TRANSCRIPTION 🔗
WEBVTT
00:00:00.000 --> 00:00:02.240
The first era of AI was about building intelligence.
00:00:02.319 --> 00:00:03.359
The second is controlling it.
00:00:03.439 --> 00:00:05.599
You know, routing, governance, security, private emails.
00:00:06.080 --> 00:00:08.480
What people didn't realize was that these things can reason.
00:00:08.560 --> 00:00:13.199
And that's to take ten vulnerabilities that might have been on the backlog and you stick staple them together and create an attack path.
00:00:13.359 --> 00:00:15.439
When you create an attack path and these things can reason?
00:00:15.519 --> 00:00:16.160
I mean, that's crazy.
00:00:16.320 --> 00:00:20.079
We're seeing like the industrialization of vulnerability discovery and exploitation with AI.
00:00:20.239 --> 00:00:25.760
I think the biggest heist ever happening right now is that OpenAI and Anthropic created their own four deployed engineering companies.
00:00:25.839 --> 00:00:26.559
What do you think they're gonna do?
00:00:26.640 --> 00:00:28.559
They're gonna afford the deployed engineer into your company.
00:00:28.640 --> 00:00:30.000
They're gonna build your workflows.
00:00:30.160 --> 00:00:31.440
I don't fucking want that, right?
00:00:31.600 --> 00:00:35.119
Ultimately, Chris, you and I both know that where this world is going to is a world of outcomes.
00:00:35.200 --> 00:00:36.159
Like they're just cut to the chase.
00:00:36.240 --> 00:00:40.159
We can talk about tech all we want, but it's like the only way you can keep up with it is that agents fix all these problems.
00:00:40.320 --> 00:00:41.759
That is like a brand new security problem.
00:00:41.840 --> 00:00:42.799
Like, who's building for that?
00:00:43.039 --> 00:00:46.079
Like, think about Anthropic, there are AI people that created the security stuff.
00:00:46.159 --> 00:00:49.119
Like, I mean, there are rumors that they actually run themselves autonomously in there.
00:00:49.200 --> 00:00:51.759
So, like, you gotta think so fucking crazy.
00:01:05.359 --> 00:01:07.359
Thank you for joining the Resilient Cyber Show.
00:01:07.439 --> 00:01:10.480
My name is Chris Hughes, and today I'm joined by Ed Sim from Bold Start Ventures.
00:01:10.560 --> 00:01:11.200
Ed, thanks for being here.
00:01:11.359 --> 00:01:12.480
Hey man, thanks for having me on.
00:01:12.560 --> 00:01:14.239
I think this is our second time, so thanks.
00:01:14.560 --> 00:01:16.719
Yeah, it was about a year ago, roughly, at this time, I think.
00:01:16.799 --> 00:01:19.040
Uh not exactly, but you know, it's been a little bit of time.
00:01:19.120 --> 00:01:25.840
But for folks that don't follow you on LinkedIn, don't follow your newsletter, you know, don't follow your investments and so on, tell us a bit about yourself and the team.
00:01:26.159 --> 00:01:30.560
Yeah, so uh I've been investing uh in the very first kind of rounds of companies.
00:01:30.640 --> 00:01:33.920
I call that inception investing since uh, I guess 30 years now.
00:01:34.079 --> 00:01:38.239
And so I I invest in AI infrastructure, cybersecurity, and physical AI.
00:01:38.400 --> 00:01:43.760
So my uh investment kind of thesis kind of spans the gamut of of hardcore technology, right?
00:01:43.920 --> 00:01:49.519
And so cyberwise, we've been the very first investor, the boards of companies like Protect AI sold to Palo Alto.
00:01:49.599 --> 00:01:52.799
We did that like a year before ChatGPT even came out, so we can talk about that.
00:01:52.959 --> 00:02:04.319
And then newer ones are like Keycard, which is building kind of agent authentication authorization, uh, Surf AI, uh, which uh just closed a new round with Excel and we did that with Cyberstarts at the beginning and a bunch of others.
00:02:04.400 --> 00:02:06.000
So really, really excited uh to be here.
00:02:06.239 --> 00:02:12.000
Yeah, I was telling you off air before we jumped on, I read a lot of your content because you know, I I live in a cybersecurity echo chamber.
00:02:12.080 --> 00:02:13.919
Like everyone's all all things cybersecurity.
00:02:14.000 --> 00:02:24.719
And it's nice to step out and see the broader, you know, uh CTO, CIO, investor perspective, you know, and I watch the market and security in terms of startups and venture capital and all that, and you have uh some great resources around there.
00:02:24.879 --> 00:02:27.680
Um so you you made this point that you've been investing for almost 30 years.
00:02:27.840 --> 00:02:31.680
I wanted to start, I'm gonna hit on all the things you just brought up, sneak, protect, surf, etc.
00:02:31.919 --> 00:02:43.039
But you know, in this era where a founder can vibe code MVPs, you know, and and like what does a day one partnership look like anymore compared to maybe eras past before before like agentic coding and things like that kind of were mainstream?
00:02:43.280 --> 00:02:44.479
Yeah, no, I think that's a great question.
00:02:44.560 --> 00:02:45.759
We're just talking about this earlier.
00:02:45.840 --> 00:02:47.759
Look, at the end of the day, it it's amazing.
00:02:48.000 --> 00:02:49.599
Anyone can create anything, right?
00:02:49.680 --> 00:02:52.000
And so I think it's the best time to be a founder and to build.
00:02:52.240 --> 00:03:02.080
On the other hand, in the years past, when you had an idea 10 years ago, 15 years ago, you build the software company, you figure, oh gee, maybe my moat is six months, 12 months, maybe 18 months, right?
00:03:02.240 --> 00:03:04.960
But now things can get copied very, very fast.
00:03:05.199 --> 00:03:10.000
So when we look at founders at Inception, there are a couple ways to make money in cybersecurity.
00:03:10.159 --> 00:03:14.080
One is take something that already exists and reinvent it and make it 10 times better, right?
00:03:14.159 --> 00:03:15.599
And which is getting harder and harder.
00:03:15.680 --> 00:03:20.240
But you're seeing a lot of people do that who are second and third time founders, perhaps raising$100 million.
00:03:20.800 -->
Load full transcript
00:03:25.680
I mean, some massive rounds of funding to say, I'm gonna do this AI native or agent agent native, right?
00:03:25.840 --> 00:03:28.800
And the other is to go after new attack paths and new attack factors.
00:03:28.879 --> 00:03:32.159
But the risk there is that the market's very early, there's probably no budget for it.
00:03:32.319 --> 00:03:36.800
But if things do take off, like a Protect AI, we funded that a year before ChatGPT took off.
00:03:36.960 --> 00:03:42.240
We were there at the right time at the right place, made the right acquisitions, and we're able to exit forever 700 million to Palo.
00:03:42.560 --> 00:03:47.520
In today's world, though, what we look at is because I gotta tell you, the world doesn't need all the cybersecurity companies.
00:03:48.000 --> 00:03:51.919
The world needs more cybersecurity, but we don't need all the cybersecurity companies that we have right now.
00:03:52.080 --> 00:03:53.360
Just flat out.
00:03:53.520 --> 00:04:02.319
And I think the biggest challenge is that people can have a great cyber idea, but what I really look for underneath that is the depth on the AI uh strengths, right?
00:04:02.400 --> 00:04:04.000
So vibe coding, it's not about vibe coding.
00:04:04.080 --> 00:04:06.319
I want to know kind of guys like, are you gonna build your own model?
00:04:06.400 --> 00:04:07.599
Are you gonna post-train your own model?
00:04:07.840 --> 00:04:12.719
Like, how deep are you going to go in terms of kind of building emote and building a flywheel?
00:04:12.800 --> 00:04:15.120
And how deeper you can go into those customer environments, right?
00:04:15.199 --> 00:04:19.759
So I like to know that you can actually hire some hardcore AI folks like researchers, et cetera, as well.
00:04:19.839 --> 00:04:24.079
Like Surf AI has got a massive team of researchers they they've uh accumulated together.
00:04:24.160 --> 00:04:25.680
So that that's something that's very big for us.
00:04:25.759 --> 00:04:31.120
And then you look at what lens and uh am I gonna reinvent something or am I looking at a new attack factor, right?
00:04:31.199 --> 00:04:32.240
So that's how we think about it.
00:04:32.480 --> 00:04:34.160
Yeah, you've so you brought up, you know, protect.
00:04:34.319 --> 00:04:40.720
I wanted to uh ask you about that, that acquisition of Palo Alto, obviously that you know, kind of very, very early, as you said, before even ChatGPT.
00:04:40.800 --> 00:04:44.160
Um, you know, I think you called it the first exit in AI security history.
00:04:44.399 --> 00:04:50.639
Like uh, you know, and there's a lot of to your comment about we don't need all the cybersecurity companies, there are a ton of cybersecurity companies.
00:04:50.800 --> 00:04:56.319
I can tell you as a practitioner, like, you know, a lot of times for security leaders, CISOs is spent just navigating, like, what are all these products?
00:04:56.399 --> 00:04:57.199
How do I consolidate?
00:04:57.279 --> 00:04:58.560
How do I get ROI?
00:04:58.720 --> 00:05:01.199
Like, how do I kind of rationalize my tech stack?
00:05:01.360 --> 00:05:09.519
Um, you know, when should an AI security founder, looking at the founder's perspective, uh sell into a consolidator versus keep building towards uh being the bigger standalone platform?
00:05:09.600 --> 00:05:10.560
How do you think about that?
00:05:10.879 --> 00:05:28.399
Yeah, I I think at every point in time when a founder is out there um raising capital, uh, it's also a a perspective up, gee, you know, maybe I should consider like, do I take$50,$100 million more at a higher price where my prep stack goes higher, where the new round investor thinks I should get a 2x on top of 700 million.
00:05:28.480 --> 00:05:30.240
So now I have to sell for 1.5 billion?
00:05:30.399 --> 00:05:40.319
Or do I kind of consider like, gee, my last round was done at$300,000,$400 million, and then I can make everyone, you know, a fair amount of return back right now, and actually I can probably do myself a better bit.
00:05:40.399 --> 00:05:42.959
And I think it, I think every decision is kind of independent.
00:05:43.040 --> 00:05:45.199
It's independent of who the founder is.
00:05:45.360 --> 00:05:47.839
It's also dependent on kind of where the market is, right?
00:05:47.920 --> 00:05:56.560
For example, in Protect's case, you know, we were out there, we're cranking along, um, having a number of new enterprise customers, but we also felt there are a number of competitors coming down the pike.
00:05:56.639 --> 00:06:00.319
And a lot of them weren't as good as what we did, but it just slows things down.
00:06:00.480 --> 00:06:05.439
And the more we talk to Palo, and then Palo's like, look, we're gonna get in the space no matter what, whether it's you or someone else.
00:06:05.519 --> 00:06:10.879
So you start doing the math, and we're like, gee, at that price, you know, everyone's gonna do really well.
00:06:10.959 --> 00:06:16.879
And frankly, if you look at it, a lot of times the first one is usually the highest one, and then eventually everything else is lower after that.
00:06:16.959 --> 00:06:24.560
And then maybe after that, when the market matures, you and that's kind of the first three years, is like the first one's the highest, and then people are just buying the tech, which they did.
00:06:24.639 --> 00:06:26.560
It started a wave of 12 or 15 acquisitions.
00:06:26.639 --> 00:06:33.439
But then once you start seeing who the leaders are, what what real revenue comes in, then they kind of value it based on revenue, and then the numbers make it higher, right?
00:06:33.519 --> 00:06:34.240
If the market is real.
00:06:34.319 --> 00:06:39.839
And so that was kind of it for us was like looking at that situation, and the founders are like, look, this is probably the best thing we can do.
00:06:39.920 --> 00:06:41.600
And they are killing it in there, honestly.
00:06:41.680 --> 00:06:45.519
I don't know if you see or hear, but they're selling the crap
out of that of that software.
00:06:45.839 --> 00:06:59.439
Yeah, I mean, there's so many things to even we can do a whole episode on that topic alone is like to your point in the beginning, it's kind of like, you know, you know you're gonna be the benchmark for subsequent acquisitions in that space and and like, you know, until revenues kind of normalize and start to become transparent.
00:06:59.519 --> 00:07:03.920
Um and then uh, you know, obviously folks like Lee made a lot of money, some of the early employees there.
00:07:04.000 --> 00:07:05.839
I mean, the whole firm was early at that point.
00:07:06.079 --> 00:07:12.079
And then also like having been on the product vendor side, uh, you're gonna be facing Palo whether it's with them or against them.
00:07:12.240 --> 00:07:16.000
And it's gonna cause a lot of friction within your sales cycles, you know, in big enterprises and so on.
00:07:16.079 --> 00:07:22.639
And they have, you know, they have that massive distribution, they have the brand recognition, like they have all these things that you're gonna be facing in these deals when you go out there.
00:07:22.720 --> 00:07:25.680
So it made a lot of sense if you think about it from that perspective as you framed it.
00:07:25.920 --> 00:07:37.680
I wanted to ask you too, like, you know, you talk about being an inception fund, uh, and you know, we've seen some of the you you talked about this in the opening, like some of these insane, you know, uh round size, round sizes, even seed rounds and so on.
00:07:37.759 --> 00:07:46.560
Like, you know, how do you think about competing in this space as an inception fund, you know, against some of the mega funds out there uh with with these unheard of kind of seed rounds and so on that are going on?
00:07:46.879 --> 00:07:47.759
Yeah, it's kind of crazy.
00:07:47.839 --> 00:07:53.279
I was looking at some stat from Carta where the 99th percentile of these seed rounds are like I saw that this week too.
00:07:53.360 --> 00:07:53.759
Yeah, yeah.
00:07:53.920 --> 00:07:57.279
Yeah, I was it 250, 250 million or something, or 500 million?
00:07:57.360 --> 00:07:58.720
I mean, the valuations were insane.
00:07:58.879 --> 00:08:01.600
And look, from our perspective, uh a couple of things.
00:08:01.839 --> 00:08:09.600
One is I always believe there's gonna be an opportunity for people to be first and to be kind of courageous and to have the conviction with the founder.
00:08:09.759 --> 00:08:12.240
Um, and just the way we've done things, we feel very comfortable.
00:08:12.319 --> 00:08:13.199
I've been doing this for 30 years.
00:08:13.360 --> 00:08:15.040
Bold start's been doing this since 2010.
00:08:15.199 --> 00:08:17.759
We can look a founder in the eye, we can diligence who they are.
00:08:17.920 --> 00:08:23.279
They may have nothing, but we can just understand kind of that, you know, these people have grit, that they're gonna build something.
00:08:23.360 --> 00:08:26.240
They have a unique insight that that has been built over 10 years.
00:08:26.319 --> 00:08:30.240
And and we've seen enough things where we can actually commit quicker than most funds.
00:08:30.399 --> 00:08:37.360
And I gotta tell you, like when you do that, the goodwill you build with the founder, they will never, you are my first kind of person that just said, I'm in.
00:08:37.519 --> 00:08:39.519
And a lot of times what we'll do is we'll catalyst.
00:08:39.600 --> 00:08:40.639
I call that ball control.
00:08:40.799 --> 00:08:43.120
In any sport you play, you want to have ball control.
00:08:43.279 --> 00:08:46.080
So ball control for us is moving as fast as possible.
00:08:46.159 --> 00:08:53.360
If it requires 48 hours on the weekend for us to do our diligence and make calls, but you know what the most interesting thing is because we do what we do.
00:08:53.519 --> 00:08:56.320
I mean, look, about a third of our investments are in cyber.
00:08:56.480 --> 00:08:58.879
The founders come to us and they've already diligenced us.
00:08:58.960 --> 00:08:59.919
They already know who we are.
00:09:00.080 --> 00:09:03.519
And so, and and they probably diligence us faster than we can diligence them.
00:09:03.679 --> 00:09:06.399
So there's already this kind of wanting to work together.
00:09:06.559 --> 00:09:08.399
It probably got referred to another founder.
00:09:08.480 --> 00:09:09.679
So the credentials are kind of there.
00:09:09.759 --> 00:09:10.960
And look, we invest in people.
00:09:11.120 --> 00:09:18.399
We want high-slipping people that can come in and tell a good story, but then also can build the foundation of you know, a big, big idea, right?
00:09:18.480 --> 00:09:21.360
A big idea, but maybe it starts as a wedge and and they grow over time.
00:09:21.519 --> 00:09:23.360
So anyway, I think that's that that's that.
00:09:23.519 --> 00:09:25.120
And then too is we have upsized refunds.
00:09:25.200 --> 00:09:28.240
I mean, like we can, you know, write checks uh in the 15th.
00:09:28.320 --> 00:09:31.440
I can I could split a$30 million round by the big fund, which we have.
00:09:31.600 --> 00:09:37.519
Uh we did a$25 million round we split with G V kind of in in Tesla, which was Guy Pajarni from Sneaks Company.
00:09:37.679 --> 00:09:45.440
Um so we can regularly do that, but then also we can write smaller checks too, because we're not so big that a million, two million, three million dollar check is not going to matter to us because it does.
00:09:45.519 --> 00:09:47.039
And so that's kind of the balance that we strike.
00:09:47.360 --> 00:09:47.919
Yeah, that makes sense.
00:09:48.000 --> 00:09:50.080
And honestly, it's funny you brought up uh Tesla.
00:09:50.159 --> 00:09:55.440
I've watched a ton of uh guys' content and like they're doing amazing work already on their new initiative and new venture.
00:09:55.600 --> 00:10:01.840
And uh you know, I feel like they're building on the momentum of Sneak and the developer-centric type approach, but now with the new
agency uh SDLC and all that.
00:10:02.000 --> 00:10:06.000
Like I watch a lot of the stuff that he's doing, um, and I think they're gonna do incredibly well too.
00:10:06.159 --> 00:10:11.759
Uh you I just want to dig a little deeper on something you said there for a moment ago, was like uh the founder's diligencing you as well.
00:10:11.919 --> 00:10:17.600
Like, you know, typically it's like you're look, you're you're doing due diligence on the founder's right to make a bet with with placing funds and so on.
00:10:17.840 --> 00:10:24.159
Um, what is what do you think founders look for in a in a venture partner that kind of helps differentiate who they select as well?
00:10:24.480 --> 00:10:33.919
I I think first and foremost is is this a is this a guy, a man and woman or a partner that I'm gonna want to call at midnight or take a call from at midnight?
00:10:34.080 --> 00:10:35.360
Um you know, do I like them?
00:10:35.600 --> 00:10:35.679
Right?
00:10:35.840 --> 00:10:40.639
I mean, because like everyone's money is green, and most of the investments that we make will be competitive, right?
00:10:40.799 --> 00:10:42.320
There'll be multiple term sheets on the line.
00:10:42.399 --> 00:10:44.480
And and first you gotta like and same with us.
00:10:44.559 --> 00:10:46.240
So like, do I like working with this person?
00:10:46.320 --> 00:10:48.399
Sometimes maybe that just a fit's not there, right?
00:10:48.559 --> 00:10:55.360
And so if there's a fit, that's that's actually kind of because look, I think the average venture deal lasts longer than the average marriage out there, which is kind of crazy, right?
00:10:55.519 --> 00:11:00.000
So I think I think people underestimate how important the people, the people aspect are.
00:11:00.240 --> 00:11:01.600
And then all the other things come after that.
00:11:01.759 --> 00:11:03.440
Yeah, same amount of money, that's fine.
00:11:03.759 --> 00:11:05.279
Valuation is all within a band.
00:11:05.759 --> 00:11:07.679
That's not kind of less less important.
00:11:08.000 --> 00:11:10.240
Um that's like kind of what are you gonna do for me?
00:11:10.480 --> 00:11:22.080
When you do inception investing, I tell you, Chris, you can't put a lens of a multi-stage firm on a on an inception investment because you know, maybe you know, because people sometimes worry about how quickly do you get to revenue and how quickly do you scale.
00:11:22.159 --> 00:11:28.720
Maybe you need to spend that extra three months saying, I should build this thing and make it a little bit deeper technically, or if I spend this extra time.
00:11:28.879 --> 00:11:34.240
And so having that patience, I think, is really important, knowing when to get out of the way and not bother them, right?
00:11:34.399 --> 00:11:37.120
Because I've seen kind of investors just kind of always hover over people.
00:11:37.200 --> 00:11:38.240
It's like, look, go build.
00:11:38.399 --> 00:11:42.000
Uh I'm not if I'm here to help you build, that's the wrong investor, right?
00:11:42.159 --> 00:11:44.240
Because I'm in I'm investing in you to build something.
00:11:44.399 --> 00:11:45.840
And then two is of course the added value.
00:11:45.919 --> 00:11:53.279
We can you open uh relationships to the right CISOs or the right kind of um IT executives and can you actually help me hire people, right?
00:11:53.360 --> 00:12:01.120
Because a lot of times I spend time interviewing um, you know, let's say heads of sales or heads of marketing or heads of engineering, just sharing the patterns that we had before.
00:12:01.200 --> 00:12:10.720
And I'm not the final say, but they like to have kind of our word, and of course you end up trying to sell them to like, hey, I've backed multiple companies, this one's gonna be a massive one, um, and therefore you join.
00:12:10.799 --> 00:12:12.320
So those are things that we like to do.
00:12:12.480 --> 00:12:14.720
But I think that kind of is the most important thing.
00:12:14.879 --> 00:12:16.320
Um relationship.
00:12:16.480 --> 00:12:18.240
And people get that in the world of AI.
00:12:18.639 --> 00:12:19.679
No, I loved hearing that part.
00:12:19.759 --> 00:12:20.960
That's why I want to dig deeper in it.
00:12:21.039 --> 00:12:22.799
Is like not all capital is created equally.
00:12:22.879 --> 00:12:29.279
Like there's dollars, but there's the relationship, there's the people, there's the experience of having been there and done this before and helped companies uh succeed.
00:12:29.440 --> 00:12:36.559
There's the CISO network and the access to the buyers and the distribution opportunities as well as well that people don't think about.
00:12:36.720 --> 00:12:47.200
Um, I do want to pivot us to something you and I were riffing on a little bit off air, is like, you know, the a big thesis around neo clouds and self-hosting and open source and control of proprietary data and uh intelligence, et cetera.
00:12:47.360 --> 00:12:54.080
Like, you know, there's a big uh core organ argument you've been making lately, I think, where uh the first era of AI was about building intelligence.
00:12:54.159 --> 00:12:57.759
The second is controlling it, you know, routing, governance, security, private e-fails.
00:12:58.080 --> 00:13:03.600
Uh if intelligence is becoming a commodity, like where in the control area are the venture venture scale scale companies?
00:13:03.679 --> 00:13:04.399
How do you think about it?
00:13:04.720 --> 00:13:11.759
Well, look, I I think the best thing that ever happened to any software company is that uh anthropic and open AI filed to go public.
00:13:11.919 --> 00:13:18.080
Uh and the reason is is because you know, they're like, holy crap, I I actually have to start maybe not subsidizing my margins as much.
00:13:18.240 --> 00:13:18.399
Right.
00:13:18.480 --> 00:13:28.799
And and all of a sudden there's a wake-up call in the world saying, you know what, man, the tokenomics, you know, the cost of actually using all these models just kind of we blew
through our budgets.
00:13:28.879 --> 00:13:31.120
And you could hear kind of people talking about it repeatedly, right?
00:13:31.200 --> 00:13:32.080
So that was part one.
00:13:32.320 --> 00:13:43.440
Part two was then like, okay, if I'm doing that, then maybe I can actually staple together models that I don't need to use state of the art or soda models all the time for everything, right?
00:13:43.519 --> 00:13:45.840
So then this whole router kind of industry got created, right?
00:13:45.919 --> 00:14:02.399
So so even look, some of the largest banks I know were already building kind of models where they would actually do a query and it would automatically route, maybe if it was a very important question to state of the art, and then they would have subagents underneath kind of managing kind of Kimmy or DeepSeek or GLM 5.2 or whatever, right?
00:14:02.480 --> 00:14:05.759
So so then you kind of now this whole industry is spawning around that.
00:14:05.919 --> 00:14:07.679
I mean, Spectre Cloud does some of that as well.
00:14:07.759 --> 00:14:10.799
And they're gonna have a big announcement in a week or two uh about this.
00:14:11.039 --> 00:14:24.879
But but so now and then the other thing I think is that people are proving that if you post-train models, uh, take an open source model, uh, add some more data to it, do some more reinforcement learning around it by creating the right rewards for certain kind of outcomes.
00:14:25.039 --> 00:14:26.799
In security, that's a really interesting area, right?
00:14:26.879 --> 00:14:36.720
For example, it's like um if I fix, if I find this vulnerability and then I can uh daisy chain multiple vulnerabilities together and turn that into a uh into a like a real hack, right?
00:14:36.799 --> 00:14:41.679
Kind of a path, you know, um if I create a patch, you know, does that work, right?
00:14:41.840 --> 00:14:43.360
Like in code, you know code works.
00:14:43.440 --> 00:14:46.399
And in security also, if you patch something, you can know that works, right?
00:14:46.480 --> 00:14:49.279
So there's reinforcement learning and reward feedback that can be very interesting.
00:14:49.519 --> 00:15:02.559
So I'm seeing more and more people now saying, okay, like now, uh it costs a lot, I can do routing, and then holy shit, why am I giving my data over to all these um you know vendors?
00:15:02.639 --> 00:15:04.320
Why am I giving it to OpenAI and Anthropic?
00:15:04.399 --> 00:15:11.759
Like, I think the biggest heist ever happening right now is that OpenAI and and uh Anthropic created their own uh Ford deployed engineering companies.
00:15:11.840 --> 00:15:12.480
What do you think they're gonna do?
00:15:12.559 --> 00:15:19.440
They're gonna forward deployed engineering to your company, they're gonna build your workflows, and they're gonna take your data and then use that to sell to every other competitor.
00:15:19.600 --> 00:15:21.039
I don't fucking want that, right?
00:15:21.120 --> 00:15:22.559
And so so I think that's another thing.
00:15:22.639 --> 00:15:25.519
So Satya keeps talking about, which he's a brilliant master of.
00:15:25.840 --> 00:15:29.840
So it's not only the data guys, but that data also includes how your company works.
00:15:30.159 --> 00:15:35.039
Like Chris, if you and I have the same company, how your company works and how my company works may be different.
00:15:35.279 --> 00:15:35.519
Okay.
00:15:35.679 --> 00:15:40.080
And the second thing is what people don't understand is that people don't give a shit about benchmarks, right?
00:15:40.159 --> 00:15:42.559
There's benchmarks or coding, those are generic benchmarks.
00:15:42.720 --> 00:15:46.240
But how about benchmark for Chris's company, benchmark for Ed's company?
00:15:46.320 --> 00:15:51.759
So if you benchmark for my company, I want to make sure that I own those workflows and I own those benchmarks, and you own your benchmarks.
00:15:51.840 --> 00:15:52.879
You shouldn't have your benchmarks.
00:15:52.960 --> 00:15:54.080
Maybe you operate better than me.
00:15:54.159 --> 00:15:55.120
You should not be giving that to me.
00:15:55.279 --> 00:15:56.960
So that's the kind of thing that Saki is talking about.
00:15:57.039 --> 00:15:59.440
And of course, he's talking his own book, right?
00:15:59.519 --> 00:16:01.440
Because he doesn't have a model, he doesn't have a game in the hunt.
00:16:01.519 --> 00:16:02.720
But the reality of it is it's true.
00:16:02.879 --> 00:16:03.039
Right.
00:16:03.120 --> 00:16:07.039
And now Alex Carp's behind it, he's talking his own book, but the reality of it is it's true, right?
00:16:07.279 --> 00:16:08.159
Yeah, yeah, those two.
00:16:08.559 --> 00:16:08.799
That's it.
00:16:08.960 --> 00:16:09.600
Yeah, no, those two.
00:16:09.759 --> 00:16:10.559
I can go on and on.
00:16:10.639 --> 00:16:12.080
I'm very passionate about this.
00:16:12.240 --> 00:16:15.919
We've invested in six or seven domain-specific models over the last two or three years.
00:16:16.000 --> 00:16:19.759
So this is an area we've been investing behind um, you know, big time, right?
00:16:19.840 --> 00:16:23.440
We've got generalist AI in the robotics space that just closed at 2 billion.
00:16:23.600 --> 00:16:26.480
We've got uh uh Topos Bio in the bio AI space.
00:16:26.639 --> 00:16:34.399
We have uh UR.ai, which allows you to build your own personalized AI, which you should you should try because it could actually read your newsletter and create literally a video of you talking about it.
00:16:34.879 --> 00:16:35.679
I saw that recently.
00:16:36.000 --> 00:16:41.600
Yeah, and we've got we've got we've got one doing uh uh mining for minerals, uh, and I've got a few more coming down the plague.
00:16:41.759 --> 00:16:43.360
So anyway, uh that's super exciting.
00:16:43.519 --> 00:16:49.120
Yeah, no, I mean just you know Alex Carp and obviously Satya and others have been kind of uh beating that drum recently.
00:16:49.200 --> 00:16:54.480
And as you said, like, you
know, of course it's in their own in interest and that they're incentivized to kind of have this narrative, but it's also true.
00:16:54.639 --> 00:17:02.080
Um and like, you know, it it people are kind of having this reverse backlash against the token maxing uh that went on, and now the subsidy era is coming to an end.
00:17:02.240 --> 00:17:20.960
But you know, for me on the security side, this opens up so many, I don't want to necessarily say exciting, but it is exciting security conversations around uh the model hosting, the security of hosting the models, the lineage, the training, the you know, all the uh security aspects that go into to kind of engineering hosting, training, utilizing these things in your own uh environments too, versus you know, using the frontier labs.
00:17:21.039 --> 00:17:29.119
Like when you look at this and it's like uh Satya had that line where like the company's private evals become the most valuable IP, like their their own you know, data and so on.
00:17:29.200 --> 00:17:34.319
Like if evals, context, and memory are the crown jewels, that I looked at that as like a brand new security problem.
00:17:34.480 --> 00:17:37.039
Like who's who's building for that, or how do you look at that landscape?
00:17:37.119 --> 00:17:42.559
Because there's gonna be a lot of exciting opportunities, both for the builders and also obviously for the folks investing in them on your end.
00:17:42.880 --> 00:17:44.319
Yeah, so so I guess a couple things.
00:17:44.400 --> 00:17:53.279
What's really funny is I kind of like say that everything we did at RSA this year was useless because a week after that, uh mythos rumors came out, and then the then the weeks after that, mythos came out.
00:17:53.359 --> 00:17:56.960
When you talk to people that use mythos, they say, holy shit, this stuff is really, really good.
00:17:57.279 --> 00:18:04.160
But it's not so good that you couldn't specifically take an open source harness and just say, let's actually build around it, but it's really fucking good and it's good for people.
00:18:04.240 --> 00:18:04.799
And guess what?
00:18:04.960 --> 00:18:10.000
Most enterprises and most security people don't have the AI people and the talent to actually go build these things.
00:18:10.160 --> 00:18:11.279
So that's that's part one.
00:18:11.599 --> 00:18:18.160
Part two is then the question is, is you know, there used to be a backlog of a thousand vulnerabilities perhaps.
00:18:18.559 --> 00:18:24.160
And maybe you always prioritize the top 10 or the top 20 based on kind of you know mission critical systems.
00:18:24.319 --> 00:18:27.599
But what people didn't realize was that these things can reason.
00:18:27.920 --> 00:18:33.680
I mean that that to me, and that's to take 10 vulnerabilities that might have been on the backlog and you stay staple them together and create an attack path.
00:18:33.920 --> 00:18:38.160
Well, you create an attack path and these things can reason, like I mean, that's crazy.
00:18:38.319 --> 00:18:41.839
And then people say, like, you know, IBM, let's talk about the IBM warning.
00:18:42.000 --> 00:18:46.480
It dropped 25%, and people are saying that it was a software thing, it was you know, blah, blah, blah.
00:18:46.640 --> 00:18:50.400
But in uh a comment was also people are worried about security spend.
00:18:50.559 --> 00:18:54.079
And what they're worried about is like, how much budget am I gonna have to spend?
00:18:54.240 --> 00:19:04.720
Because if you said to people, can I use methos all the time to continuously operate in just code, in just my cloud configurations, my dentist, my endpoints in network state, you'd say yes, I would do that all the time.
00:19:04.799 --> 00:19:05.359
But guess what?
00:19:05.920 --> 00:19:09.279
People's budgets were fucking blowing up, absolutely blowing up.
00:19:09.519 --> 00:19:16.319
And so then the question is, is like, how do I actually do continuous attack path reasoning on a protective side?
00:19:16.480 --> 00:19:16.960
How do I do that?
00:19:17.039 --> 00:19:31.599
So I think there's gonna be a whole bunch of different companies that get spawned to perhaps partner with the Methos or or Daybreak from OpenAI and also give the ability to to do I guess frontier like security at one-tenth the cost or more, right?
00:19:31.759 --> 00:19:32.799
Because you've got to balance it.
00:19:33.200 --> 00:19:43.200
I mean, if I said to you, hey, I can save you one uh you know uh 90% on your security, you're gonna be like, I would only spend that if I get the best security ever, right?
00:19:43.279 --> 00:19:44.799
So that's one area you just can't mess around with.
00:19:44.880 --> 00:19:46.319
So this is gonna be very, very interesting.
00:19:46.400 --> 00:19:50.400
It's gonna change, and I bet you, I bet you, and I look at my existing portfolio.
00:19:50.559 --> 00:20:10.319
I've got a lot of portfolio companies right now already in security there talking about deploying uh in virtual uh virtual environments or on-prem kind of very domain-specific models, could be uh vulnerability models or reasoning models or code models directly on-prem for these um companies because it's easier to do, um, it's easier to build.
00:20:10.400 --> 00:20:17.200
And if you have the talent, you know, why wouldn't the enterprise take this privacy and this data and evals built for them?
00:20:17.359 --> 00:20:18.400
So that's what's gonna happen.
00:20:18.480 --> 00:20:20.079
There's gonna be a ton of companies trying to build this stuff.
00:20:20.240 --> 00:20:22.559
Yeah, no, I mean honestly, I just a few things I want to say on my end.
00:20:22.640 --> 00:20:25.359
Like, this is why I enjoy your content is like you're not in the security space.
00:20:25.440 --> 00:20:27.680
And like I I wrote a book on vulnerability management previously.
00:20:27.759 --> 00:20:29.680
I've been in this, I've been doing that throughout my career.
00:20:29.839 --> 00:20:31.599
It's kind of like your perspective,
man.
00:20:32.000 --> 00:20:44.960
No, every everything you said there, and I've had so I've had people on the show in the past like who are deep in the weeds on this, and like they say we have mythos at home with open weight, smaller you know, models, we can use effective harness engineering uh to find you know zero days and end days and so on.
00:20:45.200 --> 00:20:55.359
And then the vulnerability chaining piece is incredibly valid because historically we always uh patch what are considered critical or high vulnerabilities, like based on severity scores, and we kind of ignore the mediums and lows.
00:20:55.440 --> 00:21:01.039
But if you can chain together several mediums and lows with these models, exploitation looks much more, much more different now.
00:21:01.119 --> 00:21:08.799
Uh so we're seeing like the industrialization of of uh uh of vulnerability discovery and exploitation with AI, and it's gonna really change the paradigm here.
00:21:08.880 --> 00:21:16.559
Uh and then like the whole ecosystem that we use to kind of manage this, like vulnerability databases like NIST, uh, and I'm going way down the geek on the security side.
00:21:16.880 --> 00:21:18.079
CV, but those are all those.
00:21:18.319 --> 00:21:23.599
Yeah, they're all starting to crumble under this volume that's happening of vulnerabilities that are being reported and discovered and so on.
00:21:23.920 --> 00:21:24.799
Can I ask you a question?
00:21:25.039 --> 00:21:37.680
Um like how do you feel about okay, like if I said, here's Fable or Bethos for you at your enterprise and said, okay, by the way, the only way this really works, because the other thing about these companies is that they find vulnerabilities.
00:21:37.759 --> 00:21:42.720
Maybe they can look at the attack path reasoning around it and kind of see kind of how this these paths are created.
00:21:42.799 --> 00:21:51.680
But how much access can you give it to have it understand your domain and your environment, to know every single endpoint, to know every single device, to know every single person, right?
00:21:51.759 --> 00:21:56.480
I I think that's a really troublesome aspect if you feed that to one of the state of the art models, right?
00:21:56.559 --> 00:21:57.680
I mean, I'm just asking you like that.
00:21:57.759 --> 00:21:59.279
Well, no, it's a so it's a perfectly valid question.
00:21:59.359 --> 00:22:04.000
Like on the We we kind of touched on it from the business, intellectual property, competitive dynamic aspect.
00:22:04.079 --> 00:22:11.920
But if you think about it from the security side, you're now turning over all this sensitive data about your environments, your vulnerabilities, your systems, how they interconnect and communicate to a third party.
00:22:12.000 --> 00:22:28.720
And now is a supply chain risk as well, because you're hopefully trusting that Enthropic and OpenAI and others safeguard this data and they don't get exposed in some way because it could be used by a malicious actor down you know down the road uh against you that you've kind of just given the the keys of the kingdom of a map of your uh attack surface to a third party, basically.
00:22:29.039 --> 00:22:40.400
Which is why I think you're gonna go to a point back to earlier the conversation you brought up about Satya is that people are gonna put uh on their own on-prem environment, they're gonna they're gonna have perhaps kind of context memory of their own on-prem environment that they own.
00:22:40.720 --> 00:22:49.279
I mean, right now, as far as I know, Google is probably the one of the only uh Frontier Lab models, if you call them Frontier, but uh models out there that actually create a VPCR for you already, right?
00:22:49.359 --> 00:22:59.279
I mean, that's already been announced, but you know, but you know, and and so you have that, and let's just say you have that model, which is generalized intelligence, but then it's tuned to your thing and and you kind of own it in your cloud.
00:22:59.519 --> 00:23:01.519
But like you have to kind of have that, right?
00:23:01.599 --> 00:23:05.759
Because ultimately, Chris, you and I both know that where this world is going to is a world of outcomes.
00:23:06.000 --> 00:23:06.960
Like they're just cut to the chase.
00:23:07.039 --> 00:23:17.519
We can talk about tech all we want, but it's like the only way you can keep up with it is that agents fix all these problems and you have the trust in the agent to have uh the understanding of your environment, which you kind of have to have, maybe something on-prem.
00:23:17.680 --> 00:23:25.200
Maybe you have access to frontier intelligence on the on the soda side, so it's pinging off both sides, and then agents have to fix these things automatically, um, period.
00:23:25.359 --> 00:23:27.039
And that's kind of where we see the world going, right?
00:23:27.359 --> 00:23:35.599
I have a company called June June, J-O-O-N.co, and it's founded by Alan Cohen, who was the former CTO and co-founder of Samplify.
00:23:35.680 --> 00:23:37.680
So he helped invent SOAR back in the day.
00:23:37.920 --> 00:23:43.119
And uh he calls it just the autonomous workforce is kind of trained here to defend your environment.
00:23:43.200 --> 00:23:51.279
And literally he has Helen, Valerie, Don, Joey, who are these very specific experts that are brought in to augment teams.
00:23:51.359 --> 00:23:53.759
And let's just say you have 10 people on your team, but you need four more.
00:23:53.839 --> 00:23:56.319
Well, we'll sell you four people and they will do the work for you.
00:23:56.400 --> 00:23:58.400
And that's kind of what we're doing, is outcome-based.
00:23:58.480 --> 00:24:00.799
Um, and that's where the world has to go almost like support.
00:24:01.039 --> 00:24:03.680
All right, I'm gonna go on one last little rant, and I'm gonna move us on to another topic.
00:24:03.920 --> 00:24:04.799
No, this is a rant.
00:24:04.880 --> 00:24:05.519
I can go on on a last one.
00:24:05.599 -->
00:24:06.000
No, no, no.
00:24:06.079 --> 00:24:07.200
I love that we this is amazing.
00:24:07.359 --> 00:24:12.880
So, like, you know, last year there was a lot of excitement around what was called serve uh uh services as a software, right?
00:24:12.960 --> 00:24:16.559
The ability to use software agents and so on to kind of conduct labor and do activities.
00:24:16.720 --> 00:24:29.039
And like this is what we need within cybersecurity is to use AI for and agents and so on as a force multiplier for cybersecurity because we've always had uh workforce constraints that you know there's we're always outnumbered by engineers, developers, et cetera.
00:24:29.119 --> 00:24:30.640
We can't keep up with the threat landscape.
00:24:30.720 --> 00:24:35.039
So using this technology as a force multiplier is the right direction to go, and I'm super excited about it.
00:24:35.200 --> 00:24:37.920
But of course, it's dual use, it can be used against you as well.
00:24:38.079 --> 00:24:41.839
So that it's kind of like a dual use technology in that way from a cyber perspective.
00:24:42.000 --> 00:24:44.720
Um I did want to ask you about you brought up Key Card earlier.
00:24:44.799 --> 00:24:51.680
So uh uh agentic uh uh agentic IAM has got tremendous amount of uh uh attention lately, both in investment from the cybersecurity community.
00:24:51.759 --> 00:25:01.359
Uh we're kind of uh starting to introduce new uh protocols like agent auth and thing things to kind of how do we kind of govern and control how agents communicate, authenticate, et cetera?
00:25:01.599 --> 00:25:09.359
Um like with Key Card, you've said that we need to stop thinking about agents as digital twins and start thinking of uh thousands of short lired short-lived problem solvers.
00:25:09.599 --> 00:25:11.119
Like permissions can't last forever.
00:25:11.359 --> 00:25:14.640
So uh is do you how do you think of it from the investment perspective?
00:25:14.799 --> 00:25:25.599
Do you think uh agent identity is a standalone category, or do you think it'll be absorbed by Octand or Microsoft, or like how do you see that one playing out when it comes to you know consolidation and vendors and things like that within the landscape?
00:25:25.920 --> 00:25:27.519
I would say yes, yes, and no, right?
00:25:27.599 --> 00:25:28.559
All across the board, right?
00:25:28.640 --> 00:25:30.960
So so look, uh I funded these guys two years ago.
00:25:31.039 --> 00:25:32.960
Uh, us and Andreessen were the first ones in.
00:25:33.039 --> 00:25:34.799
I had known the founder, Ian Livingstone.
00:25:35.039 --> 00:25:38.079
He was an engineer at one of my first companies that was sold to Salesforce.
00:25:38.160 --> 00:25:47.200
And I backed him uh again after that in a new company in Krieger, and then Jared Hansen, who's the chief architect of Al Zero, said sneaked Al Zero and a bunch of other uh real high-level experience.
00:25:47.359 --> 00:25:53.279
And I'd seen a number of things that were doing NHI level
dashboarding stuff, but I'm I'm a nerd, right?
00:25:53.359 --> 00:25:53.680
You can tell.
00:25:53.759 --> 00:25:56.240
And and I love kind of underlying security primitives.
00:25:56.319 --> 00:26:00.160
And Jared actually built Passport.js and a bunch of other hardcore techs.
00:26:00.319 --> 00:26:07.440
Uh and and what they'd said was like, hey guys, like this is gonna take a little bit longer because the infrastructure is gonna be really hard to do, but we're gonna take a fundamentally different approach.
00:26:07.599 --> 00:26:12.799
Versus kind of showing you all the non-human identities kind of working, let's talk about the problem itself.
00:26:13.039 --> 00:26:32.640
Um, if you believe that every agent should have its own identity, that it shouldn't go through Chris or Ed, that, you know, number one, if you believe they should be dynamic uh uh act access or authorization, uh finely scoped, if you believe that they should be ephemeral and dynamic, right, when it's when a job is done, uh, and you believe that there should be an audit trail.
00:26:32.720 --> 00:26:37.359
And then if you can cryptographically prove that the agent is not hacked in the whole process, that's what we're building.
00:26:37.680 --> 00:26:40.720
And so this company was coming fundamentally from the bottom up.
00:26:40.880 --> 00:26:46.799
And it's like, okay, we want to go to the builders, we want to go to people building kind of agents to kind of help them kind of do this.
00:26:47.039 --> 00:26:51.200
Two is over time, just like Protect, what they realized was that they need to expand their footprint.
00:26:51.440 --> 00:26:54.640
So we made a few acquisitions, we made two or three acquisitions UVC.
00:26:54.799 --> 00:26:59.920
So we actually have stuff, for example, that can plug into clawed code immediately and control blast radius around that.
00:27:00.000 --> 00:27:02.559
You can create rules like never push code into production.
00:27:02.720 --> 00:27:04.640
And you're like, why doesn't anthropic do that?
00:27:04.799 --> 00:27:07.759
Well, because we talked about earlier, we're gonna live in a multi-model world.
00:27:07.920 --> 00:27:15.279
So you write one kind of governance layer that's at the top of Daybright, and people switch all the time from Daybra to from Codex to um to anthropic.
00:27:15.440 --> 00:27:20.160
So we've added a whole bunch of periphery things to just kind of get in the door, even though they may not sound like an ancillary.
00:27:20.400 --> 00:27:22.480
And now we're part of this enterprise architecture stack.
00:27:22.720 --> 00:27:24.000
Uh all right, so let's fast forward.
00:27:24.079 --> 00:27:25.119
That's kind of the primitive.
00:27:25.440 --> 00:27:27.680
So um I'll give you an example.
00:27:27.839 --> 00:27:36.480
We talked to uh a bank and we've been selected from one of the large banks, and they're like, dude, we met with 88 aging companies and you were down to the final two and we picked you.
00:27:36.720 --> 00:27:47.359
Uh and we picked you because you understand the enterprise architecture that's required for us to pick one vendor that can actually manage all this stuff and think about holistically versus these five different point solutions we need to put together.
00:27:47.440 --> 00:27:48.559
And that's kind of what protected.
00:27:48.640 --> 00:27:55.680
We bought four or five companies, and so I believe, in my opinion, that the agent identity market is going to be fucking massive.
00:27:55.920 --> 00:28:02.160
Um, I think that a lot of large vendors will win, but I also think companies like Keycard, we're not the only one, there'll be a few the only ones.
00:28:02.240 --> 00:28:04.400
There'll be a few others that actually think about it holistically.
00:28:04.480 --> 00:28:09.359
Where if I said I want an own agent identity, it can't be just one low-point product, it has to be a number of different things.
00:28:09.440 --> 00:28:12.400
So that's why I said yes, yes, yes, and that means that there's gonna be consolidated.
00:28:12.559 --> 00:28:15.279
We've already consolidated some companies within our own agent space.
00:28:15.440 --> 00:28:23.440
And now we have this enterprise architecture that we're selling, and then we also have a dev motion now where devs can easily just kind of sign up and build up build for it right now.
00:28:23.519 --> 00:28:25.039
So I think it's a massive.
00:28:25.279 --> 00:28:30.880
We need, I mean, this problem has to be solved and it's gonna be solved in many different ways, but we don't need a hundred vendors to do it.
00:28:31.039 --> 00:28:31.519
Yeah, yeah.
00:28:31.599 --> 00:28:32.880
No, I think you're uh spot on.
00:28:32.960 --> 00:28:41.279
Like, of course, there's gonna be uh uh some cycles of like innovative incumbents uh winning, but also new uh a new entrance kind of disrupting and winning in the category.
00:28:41.359 --> 00:28:43.440
And I want to say a few things about uh Key Card.
00:28:43.519 --> 00:28:47.599
I actually follow an individual named Ali Howe who has a uh podcast called The Secure Agents.
00:28:48.000 --> 00:29:01.039
And like people to listen because Key Card is bringing to get together like these industry leaders like Carl McGuinness, uh former chief architect of Okta, uh Dick Hart, who wrote uh OAuth, like all these experts on the domain around agentic identity.
00:29:01.119 --> 00:29:05.839
And I'm like, I'm always following their content from Key Card because I'm learning around agentic IAM.
00:29:05.920 --> 00:29:14.799
Um so like they're doing a, they're they're not only making a product, but they're bringing the the right people together, have the conversation about where the industry is going, how do we build for this, like what do the protocols look like?
00:29:14.880 --> 00:29:19.119
Um so if you're not following insecure agents and key cards content, definitely go uh give that a follow.
00:29:19.200 --> 00:29:19.839
I learn a lot from it.
00:29:20.160 -->
00:29:21.759
Chris, I love I love that you're saying that, right?
00:29:21.839 --> 00:29:24.240
Because remember I said earlier about what we look for in founders, right?
00:29:24.400 --> 00:29:26.160
What is unique insight from what you're building?
00:29:26.400 --> 00:29:28.400
Kind of which which security lens are you doing, right?
00:29:28.480 --> 00:29:30.400
Are you 10xing something or going after a new category?
00:29:30.480 --> 00:29:34.480
And then three is technical mode, then four is how do you think about distribution and marketing, right?
00:29:34.640 --> 00:29:36.480
And so that was a very intentional thing.
00:29:36.559 --> 00:29:43.279
They bought kind of insecure agents because they're like this woman just knows how to kind of build an audience and she has all the key leaders here.
00:29:43.359 --> 00:29:49.759
And by the way, the reason they did that is she's fantastic, she's super smart, but it's also the market is still really early.
00:29:49.839 --> 00:29:55.920
Despite all the stuff that you and I talk about and everyone that listens on it, the amount of agents deployed right now is probably like three or four percent in production.
00:29:56.160 --> 00:30:00.319
So we haven't even we're at the tip of the iceberg right now in terms of the opportunity.
00:30:00.400 --> 00:30:01.440
So it's huge.
00:30:02.160 --> 00:30:02.480
Yeah, yeah.
00:30:02.559 --> 00:30:04.319
Anyway, so I think that's great, right?
00:30:04.480 --> 00:30:16.799
No, I think just that that's a good thing to call out is like there's a technical aspect, there's the market, all these kind of things, but there's also the the brand and the community and like bringing people together, having those conversations, building trust of being a trusted source of uh information in the domain.
00:30:16.880 --> 00:30:17.279
And let's do it.
00:30:18.319 --> 00:30:19.519
It's insecure agents, right?
00:30:20.400 --> 00:30:21.039
That's what's important.
00:30:21.359 --> 00:30:27.119
And so and they learn that from Sneak, by the way, because we we you know when we're at Sneak, we did the secure developer.
00:30:27.200 --> 00:30:27.599
And that's the one.
00:30:27.759 --> 00:30:30.480
I've long followed that podcast, and now I'm following Tesla's new content.
00:30:31.119 --> 00:30:31.279
Exactly.
00:30:31.359 --> 00:30:32.640
So there's this try and drew thing.
00:30:32.799 --> 00:30:36.559
Yeah, by the way, and yes, you have got to go talk to the CISOs and do that thing.
00:30:36.640 --> 00:30:37.519
We all know how that worked.
00:30:37.599 --> 00:30:51.200
But then there's also another mechanism which I think people should understand and I hope some founders kind of understand is that building the brand and building actually a community that's bigger than what you do, that's not about you, that's about kind of, you know, look, ultimately we want to help make the world more secure when we wake up every day.
00:30:51.279 --> 00:30:53.200
If you're in that, if you're in it
for the money, then get the hell out.
00:30:53.279 --> 00:31:00.720
But if you're in it to help make the world more secure and to solve problems to make sure nation states don't take us down, then these are the things that I think provide goodwill.
00:31:00.799 --> 00:31:03.279
And you know, even Ian from Protect AI, he did the same thing.
00:31:03.359 --> 00:31:08.960
He had his uh between like uh his podcast, almost like between two ferns or between two clouds, but he knew all his competitors.
00:31:09.039 --> 00:31:14.079
They would regularly meet across the board just to kind of trade notes with each other because they're like, look, rising tide lifts all boats.
00:31:14.160 --> 00:31:15.440
Let's make sure we kind of know each other.
00:31:15.519 --> 00:31:18.079
I'd rather know who my competitors are than not.
00:31:18.160 --> 00:31:19.839
And that's kind of the the way this industry works.
00:31:20.079 --> 00:31:20.799
Yeah, I agree with that.
00:31:21.039 --> 00:31:24.640
Just uh I'll say last thing for me is like, you know, I think it's a not a zero-sum game.
00:31:24.720 --> 00:31:31.839
Like you can do well financially by also doing good for the community and bringing together folks, having good conversations, helping everyone kind of learn and up level at the same time.
00:31:31.920 --> 00:31:34.960
And that's why I love uh you know the all the outlets that we just talked about.
00:31:35.039 --> 00:31:35.759
Like I learned so much from them.
00:31:36.160 --> 00:31:42.720
Including your podcast, too, because now it's becoming one where my uh founder of some of the PR firms are like, how do I get in the uh Chris?
00:31:42.880 --> 00:31:47.759
So you're gonna have to have my my buddy Yair Grinlinger from Surf AI, you know, hop on the podcast to talk about it.
00:31:48.079 --> 00:31:51.519
It's uh it's funny you mentioned Surf because that takes me to the next question I had for you.
00:31:51.680 --> 00:32:04.319
Is like, you know, you and others have been making the argument that with you know the rise of agencai AI, you know, agents can exponentially outnumber human users, changing the dynamics of of security, like you know, scaling agents while almost no one has the architecture to secure them quite yet.
00:32:04.480 --> 00:32:08.559
Like what does the architecture or I'm sorry, the agentic operating model for security look like?
00:32:08.720 --> 00:32:10.160
Like the surf AI thesis?
00:32:10.240 --> 00:32:11.440
Like uh what does it look like day to day?
00:32:11.519 --> 00:32:12.720
What happens to SOC analysts?
00:32:12.799 --> 00:32:15.680
What is you know, what does the future of this field look like when you have firms like that?
00:32:16.000 --> 00:32:16.799
Yeah, it really is.
00:32:16.960 --> 00:32:20.400
There's probably 85 SOC companies out there, so I think they're all great and everything.
00:32:20.480 --> 00:32:21.680
So we're not trying to do that.
00:32:21.839 --> 00:32:29.279
But what we're what we're trying to do is that, you know, I guess there's a couple of things is you know, Dimitri from Big ID had said something, you can't protect what you don't know.
00:32:29.440 --> 00:32:38.559
And what's very interesting, interesting, what what Surf does is we monitor kind of all your assets, but then we tie all your assets to an individual who's the owner, right?
00:32:38.640 --> 00:32:44.960
And so we call this automated security hygiene because if you don't know who the owner is and there's an issue that pops up, how are you gonna fix it?
00:32:45.119 --> 00:32:45.359
Right.
00:32:45.440 --> 00:32:49.519
And so, and a lot of that is ping pong, as you know, ping pong and the slack and everything else.
00:32:49.680 --> 00:32:58.640
And so what we do is we connect your systems, we kind of automatically understand what's happening, kind of what workflows look like, and then we suggest kind of automations to kind of do things.
00:32:58.799 --> 00:33:07.839
Um, and then most importantly, is we have kind of humans involved in the sense that the agent may ping a human and say, hey, I've already done the investigation for you, hit the button, let's fix this.
00:33:08.000 --> 00:33:10.880
Oh, by the way, eventually, would you like me to create an automation for that?
00:33:11.119 --> 00:33:14.000
And if you think about it, I call it like the 80-20 rule.
00:33:14.079 --> 00:33:25.680
Like a lot of the things that they can automate is kind of like, you know, uh dormant, dormant accounts, just like all these things that take up a lot of time, they're annoying as hell, but there's never going to be company for that.
00:33:25.759 --> 00:33:34.160
There's never gonna be, and so if we can take all those long tail things, let's just say that that long tail of 80% and kind of automate all those things, that's kind of what we're helping CSIS do right now.
00:33:34.319 --> 00:33:40.720
And then, of course, what we do is we drop models on prem, we understand your environment, we do all those things to make it really, really customized for you.
00:33:40.880 --> 00:33:42.559
So that's where it's going.
00:33:42.880 --> 00:33:49.440
You know, what jumped out to me from there from the security perspective is it goes back full circle to the to the point we made earlier about knowing the organizational context, right?
00:33:49.519 --> 00:33:51.680
Like as each organization environment is unique.
00:33:51.759 --> 00:33:56.240
And then like the the 80-20 kind of Pareto principle you talked about, like security technical debts like that.
00:33:56.319 --> 00:34:03.359
Like you see the iceberg tip, but there's so much more underneath that dormant accounts, misconfigurations, old stale, crusty systems that are misconfigured.
00:34:03.440 --> 00:34:08.480
Like there's just a lot of stuff that needs to be addressed that's just sitting there for uh you know, for the taking by attackers.
00:34:08.559 --> 00:34:10.239
So I think you nailed that.
00:34:10.480 --> 00:34:13.679
You nailed that's the most
important thing is this whole context graph that we build, right?
00:34:13.760 --> 00:34:23.679
Because we suck in security, it's not just about security, it's about you know, we suck in data from identity, from the cloud, from workday, from SaaS, from ITSM, and we add it all up together, man.
00:34:23.760 --> 00:34:27.840
That's a lot to map out, continuously understand, and we start providing suggestions for you.
00:34:28.159 --> 00:34:30.559
Yeah, personalized to your organization, which is the key, right?
00:34:30.719 --> 00:34:31.920
Because you know, it's not generic.
00:34:32.400 --> 00:34:32.800
Exactly.
00:34:32.960 --> 00:34:33.119
Yep.
00:34:33.280 --> 00:34:33.440
Yeah.
00:34:33.679 --> 00:34:35.920
So I was gonna, you know, a couple last questions for you here.
00:34:36.079 --> 00:34:40.719
Like, you know, the I think this is uh you have a line that you said like uh enterprise agent hunger games.
00:34:40.800 --> 00:34:42.480
You can't vibe code your way to into production.
00:34:42.639 --> 00:34:46.400
The last mile is security compliance, deter determinism, and more security.
00:34:46.559 --> 00:34:51.760
Um and you we talked about this, like only 80%, 8% of organizations so far have kind of deployed agents.
00:34:51.840 --> 00:34:53.519
Like we're in the infancy phase of this.
00:34:53.679 --> 00:34:55.440
Um, what do you think the real bottleneck is?
00:34:55.519 --> 00:34:57.280
Is it tooling, the trust, the buyers?
00:34:57.360 --> 00:35:04.719
Like, what do you think kind of um is gonna uh help us take, you know, to the other 92% of the untapped uh you know market essentially when it comes to agentic deployment?
00:35:05.039 --> 00:35:06.159
Yeah, I I think a couple of things.
00:35:06.239 --> 00:35:12.800
Uh first of all, I think when you talk to organizations, and I talked to a lot of like, you know, and this is outside of security, you talk to heads of AI, CIOs.
00:35:12.960 --> 00:35:18.800
I just wrote something about uh this I went to this McKinsey CTO forum with a bunch of Midwest uh heads of AI.
00:35:19.119 --> 00:35:19.599
You know what?
00:35:19.679 --> 00:35:21.039
Most people don't even know where to get started.
00:35:21.199 --> 00:35:23.920
Like, like what should I automate first to even get started?
00:35:24.079 --> 00:35:28.239
Let alone kind of, you know, and a lot of times they'll just kind of take a cookie-cutter approach to doing things, right?
00:35:28.320 --> 00:35:29.440
And I'll give you a great example.
00:35:29.599 --> 00:35:31.679
I invest in a company called X5 Labs out of Israel.
00:35:31.760 --> 00:35:33.440
It's x5labs.com.
00:35:33.679 --> 00:35:39.119
And so they drop a little download on your desktop and they understand how you work and they model it all out uh data.
00:35:39.199 --> 00:35:40.960
These are 37 workflows that you have.
00:35:41.119 --> 00:35:44.639
Oh, by the way, let's not just automate the 37 workflows.
00:35:44.719 --> 00:35:46.400
In fact, these are the five that are run the most.
00:35:46.559 --> 00:35:49.119
Let's look at those five and say, should we even reorient these things?
00:35:49.199 --> 00:35:50.559
Can we cut out some steps in the process?
00:35:50.719 --> 00:35:51.679
Can we make it more efficient?
00:35:51.920 --> 00:35:52.960
Okay, now we've done that.
00:35:53.039 --> 00:35:55.679
Let's hit a button and let's actually create the agents for it, right?
00:35:55.760 --> 00:36:02.719
And so, oh, by the way, I'm gonna drop a model on-prem, another thing, model on-prem that is that's that you own that understands your workflow.
00:36:02.800 --> 00:36:04.960
So, so A is I don't know where to get started.
00:36:05.119 --> 00:36:10.960
Uh, two is um, I still think that you know, it's just they're not sophisticated.
00:36:11.039 --> 00:36:12.480
They're just like throwing tons of tokens, right?
00:36:12.559 --> 00:36:13.760
Because most of it is just in coding.
00:36:14.000 --> 00:36:16.320
Coding actually works, it's really good, right?
00:36:16.400 --> 00:36:19.760
And so that's kind of what they haven't actually done the business workflows yet.
00:36:19.840 --> 00:36:25.119
So I think I think we're just still, still very early kind of in that process, and it's confusing as hell.
00:36:25.280 --> 00:36:26.480
Like, where do I get started?
00:36:26.639 --> 00:36:30.800
Like, if you actually look at kind of where we are right now, like what model to start?
00:36:30.880 --> 00:36:32.480
Who would I, what platform do I use, what do I do?
00:36:32.559 --> 00:36:34.880
This, which is why Palantir is almost like the easy button.
00:36:34.960 --> 00:36:41.280
I call anthrop look, there was a period of time, Chris, and and I was kind of one of them where I was like, holy crap, man, anthropic is taking over the world, right?
00:36:41.440 --> 00:36:46.800
After Claude Code, then cowork came out, and then kind of they're just dominating and opening out is nowhere to be found.
00:36:46.880 --> 00:36:52.800
But thank God, because now it just has opened up, it'd be the equivalent of like AWS will kill everyone or Google will kill everyone.
00:36:53.039 --> 00:37:00.880
And in the reality of it is, because of token costs, because of kind of all these issues and privacy issues and governments getting involved, people want alternatives now.
00:37:00.960 --> 00:37:08.000
And I think we're gonna hit this new phase where there's you know platforms where it's easy for you know, it's the easy button, but it's not just anthropic, right?
00:37:08.079 --> 00:37:09.679
It's Palantir is an easy button.
00:37:09.760 --> 00:37:10.719
It's not just Palantir.
00:37:10.800 --> 00:37:12.559
There's gonna be lots of other solutions that come around.
00:37:12.639 --> 00:37:18.320
And even some of these security companies, your opportunities to build your own easy button for very specific workflows that people own, right?
00:37:18.400 --> 00:37:19.920
Like the surf AIs of the world.
00:37:20.079 --> 00:37:21.280
And I think you're gonna see
a lot of that.
00:37:21.360 --> 00:37:23.760
There's probably a lot of that discussion in Blackout as well about these models.
00:37:24.000 --> 00:37:25.039
Yeah, no, I definitely think so.
00:37:25.119 --> 00:37:28.960
And it goes to the point of what we talked about earlier, where you were saying there's so much, you know, where do you start?
00:37:29.039 --> 00:37:30.960
There's so much confusion, so much to need to know.
00:37:31.039 --> 00:37:32.320
There's a lack of expertise.
00:37:32.400 --> 00:37:34.719
Uh, that's another reason, of course, like they're incentivized.
00:37:34.800 --> 00:37:38.880
We talked about the the forward deployed engineering and kind of embedding, uh, taking organizational context.
00:37:39.119 --> 00:37:51.039
But there are also lens of truth to that, that they know the biggest barrier to adoption, implementation, entrenchment, you know, kind of getting entrenched is helping people get started, get going, get things deployed, get things, you know, because people don't know how to do that.
00:37:51.119 --> 00:37:54.639
It's like we're in the we're again, we're in the infancy of this adoption lifecycle.
00:37:54.719 --> 00:37:56.320
Um, so I think that that's part of it as well.
00:37:56.400 --> 00:38:05.840
And it opens up a lot of opportunities, as you said, not just for the biggest players, uh, but for niche companies, companies who can really understand your organizational context, your workflows, your security, your security needs, and so on.
00:38:06.000 --> 00:38:08.239
So yeah, I think it's gonna be it's such an exciting time, really.
00:38:08.880 --> 00:38:10.719
I mean, I mean, I I I think the world's changed.
00:38:10.800 --> 00:38:15.519
Everything that was built, you know, prior, if you're not kind of built for the world of eventual outcomes, right?
00:38:15.599 --> 00:38:23.440
And and actually if you're not built to understand these attack paths, not just kind of siloed kind of fixes, you know, and this is your space, I I think your host.
00:38:23.519 --> 00:38:24.719
I I think everything has to be rebuilt.
00:38:24.800 --> 00:38:29.679
So I think the opportunity for a security fund is to rebuild everything from scratch in a, you know, in in this format.
00:38:29.920 --> 00:38:30.000
Yeah.
00:38:30.079 --> 00:38:31.760
And I'll what I'll say, I'll wrap us up here.
00:38:31.920 --> 00:38:44.960
The thing I will say is like there's gonna be two types of people, in my opinion, I could be wrong, is some people, some organizations, some individuals like us and some leaders and whomever are gonna look at that and feel totally overwhelmed, paralyzed, and like, you know, they don't they're overwhelmed, right?
00:38:45.039 --> 00:38:46.000
They they feel defeated.
00:38:46.079 --> 00:38:48.800
And others are gonna look at that and be like, holy shit, there's so much opportunity.
00:38:48.880 --> 00:38:53.440
It's so exciting, there's so much to learn, there's so much new things that we can do, and like they're gonna be invigorated by that.
00:38:53.519 --> 00:38:55.920
And I think it's gonna be the ladder that's gonna be thriving as well.
00:38:56.079 --> 00:38:57.920
And I think you nailed it.
00:38:58.079 --> 00:39:03.440
I I think it'd be because look, I really think that there's a lot of security professionals out there who are really smart.
00:39:03.599 --> 00:39:07.440
But like thinking about anthropic, there are AI people that create the security stuff.
00:39:07.519 --> 00:39:10.559
Like, I mean, there are rumors that they actually run themselves autonomously in there.
00:39:10.639 --> 00:39:12.960
So like you got to think so fucking crazy out there.
00:39:13.199 --> 00:39:16.000
Like, hey, if I can run security autonomously, I would love to do that.
00:39:16.079 --> 00:39:16.719
That'll be my dream.
00:39:16.880 --> 00:39:18.800
That's why I call it the autonomous enterprises a joke.
00:39:18.880 --> 00:39:20.400
It's like, yes, we're never gonna get there.
00:39:20.480 --> 00:39:21.599
But yes, that'll be the dream.
00:39:21.679 --> 00:39:28.000
So I think you really need hardcore AI people, people that build transform, people that build models and all those things to come into security.
00:39:28.159 --> 00:39:29.599
And you're seeing that right now.
00:39:29.679 --> 00:39:36.960
And I think the best thing and the worst thing is the anthropic and daybreaker coming out there and forcing everyone else to catch up and and and and improve their game.
00:39:37.199 --> 00:39:37.840
Yep, well said.
00:39:37.920 --> 00:39:38.960
So I'll wrap us up there.
00:39:39.119 --> 00:39:40.960
Thanks so much again for jumping on with me, Ed.
00:39:41.039 --> 00:39:44.400
And folks that don't follow his newsletter, uh, definitely do follow him on LinkedIn.
00:39:44.480 --> 00:39:45.599
He's got some amazing content.
00:39:45.679 --> 00:39:49.519
I'm always uh enjoying the insights, both inside security and beyond security.
00:39:49.679 --> 00:39:50.639
Uh so thank you so much.
00:39:50.880 --> 00:39:54.880
I look forward to maybe running into you at Black Hat and if not there soon soon enough somewhere else.
00:39:55.119 --> 00:39:55.440
Absolutely.
00:39:55.519 --> 00:39:56.400
I always enjoyed it, man.
00:39:56.480 --> 00:39:56.960
That was fun.
00:39:57.119 --> 00:39:57.440
Absolutely.
00:39:57.519 --> 00:39:58.079
Take care.
00:39:58.320 --> 00:39:58.880
Cool.