built startups before.
00:33:26.400 --> 00:33:30.160
And we want to hire people to lead teams that have led teams before.
00:33:30.400 --> 00:33:38.720
We want to hire people at this stage we're at right now where, you know, we're early but growing really quickly that have been early and grown really quickly before.
00:33:38.880 --> 00:33:48.480
I think you get a lot of interviews where someone's like, Yeah, I've worked at, you know, meta for the last eight years, but I really want to go work at a startup because I want more ownership.
00:33:48.640 --> 00:33:53.759
And, you know, it's not that I don't believe them or I think they're being disingenuous, but I don't think they know what that means.
00:33:53.920 --> 00:33:59.440
I think until you've sat in that exact seat, you actually don't really know what you're talking about.
00:33:59.680 --> 00:34:03.519
So I want people that have sat in that seat before and done that exact thing before.
00:34:03.599 --> 00:34:06.880
And it usually means you're finding more like diamond in the rough type of candidates.
00:34:07.119 --> 00:34:10.400
They don't have the big logos, they probably don't have the schools.
00:34:10.559 --> 00:34:13.679
You know, we're not averse to that, and we will hire people like that.
00:34:13.840 --> 00:34:26.079
But the our big criteria is that you're hardworking, you're humble, you've done the thing you're doing before, and you're motivated by the mission, the team, the journey, not the equity price or the comp or things like that.
00:34:26.400 --> 00:34:29.199
And what's an AI tool that people aren't talking enough about?
00:34:29.440 --> 00:34:32.320
An AI tool that people aren't talking enough about, anon.
00:34:32.800 --> 00:34:34.639
Will passwords ever fully die?
00:34:34.960 --> 00:34:36.880
Will passwords ever fully die?
00:34:37.199 --> 00:34:38.880
You know, maybe.
00:34:39.280 --> 00:34:56.159
I you know what's a funny thing is that we had all this move to certain types of passwordless auth, pass keys and stuff, which I think are okay, and CAPTCHAs, which you know now are getting solved at a lower rate with humans and AI, like AI is actually better at solving.
00:34:56.320 --> 00:35:08.960
The funny thing is the thing that is actually really effective in handling did this user actually really want this other service to use the thing that they're using is old school two-factor authentication.
00:35:09.199 --> 00:35:19.199
It's sending a text message to the person and having them enter that code in because it establishes a physical presence of a user holding a phone and access to the device.
00:35:19.440 --> 00:35:22.079
People don't change phone numbers that often, it's hard to hack.
00:35:22.239 --> 00:35:30.239
So it's funny, it's like some things never change, and that auth does, but I actually think passwords will be around for a pretty long period of time.
00:35:30.559 --> 00:35:31.039
Awesome.
00:35:31.119 --> 00:35:31.679
Well, thank you.
00:35:31.840 --> 00:35:33.039
Yeah, that was fun.
