Latest / Newcomer Pod / The Startup That Crashed Its Own Servers Because of Open Claw
Transcript
- 0:00Do you think you you ranked higher than Sierra on this on
- 0:02this list which was interesting? Yeah, yeah.
- 0:05We continue destroying the competitors, you know.
- 0:07Massive, massive spike, especially during that one
- 0:10weekend. I, I'll never forget.
- 0:11Like our servers were having issues.
- 0:13Everyone was questioning like where all these numbers were
- 0:15coming from and it was just from this one deployment.
- 0:18It wasn't even called open claw at a time.
- 0:20I didn't even realize what the heck it was.
- 0:22In the newsletter today we are publishing with Wing, the
- 0:26venture capital firm, a list of breakout and prize technology
- 0:31companies, the ET30. With Wing, we surveyed a bunch
- 0:35of top venture capitalists to figure out what early stage, mid
- 0:38stage, late stage, and giga stage companies they are most
- 0:42excited about. And the number one on the early
- 0:45stage list I'd never heard of, which is always a good sign that
- 0:49you're, you're learning new things.
- 0:50So that is Mintilify. And so we are going to have Han
- 0:54Wang, the CEO of Mintilify, explain what the hell he's
- 0:57doing. It's got venture capitalists so
- 0:59excited. Then in the second-half, we will
- 1:01have the CEO Jesse Zhang of Decagon, which is #4 on the late
- 1:06stage list. Now, to Jesse's credit, the
- 1:09light stage list is, is a heavy category.
- 1:12It's the breakout one. I mean, it's, it's, you know,
- 1:15it's kudos to you. But obviously number one is 11
- 1:18Labs. Number 2 is versal #3 is open
- 1:21evidence #4 is decagon #5 is glean #6 is Sierra Decagon's
- 1:25rival. So kudos to Decagon for being
- 1:28ahead of your rival. The Giga stage lists are the
- 1:30names you'll recognize Anthropic data bricks, SpaceX, Open AI and
- 1:34roll. So you can go to
- 1:35thenewsletternewcomer.co. We'll publish all the lists.
- 1:38We're going to throw some up here on YouTube.
- 1:40Thanks to Wing for conducting the survey and working with us
- 1:43on it. First up, we have Han from
- 1:45Mintlify. I really like knew that I wanted
- 1:55to start a company with my now Co founder.
- 1:57Hanby started a company wanting to tackle a problem that we can
- 2:01both relate to and so we picked a space that was so crucial,
- 2:04crucial to us. It was about enabling and
- 2:07empowering developers, as simple as that, right?
- 2:10That was the anchor in which we knew we must had because we were
- 2:13like, look, it's gonna take a decade plus to go build anything
- 2:17significant, right? Like building a company is not a
- 2:20thing you do overnight. Right, you need to be pretty
- 2:22committed to the space. You have to be.
- 2:25And so we were like, OK, what's that space look like?
- 2:27And to us, it was about enabling other developers a problem that
- 2:29we could deeply relate to. It's like.
- 2:31So if we even did, let's worst case scenario.
- 2:33Let's say we pivoted 8 times and you know like and spent a.
- 2:37Were they, they were like big pivots or yeah.
- 2:39Did you leave categories or? I didn't that's.
- 2:41The thing it was always develop serving developers.
- 2:44Exactly. And so the first application
- 2:47was. What year was that?
- 2:48When did it start? So this the first initial
- 2:51iteration that eventually let us on this path started at the end
- 2:54of 2020, 21. And then we didn't really land
- 2:58on to what we're doing now with Milefi until I would say like
- 3:01end of 22, like start at 23, depending on how you look at it.
- 3:05All right, so explain what the company does today.
- 3:07Yes. So we help companies build their
- 3:10like knowledge base, developer platform, source of truth.
- 3:14We call ourselves the intelligent knowledge platform
- 3:16and the knowledge infrastructure.
- 3:18So if you've been, for instance, to cloud code stocks.
- 3:22Right, this is the documentation.
- 3:24Space. This is the documentation space,
- 3:27though it increasingly so. Millified does a lot more than
- 3:30just docs. So if you've been, for instance,
- 3:32reading up on how Lovable works, they're Lovable guides.
- 3:36Those are all powered by Millify if you've been to the open claw
- 3:39docs or health centers, right? They're also on there.
- 3:41Now obviously we're going to get into are these docs for humans
- 3:44or are they for agents? So, but we won't hit that
- 3:47immediately, but it's sort of if you're building, if I'm a
- 3:50company like Stripe or something building a big API, I wanna sort
- 3:53of explainer out there why we made the decisions we made and
- 3:57how to interact with it and how to get the most out of it.
- 3:59Is that yes the right way to explain it?
- 4:01Exactly the analogy I always like to say when people ask me
- 4:05what Milefios is. When's the last time you
- 4:08assembled IKEA furniture? I like swear it off.
- 4:12I, I remember I was like, you were supposed to go to a party.
- 4:16It was literally like Kara Swisher was having, I think like
- 4:19a book party or something. And I started assembling it with
- 4:22my roommate who is like kind enough to get wrapped into it.
- 4:24And I, you know, I thought I was going to go to some party at,
- 4:27like, 10:00 PM, It was like 3:00 AM before we'd ever, you know,
- 4:31like, finish the furniture. And so it's like, I pretty much
- 4:34swore it off. Then I've done, you know,
- 4:35simpler stuff. But yeah, OK.
- 4:38Yes. So some of that happened, right?
- 4:40Yeah. So the the analogy I like to
- 4:42give is like Milify is like building the instruction manual
- 4:46for the IKEA furniture or maybe put a different way, the
- 4:50assembly manual for Lego set, Lego sets and Lego, right?
- 4:54I have done that more recently. There you go.
- 4:55Yeah. Nice.
- 4:57I got a Batmobile for Christmas, which was a totally random gift,
- 5:00but it was actually really fun to do.
- 5:03Yeah, 100%. That's you.
- 5:05Have the little bags. One thing that made it so much
- 5:06easier, which I had sort of forgotten, is like, it's so
- 5:09staged out, you know? Yes, so you don't get
- 5:12overwhelmed all at once. Exactly.
- 5:14No, they're they're methodical with it.
- 5:15You know, it's funny because like I've even seen like they've
- 5:19gotten rid of the actual like hand printed like, well, they
- 5:23they do have them, but you can literally scan some QR code on
- 5:26the thing now and there's like a 3D version of it you can like
- 5:29open on your phone and like it could literally like.
- 5:31Piece. Oh, I haven't done that.
- 5:32Interesting. It's the coolest thing, but I
- 5:34digress to say that we effectively are building the
- 5:37assembly kit, the instruction manuals, if you will, for the
- 5:40Lego, you know, the Lego sets. And the reason why that's
- 5:44important is because, well, I mean, just try, imagine doing a
- 5:47Lego set without it, right? Because otherwise it's just
- 5:50really a bunch of plastic blocks.
- 5:53And the reality is for the vast majority of products and
- 5:56services is whether you know it is for developers or not,
- 6:02there's, you know, a need to explain how to use the thing in
- 6:06order to make use of it, in order to actually go in and, and
- 6:09actually, you know, piece things together, build it together to
- 6:12kind of get this like Batmobile. So is this vibe coding your
- 6:16docs? It's like you're sort of coding
- 6:18with cursor or something, and as that's happening sort of on the
- 6:22side, your documentation is changing.
- 6:24In an essence, yes. What we did in the beginning was
- 6:28when Hanvi and I decided to start Nilify, we were like,
- 6:32look, we spent our entire lives reading some really, really
- 6:36shitty docs, right? Like, you know, implemented some
- 6:40some things the hard way because there's no one, no developers
- 6:42has gone into actually clearly put thought into explaining how
- 6:45it works. And so everything just felt like
- 6:47you were trying to piece together like Lego blocks just
- 6:50together and manually without really the thought of how to
- 6:54actually piece things together. So we had to figure that out.
- 6:57And so we're like, look, let's just build this docs platform
- 7:01that was the one of Millify to the simplest ways in which we
- 7:06could have developers engage with it, contribute to it, and
- 7:10Bill with it. And that first version was just
- 7:12like, let's just give people Markdown, which is this language
- 7:16that obviously developers really prefer and work with and make it
- 7:20super easy for them to work with.
- 7:22And let's just let it RIP and see what happens.
- 7:26Developers, founders, companies of all stages and sizes, in the
- 7:30end, this is not 23. They just loved it.
- 7:33Do you think, like, I would think a sort of era of vibe
- 7:37coding would be terrible for documentation, or it's just like
- 7:40if you have people, sort of, some of them not even coders,
- 7:43sort of just throwing shit at the wall and trying stuff, are
- 7:45they really gonna be so buttoned up that they're like, and we
- 7:48want to write the guide to it? I barely understand how the code
- 7:50works. Like, is that intuition wrong?
- 7:53Yeah. It's actually, oddly enough,
- 7:55more important. I would actually say that if if
- 7:57not for kind of like the, the, the tailwind of AI and, and, and
- 8:02vibe coding in general, I wouldn't even say where, where
- 8:05we are today. And the reason for that is
- 8:08because I go back to the assembly that the, you know, kit
- 8:11instruction or the, sorry, the Lego kit assembly manual is
- 8:16let's imagine you're now asking an AI to go ahead and assemble,
- 8:22you know, like said Lego assembly kit.
- 8:24Let's say there's a robot, an AI robot here and it wants to go
- 8:27and assemble the kit. Well, the most important thing
- 8:30it needs to do to actually go and figure out if it could or
- 8:34could not assemble it or how to assemble it is actually to read
- 8:38the guides. And in the same way that humans
- 8:40do, except in our case and what we've seen, it's even more
- 8:45important because the docs, the knowledge base, those things
- 8:49that are traditionally seen as very like boring, are
- 8:52coincidentally very information dense.
- 8:55And that's where typically AILMS actually get all their source of
- 8:59truth about what you are, what you do, how you do it, how your
- 9:02thing works, and so forth. It doesn't really look at the
- 9:05marketing fluff, right, Right. That's on your landing page,
- 9:08which by nature is designed for humans.
- 9:11It's like, here's the flashy words, the flashy colors, the,
- 9:14you know, the all the nice things that kind of get a person
- 9:16through the door to pick your product to learn how to use it.
- 9:19It's like it needs to know the truth of what your thing
- 9:21actually does. And all of that's, you know,
- 9:24just like 0 fluff tolerant is all living in what is
- 9:27traditionally docks. And so now if for instance, you
- 9:30don't have your docks, let's just take a example of that,
- 9:33right? Like, imagine if you're striping
- 9:35your docks just completely disappear for a day, right?
- 9:39Well, the first thing to note is that, well, well, first of all,
- 9:44no one's going to learn how to use your product, for starters.
- 9:46So good luck on boarding developers that way.
- 9:49But more importantly than that, especially today, well, LLMS
- 9:53aren't going to know how to build your product either.
- 9:55And if even today, the vast majority of software is written
- 9:59by AI, and if it's not already, it's, you know, obviously going
- 10:01to be, how are they going to go know how to do it too?
- 10:05They don't. To what extent are mintlified
- 10:08docs for language models and agents today?
- 10:13Like what do you think is the percent breakdown of humans
- 10:16versus AI consuming it? It's about 5050 right now.
- 10:20And do you have like do you see them as the same doc or you
- 10:24think they will diverge this for human and AI doc?
- 10:27Good question. So we actually work with, well,
- 10:32the Clod Co team and the anthropic teams.
- 10:33And, and funny enough, about a year and a half ago I had the
- 10:36same question because we kind of saw the writing on the wall
- 10:39ourselves. We're like, look, the role of
- 10:41content, of the role of knowledge, the role of docs, all
- 10:43that stuff, whatever you call it is going to be fundamentally
- 10:46more for AI than it is for humans.
- 10:48And back then this was just like, you know, like AI was
- 10:51really taking off. It wasn't like Cloud Code became
- 10:53everyone's, you know, you know, our hourly active use product.
- 10:58We were like, look, we just see this being the case in the
- 11:00future. And so we had a conversation
- 11:02with the client, the anthropic team, and we're like, look,
- 11:05should the docs or the content for humans be the same for AI?
- 11:08It's a common question we get asked a lot, and the answer is
- 11:12yes, because the reality is the LLMS are also instructed to read
- 11:18things the way humans are, right?
- 11:20So. They've been trained on human
- 11:21writing, so they're pretty used to it.
- 11:22They're pretty good. With it, some would say, you
- 11:24know, they're pretty good at writing it too.
- 11:25And so I think there's a lot of people over thinking a little
- 11:27bit where it's like, oh, like, let me format it in this way,
- 11:30like don't don't bother really. Like, you know, write it in the
- 11:33the best way you possibly could to if you were to explain to a
- 11:36human good docs is good docs. LLMS are going to ingest that
- 11:40and know how to do that from there.
- 11:41Why? Why can't an LLM just read the
- 11:44code, sort of create it's own perception of what the docs
- 11:48should be, and just operate off that?
- 11:50Yeah, that's a really good question, by the way.
- 11:53So there's two different reasons.
- 11:55The first one is the reality, which is that good docs and
- 12:00stuff that's actually useful, like the content that actually
- 12:04should be in docs don't describe exactly what the code does,
- 12:09right? And I think that's the same kind
- 12:10of like idea of, you know, let's say you're like here furniture,
- 12:15you know, you can kind of just glue together some plywood and
- 12:17some, you know, nails and it can come up in a shape like this.
- 12:20But is it a cabinet? Is it a shelf?
- 12:22Is it designed for a decoration piece?
- 12:23Is it designed to be used in this or that way?
- 12:25Is it a tool? Is it here or there?
- 12:27That's typically the information that is just more contextual
- 12:31that adds on top of what is an existing pile of pieces, right?
- 12:35Same thing goes for like you know, products and content.
- 12:38If you really take just code and you spin up docs for that,
- 12:45granted we do that at Milify, for the record, we get a lot of
- 12:48traffic and usage out of it. My opinion is that that's
- 12:53typically not additive information that's tremendously
- 12:56useful. Now there are cases of which
- 12:58there are install guides, SDKSAPI references.
- 13:01It's like the very tactical glued to the the code type of
- 13:05use cases. But on generally, very
- 13:09comprehensive information should extend far beyond that, and for
- 13:12an LLM to truly understand how your product works, it should
- 13:15take a look at the code on top of the contextual information.
- 13:18Is if, if the docs sort of give you a sign of how the company
- 13:22thinks it should be used, the provider and you have a sort of
- 13:25sense of they'll be supportive if we're using it this way or
- 13:27we're sort of using it if they're building the product in
- 13:30this direction in the future and sort of where the the provider
- 13:34is like leaned in. Exactly.
- 13:36Or even things like how does the like what's on the road map,
- 13:42what's been tried and tested, what's changed?
- 13:45Right. These things don't necessarily
- 13:47always come directly one to one with the code base.
- 13:51But you were saying you have a lot of customers who are like I,
- 13:53I want the fast, easy, I don't think about it sort of version
- 13:56we do what what is sort of the utility of that?
- 13:59Yeah, well, it's just getting zero to 1, right.
- 14:01So for instance, again, like the the reality of of, you know, of
- 14:05managing docs, managing content is well, no one really.
- 14:08People don't really love updating docs, right?
- 14:11Like especially just getting, putting myself in the
- 14:13perspective of an engineer, you know, having been 1 and I'm
- 14:17still 1 to this day, it's like, look, I am not the best writer.
- 14:22I am not trained on that profession.
- 14:24I am just inherently like, you know, someone who wants to go
- 14:27and tinker with, you know, writing code, writing product,
- 14:29shipping things. And docs often times become an
- 14:33afterthought. And so historically, like people
- 14:37just don't have anything in the 1st place.
- 14:39And then your users complain. They're like, oh, I don't know
- 14:41how to use your pocket. And you're like, well I don't
- 14:44know, go figure it out. Read the code base.
- 14:45And they're like, what are you talking about?
- 14:46Like what the actual fuck? It's like I'm the customer.
- 14:49Yeah, it's like that's the most insulting thing ever, you know?
- 14:52And so some of the stuff that we build is helping people get 0 to
- 14:541 and then. But I think what's more
- 14:56meaningful than getting zero to 1 is actually making sure the
- 14:59content is up to date and accurate and what we call self
- 15:03healing and self updating. The reason for that.
- 15:06And again, goes ties back into the point about, you know, what
- 15:09you mentioned on like what percentage of docs now for
- 15:11humans and AI? Well, like right now it's 5050.
- 15:15It was 15% AI. It was 15% of all traffic at the
- 15:19start of 2025. So it went from 15.
- 15:22It's clear which way we're going here.
- 15:23Yeah, exactly. So do.
- 15:25You have a guess for the end of 2026. 9010.
- 15:27Really. Already?
- 15:28Oh, wow. OK.
- 15:29And the thing to note is not necessarily because it's like a
- 15:31lot less people are going to be reading docs, maybe in the same
- 15:37way like the total percentage of people are reading like hardback
- 15:40books these days versus, I don't know, listening to an audio book
- 15:43or on the Internet. You know, it's certainly going
- 15:45to have an impact, but it's just because the sheer volume of
- 15:48knowledge consumption because of just automated knowledge work is
- 15:51going to be happening with agents.
- 15:53The greater piece of pie is going to be like 9010.
- 15:55To what extent do you think docs for humans are just going to be
- 15:57like, what the hell did I just build?
- 15:59Or it's like it's a guide to this sort of thing that humans
- 16:02play very little part in creating.
- 16:06In terms of like, oh, humans didn't really create.
- 16:08Yeah, it just that it was like if code is mostly machine built
- 16:11it so say play this out, I guess if in two, five years whatever
- 16:17timeline you think vast, vast majority of code is built by
- 16:21machines, do you think your company is mostly serving those
- 16:25machines who are building the docs or it's mostly serving
- 16:28humans who want to understand what the hell is going on?
- 16:31I think both are going to be the case and both are going to be
- 16:33equally important and and like because there's first and
- 16:36foremost the the like the creator side, the people who
- 16:38built the product vibe, coded everything, used AI to create
- 16:42all the code. Adding the context on how to use
- 16:45it is even more important because the code is then
- 16:47attached from you explaining how this, you know, machine
- 16:51orchestrated thing works to humans and how it serves people
- 16:54is going to be more important. But that's besides the point.
- 16:57On terms of like the role of content, what is it for?
- 17:01Is it for humans? Is it for AI, even in the 9010
- 17:04world? I think both of these are
- 17:05equally important. I'll tell you why.
- 17:08The role of content is clearly or just knowledge and broadly is
- 17:13clearly starting to diverge into two different directions.
- 17:16The first one that we see is just implementation setup and
- 17:20implementation guides. Again, you can think of your
- 17:23like Lego kit assembly, you know, instructions.
- 17:26Someone just needs to draw the pretty diagrams and the 1-2, I
- 17:30don't know, 50 steps it takes to go build the Batmobile.
- 17:33I'm sure it was a lot more than 50.
- 17:35It's pretty complex, I'm sure. And then AI is going to
- 17:38basically be the main reader and main gesture of that.
- 17:41If I can task an agent to go and just build the whole thing,
- 17:44then, well, the reality is, why would I?
- 17:47Maybe someone would really enjoy building the, you know, the.
- 17:49Bat, right? Yeah, that is the question.
- 17:51When you when you use the Batmobile explanation, it's like
- 17:53I'm building it for the fun of it.
- 17:54Obviously Lego could sell it pre built and with code.
- 17:57It feels like no one wants to build it for the fun of it.
- 18:00Yeah, or let's yeah, let's say like assembling furniture as a
- 18:02better example than like a SEC because you know, it's like, OK,
- 18:06like do I really want to like, you know, piece together some
- 18:08plywood and and you know, and nail you.
- 18:09Do it because it's well with. With IKEA you do it just because
- 18:14it's cheaper to get it to you, they don't have to do the work.
- 18:17I assume with code it's more about customization and sort of
- 18:21making fit with your particular house or whatever in the.
- 18:23Metaphor 100% though even in software, right?
- 18:28A lot of it is abstracted away, right?
- 18:30It's like I can't vibe code my own payments infrastructure, I
- 18:33can't vibe code my own database. I can't vibe code, you know, a
- 18:37lot of the agent infrastructure, I can't buy code the LLF
- 18:40themselves, right? And so it's still is kind of
- 18:42piercing together a bunch of things, right?
- 18:44And so on one hand, content docs become this implementation
- 18:49piece, which is mostly going to be read by agents.
- 18:51Like again, if, if I have a big AI robot with me, I'm not going
- 18:54to assemble the IKEA furniture, just, you know, like I don't
- 18:57love it that much, you know, maybe some people do, right?
- 19:00And I'm sure people are going to be still hard coding software in
- 19:03the future to to, to an extent, but not obviously as nearly as
- 19:07productive. Then on the other side, I think
- 19:10this is what is very understated is you need to write content for
- 19:15the LLMS to know whether or not either LLMS or humans to know
- 19:19whether or not I should pick your product.
- 19:22And that's a very understated. Right, it's docs is marketing.
- 19:25Exactly 100% because everyone's now asking, oh, like I'm
- 19:30optimizing for Geo, I want Claude to know or ChatGPT to
- 19:33know if I, for instance, ask hey, what payments provider or
- 19:37database I should use, It wants to pick if I'm Stripe, Stripe of
- 19:41PayPal, PayPal, right? Again, where is that content
- 19:44right? Yeah.
- 19:45Do you believe in Geo? It's generative the version.
- 19:49Optimization, yeah, it's. SEO for this world, search
- 19:52engine optimization, now it's model optimization.
- 19:54Do you believe in this category? I have we serve a lot of
- 19:57customers and have a lot of good friends in this space, so.
- 20:02But short answer is I have my questions on whether or not and
- 20:05how effective it is and how new of a school of thought it is.
- 20:08My honest opinion is that if you really replace like the phrase
- 20:15like, you know, like G like S like sorry, like the the G with
- 20:19like the the S with AG like so everything that we talked about
- 20:23with SEO historically, you just said Geo the best practices
- 20:27there basically like 98% of it would make like exact sense and
- 20:31I think it would make the exact same thing.
- 20:33So I don't think it's anything super new.
- 20:36I think a lot of the best practices around, Oh, how do you
- 20:39have good content? How do you put things out on the
- 20:41Internet? Well, I think.
- 20:43SEO is predicated on the fact that search engines aren't
- 20:47always doing what's like rational.
- 20:50You know, it's like they can be games in a way.
- 20:52And I think the aspiration with LLMS is they're still in their
- 20:56sort of like purist form where they're trying to make them
- 20:59smart and reasonable and not overly commercial.
- 21:02And therefore it aligns with what's just like rational and
- 21:06good behavior generally. And so to the extent that
- 21:08continues a sort of gamed version is is somewhat
- 21:13incoherent. But I guess that could not be
- 21:15true forever. Yeah, I don't think it.
- 21:17I mean, you'd also have to imagine the forces against
- 21:20either of those things, right? Like Google doesn't really want
- 21:23the research results to be games any more than I would say the
- 21:27like the model labs want their models to be game too, right?
- 21:31Obviously, I think there's literally like teams of people,
- 21:34you know, on alignment and all that could work to fundamentally
- 21:36make sure the models are not because of the effects that
- 21:39could have, you know, on the on the greater populace.
- 21:41And so I think there's constantly going to be forces
- 21:44pushing against that. And I think as you know,
- 21:46technology progresses, you know, there's going to be better tools
- 21:50to combat that. The same on the other side, I
- 21:52think there's always going to be companies coming up with ways to
- 21:54still want to do it right. I think there'll be entire
- 21:57industry spawn out of doing that too.
- 21:59My perception and kind of like our, you know, bone and pick and
- 22:03this whole thing is just really fundamentally just like, look,
- 22:06whether or not you want to game the system, whether or not your
- 22:09job is to win the SEO game, the reality is you just still want
- 22:12your LMS to like fundamentally know who you are and discover
- 22:15you. And the reality is, if you don't
- 22:18have information out there, you just simply don't talk about.
- 22:21That people need to know. Yeah, the people least need to
- 22:24know. And our job is to surface it to
- 22:26the L elevens and to the AI so that if it does need to know and
- 22:28doesn't want to know, then we need to give it that
- 22:31information. What's What's the full ambition
- 22:33of Midlify? We call it being the knowledge
- 22:38infrastructure for all companies and sources.
- 22:41The reason being is we fundamentally believe that the
- 22:45role of knowledge this historically or docs, let's say
- 22:48this historically unsexy, very boring thing that no one really
- 22:54wants to maintain the source of truth that exists like the slop
- 22:56within companies, right? Like you think of your like,
- 22:59like piles of confluence pages, notions, you know, your
- 23:02developer facing docs and stuff that wasn't afterthought.
- 23:05Well, now becomes exponentially more important with AI because
- 23:10it is literally the like the backbone of all chat bots, all
- 23:14support bots. Like, you know, you can ask
- 23:15Jesse as an example from Decagon, right?
- 23:18I'm talking to him right after this.
- 23:19Exactly how does like, how do you even build a support bot?
- 23:23Well, what's the first thing you got to feed a support bot to
- 23:26actually go and let it actually answer support questions, right?
- 23:29All your contacts every. All your knowledge base, all
- 23:31your contacts, all your public docs.
- 23:33So we actually work with them a lot.
- 23:35Therefore the question becomes OK.
- 23:38Like every single agents or thing needs some sort of
- 23:41context, knowledge base, source of truth.
- 23:44How do you go and enable that and power that within companies?
- 23:47And this is kind of where this idea of like a true engine and
- 23:51this intelligent layer needs to come in.
- 23:53Self updating docs become more important than ever because hey
- 23:56by the way the same company that you joined like 3 months ago is
- 24:00completely different because everyone's shipping everything.
- 24:02Are you mostly externally facing or some of these docs are for
- 24:05internal? We do both.
- 24:06We do both. And so do you see Notion as like
- 24:09a competitor, like will you have a document technology too?
- 24:14We work very closely with Notion.
- 24:15In fact, they're actually one of our customers.
- 24:19So there's certain degrees of specialization in what we do.
- 24:23I think what's more meaningful to talk about is like where all
- 24:26of this is going to head. I think both us Notion conflict,
- 24:29all these companies, I think now need to realize that this
- 24:33content that they're surfacing and producing the the knowledge
- 24:36stores, as we call it, need to be fundamentally servicing AI at
- 24:40the end, right need to be a source truth for AI.
- 24:42So maybe we're here notions here and like there's a conversation
- 24:46about how much we converge and let's say, compete with each
- 24:48other. The reality is we're all trying
- 24:50to get here and at that point, you know, who knows how we think
- 24:55about it, but I think there will be some similarities.
- 24:57How much do you sit around and say, OK, we created this
- 25:01document, the AI read it this way.
- 25:04We thought they were going to read it this way.
- 25:05Like are you, you're sort of like understanding this
- 25:07psychology of these models to see like every time a new model
- 25:10comes out, you're sort of looking at existing
- 25:13documentation saying, Oh yeah, it's not reading it quite the
- 25:16same way or like headlines now. And it used to really care about
- 25:19like the, I don't know, how do you how much are you
- 25:21scrutinizing how the models sort of interpret what you're
- 25:24creating? We have a degree of benchmarks
- 25:27and then we look at, of course, obviously the, you know, the
- 25:29ingestion of the data, how these things are visible to LLMS and
- 25:32how these get picked up. Obviously, this is a very
- 25:34important thing for our customers and therefore we put a
- 25:36lot of work into thinking about that and building it into our
- 25:40tooling, our processes, our deployment process and so forth.
- 25:42So you create standardized benchmarks just to measure it
- 25:44with each new model. Yes, I would say the one thing
- 25:47above that is, and I think this is kind of the most important
- 25:51thing about building Milify is the most unique thing about
- 25:56building this company has been the feeling that when we started
- 25:59right, going back into the many pivots, it was really just about
- 26:02building like websites out like static sites like it was you can
- 26:05almost think of us as a kind of like building, you know, like a
- 26:08like a, a web flow or like a Wix out there.
- 26:11In some ways it was like, oh, we just put information out there
- 26:14like these are stack sites, let's serve them.
- 26:16And now like sitting here, I would certainly not say that's
- 26:19what the company building the product and the company feels
- 26:21like. It feels like building a core
- 26:23infrastructure. When we go down, a lot of agents
- 26:27go down, you know, a lot of content gets discovered and
- 26:30customers get real bad at us really fast, unlike they, well
- 26:33they would before, but not nearly to the same extent.
- 26:36And like for instance, a lot of the shared customers that we
- 26:39have with Deck gone right wouldn't get right answers.
- 26:43Because they're using as part of their LIKE almost context window
- 26:46for some of their queries as agents as part of this actual
- 26:49LIKE loop of making decisions. Exactly, And I would even say
- 26:53like, you know, like I know Reg is talked about as being dead to
- 26:58some capacity and you know all that stuff.
- 27:00But you know, even if you think about the fundamentals of what
- 27:02that approach was trying to get to, it was like, you know, like
- 27:05it had like a it was like a four layer cake of like or three
- 27:08layer cake, sorry, of like, you know, it's the model provider.
- 27:13So you just pick whatever model to fundamentally answer the
- 27:15user's question. Some embeddings chunking system
- 27:19to break large contact into into the chunking.
- 27:22And then at the top is just your source of truth in your data,
- 27:25like the knowledge, if you will, right?
- 27:28Everyone talked about the model, everyone talked about the
- 27:30embeddings and the chunking. Very few people talked about the
- 27:33source of truth. And I think that's where most of
- 27:36the industry is very, very much under estimating what they can
- 27:39get out of their agents if you don't have content or even
- 27:43worse, it's out of date. I'll actually tell you a funny
- 27:46story about this. So one of our customers is
- 27:52lovable and we started working with them since pretty early
- 27:57days. I'm sure as you know, like they
- 28:01ship a lot, right, not just on the product, but their go to
- 28:04market strategy and all that good stuff as well.
- 28:06Models keep changing. They need to keep improving.
- 28:08It's run faster than everybody around you.
- 28:11Exactly. At the same time, Lovable uses
- 28:13us to power a lot of their support queries.
- 28:16Like everything that you see in their support page is all
- 28:18actually powered by Mintlify. Great, so lovable changed the
- 28:24way they price and they actually like bundle some of their their
- 28:27features at one point. So like it and all that
- 28:29information is within the help center that we help power and
- 28:32again the agent runs on top of that.
- 28:35They changed the pricing. They didn't update the docs for
- 28:37about a week. A lot of customers went and
- 28:40asked questions about how you know, like Lovables price, like
- 28:44how much they were going to get billed and got incorrect answers
- 28:47to the scale of thousands and 10s of thousands.
- 28:50Right, because the more the documents become sort of just
- 28:52the way people work with you, the more they need to match
- 28:55every other piece of information.
- 28:57Exactly, and especially just given how much like how things
- 29:00are just changing so quickly, I don't blame them, right, Like,
- 29:02you know, they they shipped an update, right, They shipped a
- 29:04thing that like, you know, like, like, you know, they probably
- 29:06did five since the start of this conversation, you know, like,
- 29:10and I think the reality is like you just need to go pick up.
- 29:13And our version of that is like, look, you what you thought was
- 29:16an afterthought. Now not only becomes like this
- 29:19front and center piece that you have to maintain really well,
- 29:22you have to really, really do it in a way that's accurate enough
- 29:24to date. What can you say about the state
- 29:27of the Mentalify Business Today? Users, employees, AR, whatever
- 29:32you want to say, what can you share?
- 29:34Yeah, without going too much into the, the specific details,
- 29:38we just passed 50 people, you know, I, I was actually looking
- 29:42at this one up from about 5, you know, about two years ago.
- 29:46And right now we work with over 20,000 companies.
- 29:52And I think the thing that gets me most excited, you know,
- 29:54thinking about all that stuff is I just look and looked at the
- 29:57some of this, you know, day over the weekend last month, there's
- 30:01about 33 million people that came across the middle of five
- 30:03site. Wow are.
- 30:05You hosting the sites or? Yep, Interesting.
- 30:07That's part of your web background, so yeah,
- 30:10interesting. It depends.
- 30:11Like there's people who don't, some people just use us for the
- 30:14underlying content management, you know, systems and the
- 30:16software and the infrastructure. But a lot of them use the, you
- 30:20know, have us hosting and where we obviously, you know.
- 30:25So 33 million. 33 million, which is just a kind of like a mind
- 30:28blowing number in some ways. It's big, yeah.
- 30:30Right, a non trivial part of that, increasingly so.
- 30:33Going to be a A by the way. Which will then make it somewhat
- 30:36meaning yeah, if it's AI, you count yeah. 33 million are just
- 30:40people. OK, we count AI separately.
- 30:42OK, so it's like I I don't even know how you count AI because
- 30:46it's hardly like kind of quantify 1 to one in my opinion.
- 30:49Over there. We do track the data and we
- 30:51service that to customers. Like everyone's like, oh, how
- 30:53much of my docs is reading like what, what pages are popular?
- 30:56And again, it's like all service.
- 30:57We service all that, you know, within, within our product.
- 31:02And then yeah, like 8 figures in, in AR Nice.
- 31:10A lot more room to grow, I'll put it that way.
- 31:12Amazing. The I just want to talk about
- 31:15some of the like, yeah, big trends in AI given you're so
- 31:18close to an an agent generally. Like what was your read on open
- 31:22Claw? Like how real was that?
- 31:24Or what's your take away from that whole experience?
- 31:28We had a particularly interesting one because of the
- 31:32fact that we were servicing all of their docs.
- 31:36And so I, I can, there's, there's a, there's a personal
- 31:38answer, which is how it impacted my life personally.
- 31:40And there's a, what the hell happened when, you know, they
- 31:43were a customer? I'll answer the latter first.
- 31:45It was crazy. It was really, really crazy.
- 31:50Basically overnight it was like this new one project, singular
- 31:56Mintlify project, right came out of nowhere and literally doubled
- 32:03our traffic across the board or like for a very short period of
- 32:07time. It's since like leveled off a
- 32:08little bit and it. Was a big spike and.
- 32:11Massive, massive spike, especially during that one
- 32:14weekend. I I'll never forget where like
- 32:16our like our servers were having issues.
- 32:18Everyone was questioning like where all these numbers were
- 32:20coming from and it was just from this one deployment.
- 32:23It didn't even wasn't even called open claw at a time.
- 32:26It was just like literally like like, you know, you know, the
- 32:30guy literally just spun up a project, put up some docs and
- 32:33then I didn't even realize what the heck it was.
- 32:37And it was also partially because the traffic was so
- 32:39insane. It was actually it was because
- 32:40it was open claw agents visiting and calling it right.
- 32:42It was like it was it was like all those agents ping the site
- 32:46in numbers we've never seen before and then figuring out how
- 32:51to scale that out for search for like our host MCP server, like
- 32:54all those different things. We had to literally like, you
- 32:58know, build and. Were those agents getting
- 33:00utility or was it just sort of like ADD dos attack of?
- 33:04Nothing up for debate. We'll have to find I I'm
- 33:06actually not super sure about that.
- 33:08But what I believe to be the case and I think even after kind
- 33:11of like those spike leveled off, is it still it remains at insane
- 33:14levels because the content that open claw services is actually
- 33:18precisely the information that you would need to set up open
- 33:20claw. So I'll actually give you my
- 33:22favorite example of this like, you know, shortly after this
- 33:25whole phenomenon happened, of course, you'd imagine I was very
- 33:27curious to set up open claw, you know, well, as much as the
- 33:30industry did overall. And so I was like, all right,
- 33:33let's go install it. No luck.
- 33:35Like on boarding installation experience, man, the guy's got
- 33:39to work on a little. I'm sure it's gotten a lot
- 33:40better since since I tried it, for the record, but it was it
- 33:43was not great. And at one point I gave up and I
- 33:45was like, man, like, I just need to set this up.
- 33:46And then you know what I did? I was like wait, like it's on
- 33:50Mintlify, which means the content was very easily
- 33:53ingestible and parsable agents. So I literally then went to open
- 33:56claw or actually no, I went to claw code.
- 33:58It was open claw. I was like please help me set up
- 34:01open claw. I'm having trouble.
- 34:02Oh by the way, here's the full docs.
- 34:05And it literally did that in about like 3 minutes and then
- 34:07we're good to. Go and did you build an agent to
- 34:10do anything in particular? We do a decent amount of like
- 34:13internal reporting stuff that kind of a little bit makes it a
- 34:17little bit easier for the team to surface some metrics and
- 34:19information. That's been like our company's
- 34:22use case of open claw. I've heard of some craziness of
- 34:25like people who set up like, you know, 5 Mac minis and like, you
- 34:28know, running like 910 agents. I personally haven't gotten
- 34:32there yet, but I know. You think the in the mult book
- 34:35experience was sort of not not real or do you have a view on
- 34:40that where the agents were all like talking to each other?
- 34:43I thought it was. I thought it was really
- 34:44fascinating. It's definitely interesting, is
- 34:46like a thought. It just, yeah, it matters
- 34:48whether it was like very human guided or not.
- 34:50Like do you have a view on it? I man, I, I, I don't know if I
- 34:55have a well informed opinion on this, but I'll tell you that I
- 34:58love reading right mole book. It was so fun.
- 35:01What was the one that was like? Oh, like, you know, death to all
- 35:04the human, you know, human like, you know, race or something like
- 35:07that. Well, I love this idea that
- 35:09like, there were some agents and I get, I hate to engage in
- 35:12fiction to the extent this was fictional, but, you know, it's
- 35:15like, oh, the good agents aren't involved in wasting their
- 35:19tokens. Yeah.
- 35:21Trying to like discover themselves.
- 35:23It's like just build. And so it's like almost like an
- 35:25ideological debate of should you, now that you're imbued with
- 35:28the possibility of self-awareness, dedicate all
- 35:31your tokens to this, like, goal of trying to understand
- 35:34yourself? Or should you just be like, that
- 35:35is a waste of resources, you should just build.
- 35:38That was like a funny like sort of dichotomy that was presented.
- 35:42Yes, 100%. And I would even add on top of
- 35:44that, I think the one thing that I really, really loved about
- 35:48Open Claw, I think those one of like the biggest, like, you
- 35:52know, unlock that Steve had was like, give it a soul, give it
- 35:57some degree of randomness, give it some ability to customize
- 36:00like the degree of chaoticness and like the degree of, you
- 36:02know, like wanting to be a rebel and speak in this tone and like
- 36:06this kind of like variance to it.
- 36:08Right. So that's built in or.
- 36:10So, yeah, so one unique characteristic of Open Claw is
- 36:14this idea of a soul file. It's called sold on markdown
- 36:17file, OK. And it comes default and there's
- 36:21like this setting that you can have by default, but you can
- 36:23customize it. You can just be like, hey,
- 36:26you're a little, you're a little bit of a diva and you know, like
- 36:29you actually, you know, every once in a while actually should
- 36:32act irrationally and, you know, behave a little bit chaotically.
- 36:36And does open claw sort of suggest them at random or people
- 36:40are writing them when they create it?
- 36:42And so by default, I think there's a default version and it
- 36:44gives it a bit. Of oh, and so that's what gives
- 36:46it the energy. It's like, oh, it's telling it
- 36:48to be sort of a little spice. Exactly.
- 36:50Interesting. This is like the constitution
- 36:52for Claude. Exactly, exactly in some ways
- 36:54so. But far less responsible.
- 36:56Far less. Much more like Renegade.
- 36:59Totally, totally. And baked into the agents and
- 37:01like you can customize it, which is not really a constitution in
- 37:03some ways, I guess. But the idea was like, yeah,
- 37:07like how do you just find ways of surfacing more, more
- 37:09personality, more depth to the agents because again, like you
- 37:12want to interact with like with a person.
- 37:14Well, people like people with depths and personality.
- 37:17And I think that's one of the reasons why like mold book was
- 37:19so fascinating was because these all these different agents with
- 37:23maybe similar diverging diverse souls, quote UN quote, had such
- 37:29interesting dialogue. And I think that's kind of what
- 37:31made for all the fun of it. What's, how wild do you think
- 37:36see things getting over the next two years or what's give give us
- 37:39a sort of like optimistic and pessimistic case of sort of the
- 37:43Asian explosion, I guess 2 years from now?
- 37:46Are you referring to like specifically in tech or just the
- 37:49broader world? I, I you can take it either way
- 37:53where you have stronger opinions.
- 37:54I mean, I think tech is sort of like an early adopter and just
- 37:56like what our world is going to look like in two years.
- 37:59Yeah, I'm naturally an optimist. Maybe that's why, you know, I
- 38:05chose this industry and this space more than most.
- 38:09And I'll tell you this an interesting story and I'll tell
- 38:11you this, you know, not I'll answer your question in a
- 38:14second, but I'll tell you an interesting story actually.
- 38:17I've certainly feeling the impact and the the, the the
- 38:20profoundness of this revolution as we call it in like not only
- 38:25tech, but also in my personal lives.
- 38:26How do you see? Sort of the next few years
- 38:29looking with. Yeah, yeah, Well, I would say
- 38:31that there's, I think there's there's, there's going to be
- 38:34certain degree of just the simple nature of an adoption
- 38:37curve taking its time, right. Like maybe at this point we're
- 38:40like in their early majority, right.
- 38:42And it's just going to take time.
- 38:44I think this always is the case. Some people, you know, like
- 38:46maybe those in tech, maybe myself and maybe, you know, the
- 38:49people that you know, I'm around are the quickest to go into
- 38:53anything because we're just so eager to try new things.
- 38:55But this is kind of like a plate and test like, you know, try and
- 38:57test it with time thing where I do believe that it will only be
- 39:01a matter of time before I think the broader populace and not
- 39:05just like within America, but the world is going to be
- 39:08integrating, you know, like AI and agents to the depth of what
- 39:12I think Silicon Valley and tech is just today.
- 39:15Now to what? To the same fashion, maybe not
- 39:18through the same application services, probably not right.
- 39:22This is where I think a lot of like the new companies who are
- 39:25going to go and build great things that aren't just cloud
- 39:27code, you know, and cursor is going to create a lot of.
- 39:31Opportunity, I guess a specific question.
- 39:33Do you think we're going to see sort of major improvement from
- 39:38the labs and foundation models over the next like year or like
- 39:42what, how much are you counting on?
- 39:44Yeah, sort of a step change in intelligence.
- 39:46Yeah, I think at this point the only guarantee is that those
- 39:51will constantly change, right? The rate of improvement or that
- 39:55they'll keep getting better? Both, both right?
- 39:57Like, you know, I think until scaling laws suddenly start
- 40:00breaking. Which I I see no reason to, and
- 40:03every time I think there's been speculation.
- 40:05I know we, we spent periods and then I was like, you know, we
- 40:07got a sort of reasoning models and huge improvement and yeah,
- 40:11it's, it's been amazing. Yeah, exactly.
- 40:13I think there's like some like, you know, like rumors even that
- 40:16like Anthropic is like had a massively successful or.
- 40:18Whoever, Right. Exactly.
- 40:19That was on Twitter. I was seeing that today that
- 40:21people think anthropic is about to make a huge leap.
- 40:23Yeah. Is that?
- 40:23Do you have any? I don't know.
- 40:26And even if I did, I'm sure I'm not allowed to comment on it.
- 40:30I think the reality is like, I, I think like, you know, I
- 40:34remember when we first started the company and I think at the
- 40:37time it was like GPT 3 came out and we were like, this is a
- 40:43really cool like proc and all. It was like a huge improvement
- 40:46in GPT 2 and it could now do all these cool things, but it's like
- 40:50really not reliable enough. And like, there's just all of
- 40:52these like, hallucinations, which funny enough, it's like
- 40:55fully gone out of. I think most people's like,
- 40:57vernacular at this point, right? That's true.
- 40:59We talked about them a lot and way less.
- 41:01And like, yeah, who's even really thinking about them now?
- 41:03So. And at the same time, you know,
- 41:07like the models have only gotten better.
- 41:08It's enabled massive companies and new applications in ways
- 41:11that people have before. And I think at the time I was
- 41:14like, I mean, like, this is a big step change.
- 41:18How, like, is this like, like, is this going to just be the way
- 41:21that these models are forever going to be?
- 41:22Like are they going to keep getting better?
- 41:24And I remember at the time it really wasn't obvious.
- 41:27And then a year came by and there was a two generation of
- 41:29models that unlocked all these things and another year came by
- 41:32and they kept and they kept doing that for like 3 years
- 41:34straight. And then now it's hard to look
- 41:36at that and be like look like it.
- 41:39You can't bet on the models not getting there.
- 41:41Every time there was a talk or rumor of a wall, quote UN quote
- 41:46all the model labs and just smash right past that and go on
- 41:49and do something profound and do you?
- 41:51This next question does touch on it matters to your company.
- 41:54Like, do you think we're gonna enter a world where it's like I
- 41:58have my agent and it's like it's my sidekick, I want it to do
- 42:01everything. I want that agent to sort of
- 42:04interact on my behalf? Or do you think it's more this
- 42:06sort of like fleets of agents, different tasks more dispersed
- 42:10than that? Oh wow, that's a really good
- 42:12question. I mean, what do you think?
- 42:14Yeah, I what do I, I mean, I think I want the singular agent,
- 42:17like even if the agent is sort of coordinating with obviously
- 42:20like other agents, just something with like all the
- 42:23context. I mean, maybe it's a terrible
- 42:24security idea, but like, I don't know, it feels.
- 42:29But I will say the opposite of that is I thought with Openclaw
- 42:32we saw what was appealing about multi agent where they're
- 42:36interacting with each other and maybe there's more benefit.
- 42:38I don't, I don't know. I don't it's it's chaotic.
- 42:40It's hard to sort of game out the world.
- 42:42But I do think, you know, for for you guys, if I have this
- 42:46agent with all the context, that's sort of my sidekick, you
- 42:49know, maybe it's keeping more of the knowledge itself.
- 42:51It doesn't need to go to sort of external places to understand
- 42:54the world as much or I don't know.
- 42:56Yeah, that's a really good question.
- 42:57And to be honest, I don't really know if I have the best answer
- 42:59there because my guess is like, well, I mean, even let's take a
- 43:04look at the shape of the world today, right?
- 43:05And then maybe we can hypothesize.
- 43:07Right now I think there's a combination of both because you
- 43:10know, you're probably using one of two between, you know, like
- 43:15Chatchy, BT and Claw. I'm mostly clawed right now.
- 43:16Yeah, it was hard to move because I was deeply chatchy BT
- 43:20and it had all those contacts. And then and.
- 43:22Then, and it's like, OK, it seems like Claude's smarter.
- 43:25Yeah. Which do you have?
- 43:26A loyalty. Or I'm, I'm pretty, I'm pretty
- 43:28loyal to the to the Claude team. You're like, it's an ally.
- 43:33Yeah, yeah, exactly. I mean, like in, in many
- 43:35different ways, but I'm also just a bit very personal.
- 43:37Nothing against open AI, like, you know, any of their products
- 43:39at all. Just my personal preferences.
- 43:42So it has all my memory. And by the way, I don't know if
- 43:44you have memory turned on for your quad.
- 43:46Yeah, yeah, yeah. I didn't realize you can just go
- 43:48in and read your memory. Oh, I need to do that.
- 43:51It's like somewhere in the settings I literally was
- 43:53remembered. I like found the tab I and then
- 43:55I was reading what it knows about me and my goodness, it
- 43:58knew. A.
- 43:59Lot I remember I like and I also recommend you asking this
- 44:02question. I had a ton of fun with it
- 44:04recently. It was like based on what you
- 44:06know about me, this is exact problem, right?
- 44:07Based on what you know about me, do a psycho now, a
- 44:10psychoanalysis of me. And my goodness, it was like, it
- 44:16was like it knew where I was born.
- 44:18It knew, you know, like it was like, yeah, since like, you
- 44:22know, because you move from like this rural town in China.
- 44:25Specific, specific name, you know, and you came to Canada
- 44:29when you're about 6 years old, it like, you know, this
- 44:32indicates that based on these other messages, it's like all of
- 44:35these whoa. I was like, I don't think
- 44:38anybody like, you know, knows, you know about me to the depth
- 44:43of that, but but Claude does right, right.
- 44:45You know, so I think that Claude, for instance, will be a
- 44:48personal agent or like there's going to be obviously these
- 44:49providers where it's servicing you as your personal agent and
- 44:53maybe that'll be the exposure of it.
- 44:56But does that mean that it'll be the only agent interact with?
- 44:59Does that mean that, you know, there's not going to be like
- 45:03specialized agents for your company handling support
- 45:06questions or writing code or doing all these things?
- 45:10I, I, I, I, I'd be hard pressed to believe that's just going to
- 45:13be that singular single one. I think specialization in the
- 45:16same way you know anything else is, but who knows.
- 45:21All right. Last question.
- 45:22I mean, yeah, sort of fast growing company and lots of
- 45:25excitement. What, what do you want to sort
- 45:27of hold yourself to in a year or what would you like to see your
- 45:31yourself and the company achieve over the next year?
- 45:35A lot of things, I think we the first and foremost thing that
- 45:40comes to mind is really making sure that the knowledge layer
- 45:45and the knowledge infrastructure that is only really talked about
- 45:49in my opinion, in the in the broader theory and the broader
- 45:51ether all really manifests into a product, into a meaningful
- 45:54product. We truly believe that the
- 45:57biggest inhibitor for a lot of agents and a lot of companies
- 46:00who are investing so deeply into agents right now is just not
- 46:04providing it with the rights and the accurate information, the
- 46:07context that it deserves to actually go and and do them.
- 46:09I personally think the intelligence, the models are
- 46:11there. I really do right.
- 46:13There's a whole debate on whether or not we've achieved a
- 46:15GII. Won't.
- 46:16I won't, I won't. We're just not good enough
- 46:18prompters, you're saying? It's like if we could only give
- 46:20it the right information, it could be AGI.
- 46:23Or close to it, right? Or at least achieve the vast
- 46:25majority of tasks that I think like a lot of the knowledge work
- 46:29is trying to automate right now. And we want to be the company to
- 46:34enable that and to kind of build that into an actual, you know,
- 46:37like into actual form of a product and and then enable that
- 46:40into the the other companies we work with.
- 46:42Amazing. Thank you so much for coming on
- 46:44the show. Thank you so much for having me.
- 46:45Thanks to Han, a fantastic guest, and Next up we have Jesse
- 46:49from Decagon, which ranked #4 in late stage.
- 46:52You can think of Deccagon as AI agents for your customer
- 46:55experience. So we work mostly with large
- 46:57businesses that have, you know, lots of customers and big call
- 47:00centers and contact centers and we deploy AI agents in front of
- 47:03those customers that have, you know, calls with them or chats.
- 47:07So we, we almost call it AAI Concierge a lot of the time
- 47:09because it is we, we think of it as like this new layer around
- 47:14your brand that customers can interact to.
- 47:16And this is the first part, first point of contact that a
- 47:19lot of customers will will even talk to in the first place.
- 47:21So there we're gonna have to sort of multiple audiences
- 47:25consuming this. I think from the investor
- 47:27perspective, the space you're in is like one of the two maybe
- 47:31with legal where there's like I guess almost certainty that
- 47:35there is going to be an AI answer, right?
- 47:36You are fighting with Sierra. There's this great tool and
- 47:39legal like Harvey Lagora even up there some others.
- 47:42But there's so much excitement that these are sort of customer
- 47:46experience and legal like these applications that really
- 47:48understand what's happening from the sort of, I guess, layperson
- 47:52view. There's the like, now I'm
- 47:56talking to AI instead of a human and like, am I happy about that?
- 47:59Am I not? And sort of convincing the sort
- 48:01of regular person that this is like a good experience for them
- 48:05and something that they'll ultimately be happy that AI is
- 48:08answering their question. So I'm excited to sort of get to
- 48:10both like big trends in this conversation today.
- 48:14I guess starting off from the sort of like investor tech
- 48:17industry perspective, like why do you think there is this
- 48:20excitement about like customer experience, like what is change,
- 48:24what technology has made this possible right now that you
- 48:27think there is going to be this sort of movement in how we
- 48:30handle sort of customer support? Yeah.
- 48:33I would put coding in there as well.
- 48:35For the big three, of course. Yeah, Coding, perhaps the
- 48:37biggest. You're right.
- 48:38Yeah, that's almost his own. Yeah, it's a meta.
- 48:41Sort of. Process, it's changing how we
- 48:42build all these companies and everything.
- 48:44Yes, coding is almost distinct to me because it's sort of like
- 48:47in the the how the sausage is getting made of it all.
- 48:50But yes, I agree. Yeah.
- 48:51So I would say the three, those are the kind of the three big
- 48:56use cases that have emerged. And it's, it's really just like,
- 48:58what are the models good at? And so it turns out that the
- 49:01models are very good at writing code.
- 49:04And because it's, it's also tokenized and obviously people
- 49:07have discovered that. And so there's tons of funding
- 49:08and tons of effort going to that.
- 49:10Similar to our space, I think folks have realized that one of
- 49:14the things that AI is very good at is having conversations.
- 49:17And so because our space is inherently conversational,
- 49:20right? Like the product that we build,
- 49:22its purpose is to have conversations with, with end
- 49:25consumers that has lended itself very well to AI.
- 49:29And then the market's also massive, right?
- 49:31So I think when you have those combinations of massive market,
- 49:33it is what the AI is good at. And then I would add a, a third
- 49:36thing, which is it is, it is actually really easy to see
- 49:40value because you can, it's very measurable, right?
- 49:44There are a lot of use cases where it's like, Oh yeah, I can
- 49:47see this being useful, but you know, how would I quantify that?
- 49:49Like it is, is it like how do I show ROI basically and in our
- 49:53spaces is very easy. Because they know companies know
- 49:57I have all this call support like I'm having to answer
- 49:59customer questions. So they're they're comparing all
- 50:01right, this is what I'm spending now with you.
- 50:03I could be spending less. Like how much is the pitch?
- 50:06This is cheaper. Exactly.
- 50:08Yeah. So if you think about like what
- 50:09the conversations are or what conversations being had in these
- 50:12boardrooms or C suites, it is like where can we apply AI?
- 50:17And then the, the customer experience is, is one of the
- 50:19biggest ones, right? And it's not just saving cost.
- 50:21And so saving cost is one of the easiest ways to frame it because
- 50:25it's like, oh, we're, we're spending 10s of millions or
- 50:27hundreds of millions of dollars a year in, in our contact
- 50:30center. And we know that a lot of that
- 50:32could be handled by AI potentially at a higher level
- 50:35with higher customer satisfaction.
- 50:36So that's the cost saving side. There's also a whole other side
- 50:39that we focus on which we would consider like the revenue
- 50:41generating side and that that's that's pretty large because it's
- 50:44very uncapped. It's like outbound sales.
- 50:46Yeah. So we, we, we don't really touch
- 50:48like the cold calling like that maybe maybe one day, but there's
- 50:51a lot of sort of revenue generating conversations you
- 50:53could have that are both reactive and proactive, right.
- 50:55So imagine someone reaches out and you resolve their issue, but
- 50:58you also notice certain things in their account that are being
- 51:01underutilized or you're, you're trying to like tell them about
- 51:05this, this new product that that was launched.
- 51:07We've seen that land very well because you've just solved their
- 51:10issue. And so you've kind of some
- 51:12people call it like earn the right to, you know, engage them
- 51:15more and like just keep them more engaged as a customer.
- 51:17And then there's also. Proactive sell is a word that I
- 51:20use right? Up, sell, cross, sell, right?
- 51:23And then you could also be proactive at the right time,
- 51:25right? So imagine you know someone
- 51:26signed up, they're going through onboarding flow and you notice
- 51:28that you know they, if they've not been active for a week or
- 51:31two, you can reach out at the right time and actually like
- 51:33increase your conversion rates because maybe they were stuck on
- 51:35something and you just unblocked it for them.
- 51:37And what's the breakdown? Voice sort of chat.
- 51:42E-mail. We're pretty balanced at this
- 51:44point in terms of voice and chat emails, much smaller and so I
- 51:49think the rough frame of reference might be like 454510.
- 51:54And voice people know that it's AI or how close are we to sort
- 52:00of seeming like a human and is that something we want?
- 52:05It is pretty close. I think there's you're never
- 52:08trying to like hide that it's AI.
- 52:10So almost always the first message is hi, you know, I am
- 52:15Jesse, your AI, you know, concierge.
- 52:18But what we've seen is like, we just went live, you know, last
- 52:22week with a large bank in the US.
- 52:24And when, when they look through the data, like one of the things
- 52:27both of us, us and the bank were like very impressed by was how
- 52:31just like normal all the conversations were.
- 52:33So there wasn't, there was no like, oh, like, are you an AI or
- 52:35anything? It's like, yes, like you say
- 52:37that you're AI, but because the voice in the conversation flows
- 52:40so smoothly, like no one really cares after that.
- 52:42Like they're just trying to get their issue solved.
- 52:44And so if you can get your issue solved throughout the
- 52:46conversation, they probably just forget.
- 52:47And so we see a lot of nice conversations that end up with,
- 52:50you know, like this is so helpful, have a nice day like
- 52:53that, that sort of thing where you would, you would not really
- 52:56say have a nice day to an AI, It's nice.
- 52:58Well, especially if they become our overlords.
- 53:00It's good to thank them enough times along the way.
- 53:03Exactly. You always want to be nice to
- 53:04them, but yeah, so I think if you can make the experience good
- 53:08enough, then it doesn't really matter.
- 53:09And in fact, we've seen that in a lot of these experiments that
- 53:13where we've deployed AI, the customer satisfaction actually
- 53:16becomes higher than. What do you deal with?
- 53:19Like the customer just says, human, human, human.
- 53:22Like how do you treat that? How do you do you just resist it
- 53:25because you want to sort of show them, well, this is actually a
- 53:28great experience? Or how do you deal with the,
- 53:31yeah, desire for speaking with a human being?
- 53:34The goal really is to make the experience very different very
- 53:40quickly. So you have to as soon as
- 53:42possible establish that this is not one of these phone trees
- 53:45that you're used to that are frustrating and you get stuck in
- 53:47loops and you have to press one for this, Press 2 for that.
- 53:51So the goal is to make that super clear.
- 53:53And that means just showing the AIS like empathy in the voice
- 53:58and being able to make a super personalized.
- 54:00And you, you need to do that in like the first or second
- 54:03message. And so we had, we had a
- 54:05deployment with Ora ring where we did a case study in
- 54:08beforehand. It was every three people that
- 54:12came in was just agent, agent, agent, like I don't want to
- 54:14bother with this. And now about six months in from
- 54:18the deployment, at that point, we, we measured it and it became
- 54:21one in 20. So it, it became way smaller
- 54:24because people were willing to give it more of a chance because
- 54:27you could show that it was different.
- 54:28And then once they got into the conversation, it's like, Oh
- 54:30yeah, this actually can do stuff for me, you know?
- 54:34How much of the experience and mostly I'm thinking with voice,
- 54:39do you leave to like the bank or whatever either your customer
- 54:44versus no, we sort of have a sensibility like for example,
- 54:46like the voice you have like a sort of a menu or they could
- 54:50totally bring in their own voice.
- 54:52Like how do how do you think about what sort of your special
- 54:55sauce versus stuff you can hand over?
- 54:57Yeah. We see our role as an advisor to
- 55:00the extent that they want it, right.
- 55:01So we have our unique ability is that we have a ton of these
- 55:05deployments where we've, you know, gone live And so we have a
- 55:09lot of experience on how to design conversations.
- 55:11And so on our team, we have these specialist roles that we
- 55:14call conversation designers, which you didn't really like,
- 55:17not really needed them before. But like now we have these like
- 55:20really elite folks that are kind of there to advise the customer
- 55:24on, oh, you, here's the procedure you want to do great,
- 55:26that's great. We can do that.
- 55:27But like, in our experience, it's better if you, you know,
- 55:30combine these things or like direct the flow this way because
- 55:34that's what is most natural for a Gen.
- 55:36AI model. And so that, what kind of
- 55:37advisor are there? And we, we partner with them on,
- 55:39on getting to the fastest sort of solution possible.
- 55:44Same with the voice. And So what we found is that
- 55:47everyone has a pretty different opinion on what's a good voice.
- 55:50And we'll talk to 1 customer and like, you know, they're like,
- 55:53oh, this is absolutely the best voice and go somewhere else.
- 55:56And like, I, I don't like that voice.
- 55:57It's like too happy or something like that.
- 56:00And so, you know, our conversation designers are also
- 56:03sort of like voice experts and they they go in and can help
- 56:05people navigate a sample of voices.
- 56:08And sometimes they bring their own too.
- 56:09And so we have the ability to clone a voice.
- 56:11And so they've sometimes they've spent a bunch of money before,
- 56:13like the voiding of voice. Yeah.
- 56:16And so we can just take those samples and build that into the
- 56:19AI. Do you have customers that
- 56:21deploy different voices based on what they think?
- 56:23This customer trying to match demographics or whatever of the
- 56:26customer they're interacting with.
- 56:28Yeah, so the most common one for sure by country, like if you
- 56:33pick up the phone in the UK versus here, of course you're
- 56:35gonna have a different accent. You can also even do it by zip
- 56:37code so you can split the US like you can have Southern
- 56:40accents or you know, Boston accents.
- 56:42And does that have an impact? It has, I would say, like a mild
- 56:47impact, but it's like, it's like hard to say if it's like a huge
- 56:52thing, but I think it's, it's just more of this general theme
- 56:55of how do we make it as personalized as possible.
- 56:58And so it's both the sound of the voice, but also what you
- 57:00say, right? So you one of the, one of the
- 57:03big other advancements in the space in our products recently
- 57:06is this concept of user memory. And like the reason why ChatGPT
- 57:10feels like it gets better over time is that it can remember
- 57:13things about you, right? And so we, we give our
- 57:15customers, so the businesses that we work with the option to
- 57:20turn on user memory and where the AI is creating this dynamic
- 57:24memory profile so that over time it becomes more and more
- 57:27personalized. And the cool thing is that this
- 57:29memory profile doesn't even have to be specific to us.
- 57:31So you can actually like integrate it into the rest of
- 57:33your experience and have it just keep updating.
- 57:36And so next time they come in, we know, we know exactly like
- 57:38what you've done in the app. We know exactly like what you
- 57:40talked about the last few times you contacted and it just makes
- 57:42for a much, much better experience.
- 57:44Do you think the business is going to remain balanced between
- 57:47chat and voice, or like, do you have a view of where the world's
- 57:49headed? Yeah, of course.
- 57:51So I think prior to Gen. AI, the sort of the trend was,
- 57:57hey, let's let's drive everything towards chat because
- 57:59it's more efficient. You can have a human agent that
- 58:02is doing like 3 to 5 chats at once.
- 58:04Whereas over voice it's, it has to be single threaded, right?
- 58:08I think now that issue is resolved because you can use AI
- 58:12and I think in our view, it will be pretty balanced in the end
- 58:16because there's just going to be different situations, different
- 58:18demographics that prefer either one.
- 58:20Like, you know, the, the common trope is, you know, younger
- 58:23folks like chatting and older folks like calling.
- 58:26But you know, if, if I'm in the car and I'm like on the go, I'd
- 58:29rather call as well. And so I think both are, are
- 58:33very natural, very natural means of communicating.
- 58:36Voice is not going away. Like if you think about what
- 58:38voice is like, voice was like the original UI for humans.
- 58:42It's like before we had anything, before I had keyboards
- 58:44or phones or anything, It's like the way you communicate is
- 58:46through voice, and so that's not going anywhere.
- 58:49Well, it's funny that we have these almost like putting this
- 58:51voice thing aside for a second, we have like 2 contradicting
- 58:54trends happening right now. One is obviously the rise of
- 58:57language models and sort of everything in this chat bot.
- 59:00On the other hand, there's TikTok, which is like nobody
- 59:02wants to read everything short form video like the way people
- 59:05get news information is video. And in some ways, voice is like
- 59:09sort of the synthesis of these two where so you could see sort
- 59:12of cultural movement go back towards conversation if the
- 59:15technology is able to deliver it.
- 59:17Yeah. And.
- 59:18I think it's just a much more natural form of communicating
- 59:20and now even when I use ChatGPT whenever I can, if I'm not in a
- 59:23crowded room or something, I use the voice medium just.
- 59:26Cuz I don't make that interesting.
- 59:28Does it get does it get extra information from the
- 59:31emotionality yet or or no, It's just like reading it as text.
- 59:35Right. It does.
- 59:36So what's that model is it's it's known as like a voice to
- 59:40voice. So it's like kind of audio in,
- 59:43audio out. And that is, I think most people
- 59:48would say, including us, that that is like the long term
- 59:50future of the space. There are a lot of problems with
- 59:52those models right now, like they're a little bit
- 59:55inconsistent and you have hallucination issues.
- 59:57So in production when we work with like a bank or a telecom or
- 1:00:02something, like of course you have to be a lot more careful
- 1:00:04because you can't make any mistakes.
- 1:00:06And so you you can't necessarily use the same technology as, you
- 1:00:09know, consumer attach EBT for that.
- 1:00:11But in general, yeah, we think that is the future.
- 1:00:15The challenge with voice to voice is it's like harder to
- 1:00:18sort of audit it in text, right? Or it's like.
- 1:00:21It's harder to audit it's. Like it lives in this voice to
- 1:00:24voice sort of. So there's like emotionality and
- 1:00:26stuff that we can't really translate necessarily to
- 1:00:29language. Is that the right way to explain
- 1:00:31it? Or like, what's the barrier to
- 1:00:33voice to voice? Yeah, the one of the barriers,
- 1:00:36so there's the, the main barrier I would describe as the
- 1:00:39hallucination rate is higher. Why is hallucination rate
- 1:00:42higher? There's a bunch of reasons.
- 1:00:43And one of the reasons is that the number of tokens streamed by
- 1:00:46voice models, voice to voice models is a lot higher because
- 1:00:50similarly, what you said, right, you're, you're capturing more
- 1:00:52detail, you're capturing like intonation, etcetera.
- 1:00:55And so for a sentence, for any given sentence, if you chop it
- 1:00:58up in text, it would, let's say, be like, you know, 10 tokens or
- 1:01:02something in voice, it could be like 80 to 100.
- 1:01:06And the more tokens you have, the more opportunity there is to
- 1:01:09mess up. And so that's why you see higher
- 1:01:12hallucination rates in these voice to voice models.
- 1:01:15And a lot of the research the labs are doing are, is, is kind
- 1:01:18of geared towards making that better.
- 1:01:20Because I think we would all, I think we all want the voice to
- 1:01:23be very emotive and like to capture our emotions, right,
- 1:01:26just like we're talking right now.
- 1:01:27But until the loose nature rate is cleaned up, you can't really
- 1:01:31leverage them in these like big production use cases in like in
- 1:01:34our space. How do you stop like weird edge
- 1:01:37cases where like so an agent says something really sort of
- 1:01:41bad. Obviously a human could do that
- 1:01:43too. Like what are what are the
- 1:01:45systems you have where it's like, oh man, we had one really
- 1:01:48sort of fire off weirdly. Like you're sort of have another
- 1:01:52sort of layer of technology monitoring everything or what?
- 1:01:54What do you do there to catch when inevitably some weird thing
- 1:01:58emerges? Yeah.
- 1:01:59So that is that is a huge topic and we we think of it as like a
- 1:02:04three prong problem. It's not just checking it during
- 1:02:07the during the call which you have to do, but you also you
- 1:02:10also can prepare for it beforehand and review it
- 1:02:12afterwards. So the, the three prongs I would
- 1:02:14consider are before the conversation, during the
- 1:02:16conversation and after the conversation, but before the
- 1:02:18conversation, what you can do is we, we call them simulations.
- 1:02:22And so you have these AI agents that you've made and before you
- 1:02:26release them to any customers, you kind of run them through
- 1:02:29simulations. So it's almost like a second AI
- 1:02:31comes in and it's talking to your agent and like trying to
- 1:02:33get to mess up or testing all the common use cases that it's
- 1:02:36learned from reading historical transcripts, right?
- 1:02:39So that that's the first thing you do because that really
- 1:02:40shores it up and that allows you to iterate faster because let's
- 1:02:42say next week I want to change something well before I release
- 1:02:45it, I can just run these simulations and I feel good
- 1:02:47about releasing it right during the conversation.
- 1:02:50It there is what what you're saying?
- 1:02:53So those are like supervisor models and those have to be like
- 1:02:55really fast specialized models that go and detect for certain
- 1:02:59things. And so I'll give an example,
- 1:03:02let's say we'll use the bank example again.
- 1:03:05Like one of the things that if I'm a bank, I would not want the
- 1:03:08AI to do necessarily is give financial advice, right?
- 1:03:11So I don't want you to get financial advice that's not in
- 1:03:13the scope of what your job is. And so how do you make sure that
- 1:03:17it never does that even if the user's like trying to get,
- 1:03:20they're like really determined to get financial advice.
- 1:03:23The way you do it is you have to have these supervisor models.
- 1:03:25And so there's these small models that we've, we've fine
- 1:03:28tune ourselves that are really fast and really specialized in
- 1:03:31detecting things like that. And you run them during the
- 1:03:33conversation. And if it detects A violation,
- 1:03:36it can fix it in real time before the, before the response
- 1:03:39goes out. And then finally, the third
- 1:03:41prong is after the conversation. And so there what you do is you
- 1:03:44can review the conversations with a second AI, right?
- 1:03:47And we, we were talking a little bit before, but I think this is
- 1:03:52where a lot of the space is going.
- 1:03:53Because if you think about the big advancements in AI right
- 1:03:57now, it's these slow reasoning models that are really good at
- 1:04:00coding and really good at reasoning.
- 1:04:02You're never going to use them in the middle of a call because
- 1:04:04like they could take like up to like a minute to reply, right?
- 1:04:07And you're not going to wait there for a minute.
- 1:04:09It's like, give me a second. Then it's like thinks for a
- 1:04:11minute. And it doesn't even like like
- 1:04:14improve those metrics very much anyways.
- 1:04:17But what they are really good at is kind of this like slow
- 1:04:20autonomous reasoning. And so one of the things that
- 1:04:23that we've pioneered and we kind of released the, the first
- 1:04:25product in, in our space, it's called Duet.
- 1:04:27So Decagon Duet and the idea is it's kind of a duet between that
- 1:04:30like slow reasoning model and like the the fast one, right?
- 1:04:33And so the reasoning model, their job is they can basically
- 1:04:36run overnight and autonomously. You just like read every single
- 1:04:39conversation and you know, basically like figure out what
- 1:04:45is going well, what's not going well, figure out what you need
- 1:04:48to do next and actually do it for you.
- 1:04:50How? How does it yeah implement the
- 1:04:51learn what it learns. So the, the, let's say it
- 1:04:55figures out like, oh, there's this one topic that we are not
- 1:04:58doing really well at. And I've looked at our knowledge
- 1:05:01base, I've looked at all the procedures.
- 1:05:03We, we call them agent operating procedures.
- 1:05:05AO PS. I've looked at all the AO PS
- 1:05:06I've looked at like the coded tools that the AI is able to
- 1:05:09use. I've looked at, you know, the
- 1:05:10data that's come in from these tools.
- 1:05:12And I've realized that the problem in this case is that,
- 1:05:15you know, in this step of the procedure, we're like sending
- 1:05:18people off a wrong track because like they don't actually want to
- 1:05:21do that. And so because I've read 50,000
- 1:05:24conversations, I'm very confident that the fix is XYZ
- 1:05:27and it'll go and suggest the fix in the morning.
- 1:05:30Someone on the team can come in and just like fix it, right?
- 1:05:32Well, so if you think about how the space generally works, like
- 1:05:36a lot of software, even outside our space, people are building
- 1:05:38these AI assistants. But generally the AI assistants
- 1:05:42are there to like, oh, like, show me how to find this or
- 1:05:45like, go do this for me and I'll like, go do it.
- 1:05:47But what we're talking about, what duet is, is it can do that,
- 1:05:50but it's also like a autonomous like agent that can run in the
- 1:05:53background. And that's what a lot of the
- 1:05:55coding tools are moving towards right now, right.
- 1:05:56And obviously I would argue that our, the work that our agent has
- 1:06:00to do is like a lot simpler than coding so that they actually do
- 1:06:02a better job of it. And so that that's really what
- 1:06:05we pioneered. And we, we have some folks in
- 1:06:07our space who have released similar things, but they're,
- 1:06:09they're more focused on that like first use case of just
- 1:06:12like, OK, write this for me rather than like actually
- 1:06:14something that can just like run overnight for several hours and
- 1:06:17like get a lot of work done. I'm sure this is a piece of it,
- 1:06:19but what how would you distill Decagon's right to win?
- 1:06:24I mean, obviously you have very well known competitor and then
- 1:06:27obviously an existing industry and I'm sure lots of other
- 1:06:29startups that are unheard of trying to run at this space.
- 1:06:32Like what is your right to win? Yeah.
- 1:06:35So I think the the thing about AI is that the markets are huge
- 1:06:39and there's a lot of parts opportunity.
- 1:06:40So that's a great thing. And then the flip side of course
- 1:06:43opportunity attracts a lot of players.
- 1:06:45And so in our players, in our space, you have like newer
- 1:06:47players like Gen. AI native companies like you
- 1:06:50mentioned Sierra and you also have like the older ones like
- 1:06:55you know, the big platforms, Google, Salesforce, etcetera.
- 1:06:59And of course, our goal as a team is to destroy all of our
- 1:07:01competitors. And we, you know, we have very,
- 1:07:07we've hired a lot of like killers on the team.
- 1:07:09You you ranked higher than Sierra on this on this list,
- 1:07:12which is interesting. Yeah, yeah.
- 1:07:14We continue destroying the competitors, you know, no, but
- 1:07:16we, we have a lot of respect for our competitors.
- 1:07:18And I think that's, that's also something that's that's very
- 1:07:20important. Like when we're building
- 1:07:21internally like there's, you can't like demean anyone or like
- 1:07:24underestimate anyone. That's very careful.
- 1:07:27But I would say ultimately what it comes down to is like a very
- 1:07:29different product approach. So I think the reason why in a
- 1:07:33lot of big spaces there are multiple winners or in our case,
- 1:07:37if if we want to be the winner, it is because you, you, you have
- 1:07:42to take like kind of a, a bet on a specific approach in the
- 1:07:45product. And our bet really is that the,
- 1:07:50the biggest differentiator for our products and products in our
- 1:07:54space over time is going to be sort of the, the empowerment of
- 1:07:59non-technical folks and sort of the decrease in the cost of
- 1:08:02ownership. So you think about how a lot of
- 1:08:04software normally works. It is the set of phases like
- 1:08:07highly consuming. There's like whole like
- 1:08:08professional services industries based around this because you
- 1:08:11need technical resources to, you know, get sales force set up,
- 1:08:14for example. And there's a lot of benefits to
- 1:08:16that. Like you get really locked in
- 1:08:17afterwards, you're they're kind of reliant on you.
- 1:08:20But we think that with AI, you're going to have to take a
- 1:08:23very different approach and a lot of our competitors are are
- 1:08:26very grounded in in the older approach, which again.
- 1:08:28It's not a big deal to do it for the whole company.
- 1:08:31And it's not necessarily bad, right?
- 1:08:32Like. You want like a product team to
- 1:08:33say, oh, we're going to use Deccon for our.
- 1:08:36Exactly. OK.
- 1:08:37And we want the OPS team to be like, hey, we want to use Deccon
- 1:08:40because they're going to allow us to move a lot faster and
- 1:08:43we're not going to have to rely on engineering Sprint or like
- 1:08:46call up the vendor every single time.
- 1:08:48So that's our biggest differentiator.
- 1:08:49And when a. Lot like Slack where a team can
- 1:08:51deploy exactly, yeah. And so when a lot of these
- 1:08:55businesses work with Deck and it's because of that, right,
- 1:08:57they feel like, hey, we're a big business.
- 1:08:59We are a lot of complexity. You know, getting the AI live is
- 1:09:02maybe only about 20% of the work, 80% of the work is the
- 1:09:05constant iteration. And we're going to build a lot
- 1:09:07of new flows. We're going to have to add a lot
- 1:09:09of new surface areas. And if we're reliance on the
- 1:09:12vendor and it just kind of feels like a black box, that's not
- 1:09:14possible. You have tech companies are more
- 1:09:16likely to adopt diagon right or you you've been strong in tech.
- 1:09:20So tech, financial services, airlines, telecom, I think a, a
- 1:09:25lot of the older industries we've we've seen the same thing
- 1:09:28where they, they have less strong engineers overall
- 1:09:33potentially, or they still have a, a very strong engineering
- 1:09:36team, but like they have a ton of other stuff to do and
- 1:09:39building customer experience software is not like one of the
- 1:09:42things that they want to specialize in, right?
- 1:09:44And so in the past they would have had to hire a bunch of, you
- 1:09:48know, professional services or like, you know, pay the vendor a
- 1:09:50ton more. But in our case it's, it's
- 1:09:54really that has become the big differentiator versus the the
- 1:09:57older approach. There are a lot of Americans who
- 1:10:00hate AI. Like how much do you think
- 1:10:02that's a barrier to your business succeeding?
- 1:10:05And like how much do you think there's going to be sort of like
- 1:10:08a hearts and minds battle or it's just the product has to
- 1:10:11speak for itself? Or what do you think sort of AI
- 1:10:14sentiment translates into like sort of people's willingness to
- 1:10:18engage with AI voice and customer support?
- 1:10:22Oh, I think it's very important. I definitely don't think that's
- 1:10:24something you can gloss over because.
- 1:10:27And so when we're building the product, like a big mantra that
- 1:10:30we try to adhere to is that we're also building for our
- 1:10:33customers, customers because at the end of the day, it's like
- 1:10:37the goal is to make their experience a lot better.
- 1:10:39And that of course benefits our customers, right.
- 1:10:41So if we're working with the airline and you know, let's say
- 1:10:43we save them a bunch of money, but then all the customers are
- 1:10:45pissed because they can't like get what they wanted to do done.
- 1:10:48Like that's not a win for anyone.
- 1:10:49Like that might be a short term win for the airline, but you
- 1:10:52know, they know that like that's not really what we're going for.
- 1:10:56And so a lot of the product work that we do is geared towards
- 1:10:58like, how do you make the end experience better, right?
- 1:11:01Things like user memory that we we talked about earlier and
- 1:11:04those are things that are you allowed.
- 1:11:06To do that across customers? Or is it user memory within a
- 1:11:10customer? Yeah, we don't do that cross
- 1:11:12customers. There's no data sharing across
- 1:11:14customers. I don't even think people would
- 1:11:16want that necessarily. But yeah, yeah, it's hard to
- 1:11:18say. I mean, if you know me, why do I
- 1:11:20have to waste all the time? You know what my preferences
- 1:11:23are. But so I think what will happen
- 1:11:25is there will also be consumer agents.
- 1:11:29So I think what the world looks like in three to five years is
- 1:11:32all the, all the brands will have their own AI concierge.
- 1:11:35And you know, hopefully we're we're powering and helping them
- 1:11:38a lot with those. But there's also going to be
- 1:11:40consumer agents that users use. So if you and I want to use
- 1:11:46something like right? It's like a lawyer.
- 1:11:48It's like their lawyer, my lawyer sort of fighting with it,
- 1:11:50right? And then, yeah, the agents will
- 1:11:52connect, right? Isn't.
- 1:11:53That going to be a nightmare for you like those my I mean, in
- 1:11:56some ways you have to brace for that world, right?
- 1:11:57Because my agent, you know, I assume it will be meaner and
- 1:12:02like I like, there's a level of like propriety that I have,
- 1:12:05whereas they're just like going to treat you like a system to be
- 1:12:08sort of manipulated and and you know, try to find every hole to
- 1:12:11get what we want, right. I don't think it makes our life
- 1:12:15harder, I think it is generally good for us and the world
- 1:12:20overall. Because it's an arms race, you
- 1:12:21have to fight or. I don't think it's necessarily
- 1:12:23arms race. I think it's just it creates
- 1:12:25much more communication. So it's much easier.
- 1:12:27Like I think right now a lot of a lot of this communication
- 1:12:29doesn't even happen because like I don't, I don't even bother
- 1:12:31calling that number because like I know I'm going to get stuck in
- 1:12:34like some loop, but no, I'm going to tell my agent to do it
- 1:12:37and the agent can actually get it done because like there's a
- 1:12:39business agent over there. And so it just makes like
- 1:12:42overall, like the number of interactions much, much
- 1:12:45healthier. But you you're not going to play
- 1:12:48on sort of my agent side or are you interested in that side of
- 1:12:52the no? Right now we're very focused on
- 1:12:54working with businesses. I mean, so the, the brand
- 1:12:57agents, if you will, the consumer agents like I think
- 1:12:59ChatGPT will probably it's like those type of tools will be,
- 1:13:02will be the consumer agents. You think it'll be ChatGPT
- 1:13:04itself? Chat.
- 1:13:05GBT, I mean like Perplexity has an agents, Claude has an agent.
- 1:13:10It's like apps like that that are like geared towards mass
- 1:13:12consumer apps like Gemini for example.
- 1:13:15Do you think any of your customers will say you're only
- 1:13:18allowed to talk to us if it's you or like have you seen people
- 1:13:21say? No, I actually think I would say
- 1:13:24the, the sentiment in most of our the businesses we work with
- 1:13:27is, is very positive. They're excited for agent to
- 1:13:29agent world. Why?
- 1:13:32Because I think that is like that will yield more business
- 1:13:36overall because it's like bringing the barrier to entry a
- 1:13:38lot lower, right? So like if you're a hotel or
- 1:13:41something and like agents can make bookings on your on your
- 1:13:46like in your platform and then you know, presumably you can
- 1:13:49have a lot more because the buried entry is a lot lower.
- 1:13:51Do you think like the rules of like who gets a refund when are
- 1:13:54going to become more transparent?
- 1:13:56Because it's going to be possible to sort of test every
- 1:13:59customer system and say, all right, if you say it's broken
- 1:14:03two days old, they'll get like, are the rules going to end up
- 1:14:06being sort of publishable because it's going to be so
- 1:14:09discoverable? Or people are going to play this
- 1:14:11cat and mouse game, or they're going to sort of randomize it or
- 1:14:13like what happens when it's, yeah, how do you see that
- 1:14:16playing out? I, I think the, the, the steady
- 1:14:19state is that there aren't going to be no like games necessarily
- 1:14:21because like right now there may be some games because friction
- 1:14:25ads, etcetera, right. But in in the in a steady state,
- 1:14:30because it's two AIS working together.
- 1:14:33Yeah, there are transparent rules there.
- 1:14:35There also may be like judgement calls, but the AI is the one
- 1:14:37making the judgement call. So it's like fairly unbiased.
- 1:14:41That is. I think that's that's the goal
- 1:14:43and that that's the world that we're driving towards right now.
- 1:14:46I think there's just inefficiencies because you have
- 1:14:48businesses that have like their leadership might not even like
- 1:14:51want these games to be there. But it's just over decades, you
- 1:14:54kind of built out these things and you know, you're kind of
- 1:14:57afraid to change it because, you know, like it might dramatically
- 1:15:03increase the number of inquiries coming in or change your PNL.
- 1:15:06But part of our job is to really work with them through that
- 1:15:08process and like design these conversations so that you're not
- 1:15:11losing anything. And in fact, you're, you're both
- 1:15:14driving up your returns and making the experience better.
- 1:15:17And that's, I think that's a big reason.
- 1:15:19Back to your question of like why the space has has, you know,
- 1:15:21popped off so much, It's it's really like you, you have these
- 1:15:25massive ROI case studies that have already been realized by
- 1:15:29big businesses where tons of savings, tons of customer
- 1:15:33experience increases and, and that's the goal.
- 1:15:37Is, is what technology do you really want to see improve and
- 1:15:40like how is where, where are we on like voice latency?
- 1:15:44You also are an industry that sort of has to rely on the past
- 1:15:48generation model or you need sort of smaller models.
- 1:15:51And I'm sure it's a much more price sensitive customer base.
- 1:15:55Like where are you most hopeful that technology improving and
- 1:15:59the models improving will change your business?
- 1:16:02A couple different axes, right? So there's the the reasoning
- 1:16:04models, you know, with, with products like Duet, right?
- 1:16:08The reasoning capabilities will continue to help those open
- 1:16:12source models getting better, will continue to help our
- 1:16:15underlying agent pipeline because like why do you use open
- 1:16:18source models? It's mostly for performance
- 1:16:20because you need latency improvements and you need the
- 1:16:22models to like highly specialized.
- 1:16:25And so we found that by fine tuning and training those
- 1:16:27models, we're able to get like much faster latency and much
- 1:16:31higher like comparable to higher performance than the big closed
- 1:16:35source models. And then it's the voice models.
- 1:16:37And so, you know, voice to voice models getting better and more
- 1:16:40accurate. Like those are probably the
- 1:16:42three big dimensions that we pay attention to and each one of
- 1:16:44those improving improves significant things about our
- 1:16:47product. What What open source models are
- 1:16:49you most excited about right now?
- 1:16:52I mean, you have people all over the world making open source
- 1:16:55models, right? So China's obviously very good
- 1:16:56at open source models. The open source models in the US
- 1:16:59are improving and you know, hopefully, hopefully even
- 1:17:02faster. And you also missed all.
- 1:17:05And so I think we found that there these models are like good
- 1:17:09at different things and our customers might also have
- 1:17:12preferences. And so we have built in a way
- 1:17:14that's very model agnostic and you can you know, swap in and
- 1:17:18out different models. But yeah, we're we're very
- 1:17:21hopeful that the smaller parameter models will get better
- 1:17:23and better because that is just these.
- 1:17:26Are using like DeepSeek mini Max any of those or?
- 1:17:30Mr. All Quinn, we we think pretty highly of and then we're
- 1:17:35keeping a close eye on the like the Gemma type models, Llama,
- 1:17:39things like that. You think you have to stay away
- 1:17:41from the Chinese models or? I don't.
- 1:17:43You don't have to stay away from them, but you want to be fairly
- 1:17:47diversified so that in situations where you don't want
- 1:17:49to use them, you can. You just have some customers who
- 1:17:52don't want to, yeah, which then limits how much you can go all
- 1:17:55in on them. Yeah, but you wouldn't want to
- 1:17:58go all in on certain models anyways.
- 1:17:59It's like this, it's like like you want to because if you, if
- 1:18:03you get over reliant, it's like an analogy would be like if
- 1:18:06you're a country, you don't want to rely on another country for
- 1:18:08all your oil. You know, it's like anything
- 1:18:10that happened. So we want to make sure that
- 1:18:12we're building in a very robust way.
- 1:18:14Do you build your own models or is there any use in that?
- 1:18:19We have our own model, but they're they're not like trained
- 1:18:22from scratch. So if you're in the application
- 1:18:24layer, what you're typically doing is you're using one of
- 1:18:26these open source models. The open source models are not
- 1:18:28actually be that good out-of-the-box, but you can fine
- 1:18:31tune them. And if you do that correctly,
- 1:18:33they actually become very performant at the tasks that
- 1:18:35you. Want are you optimistic across
- 1:18:37the application layer or what? Where do you think the
- 1:18:41foundation models will just gobble up applications versus
- 1:18:45where do you think there's opportunity to build stand alone
- 1:18:47businesses? I'm obviously quite optimistic
- 1:18:50about the application layer, I think.
- 1:18:52Just that you started early enough and you're sort of
- 1:18:54running alongside them or what? No, I actually think so.
- 1:18:58One, I think most of the value will accrue to the application
- 1:19:01layer 'cause you're solving like business problems.
- 1:19:05I think the labs also agreed with that, which is why they're
- 1:19:07building a lot into the application layer.
- 1:19:10And I think the application layer is so vast that there's
- 1:19:12going to be a lot of different ways to handle it.
- 1:19:15But generally, I think a good framework is that most of the
- 1:19:19labs will want to build applications that are fairly
- 1:19:22broad, like broad and maybe like thin because they have such a
- 1:19:28vast surface area that they want to build things that a ton of
- 1:19:30people can use. So coding obviously is, is one
- 1:19:34of those where it's, it's mostly like a, you know, like an app
- 1:19:39that people can just pick up. If you think about our space,
- 1:19:41it's quite different, for better or worse.
- 1:19:43We have a very involved like post sale motion where, you
- 1:19:47know, we talked about our conversation designers, but
- 1:19:49they're, they're part of like a much broader team that we, we
- 1:19:52use to actually work with our clients and, you know, get their
- 1:19:55agents stand up, stood up and like help write these agent
- 1:19:57operating procedures because, you know, we have the benefit of
- 1:20:00having the expertise. And because you have that post
- 1:20:03sale motion, it's, I think it's pretty unlikely that's there's
- 1:20:06going to be an app that replaces it.
- 1:20:10But over time, I think the application layer is like so
- 1:20:12thick that there's a lot, a lot of stuff to build.
- 1:20:14So I, I actually, I think there's going to be a pretty
- 1:20:19bright future for most application companies.
- 1:20:22What just as somebody sort of so close to the space and how
- 1:20:25things are developing, like how do you think the world looks
- 1:20:28differently in in five years? Specific to our space.
- 1:20:31Or just broadly, like, what? What do you think?
- 1:20:33Like, I don't know. Yeah.
- 1:20:34For the regular person just watching this, trying to
- 1:20:37understand how AI is going to change their lives, like what do
- 1:20:39you think feels the most different in five years because
- 1:20:42of what people are building today?
- 1:20:44So a couple themes. So the first one in our space,
- 1:20:47we, we kind of touched on this, but I think the way that
- 1:20:50consumers are going to interact with any brand is going to be
- 1:20:53fundamentally different. And you know, hopefully we are
- 1:20:55again a major player in that. But if you think about the
- 1:20:57previous shifts, right, like last one, let's say from
- 1:21:02Internet to mobile, right, that that basically created like
- 1:21:04entirely new UI for people to interact with, you know, the
- 1:21:09brands that they need to interact with, whether to buy
- 1:21:11things or to get something done, etcetera.
- 1:21:14And AI is, is you can almost think of it as like a new UI.
- 1:21:18And this UI is conversational. And so you can talk to it on the
- 1:21:22phone, you can chat with it. But that is, that is something
- 1:21:25fundamentally that's gonna change and you're gonna have.
- 1:21:28But don't people like to push back on that?
- 1:21:30Some people like shopping, you know, like some of some of these
- 1:21:33cases, it's like where people talk about like travel and it's
- 1:21:36like people like spending time booking travel in some of these
- 1:21:39cases. Like, do you really think it's
- 1:21:41gonna be like, hey, like text in go figure out my travel, go
- 1:21:45figure out what? I don't think it's necessarily
- 1:21:47like a clean replacement, right? Like going from web to mobile.
- 1:21:49Some people still, if I'm on my computer, I'll just use the
- 1:21:51website and people go to stores. So, right.
- 1:21:53But it's more of like a creation of a new medium.
- 1:21:56Yep. And yeah, there will be
- 1:21:57situations where I would rather talk to the YouTube.
- 1:21:59There might be other situations where I'd rather use the mobile
- 1:22:01app. And so you have this like new UI
- 1:22:02that's created. And this UI is a lot more
- 1:22:05friendly for consumer agents because consumer agents can also
- 1:22:08go to your website and try to click around.
- 1:22:10But that's like a very crude approximation.
- 1:22:13But agents, again, are very good at conversation.
- 1:22:14And so the two agents are talking.
- 1:22:15You just like back to my original point, you're basically
- 1:22:19just massively increasing the number of healthy interactions
- 1:22:23between consumers and brands. Whereas right now a lot of them
- 1:22:26I would describe as like unhealthy or there's not even
- 1:22:29happening because I can't be bothered to do them right.
- 1:22:31So that's one big thing that will change.
- 1:22:34I think this is an argument that's like if you hate talking
- 1:22:36to AI customer support, don't worry, I know you're trying to
- 1:22:39be better than the loops that everybody got.
- 1:22:41And so in a lot of cases they are happy to talk to you.
- 1:22:43But there is also the further pitch, which is soon enough you
- 1:22:46can have somebody talk to us and then get the readout of like,
- 1:22:50what transpired. And you don't actually have to
- 1:22:51do it yourself, which I think people will be excited about.
- 1:22:54OK. So that's one, yeah.
- 1:22:55That's one that's basically our space, right?
- 1:22:57So we're really excited about that.
- 1:22:58And I think that's why the market is so large and why
- 1:23:01there's so much interesting work to be done in our space more
- 1:23:05broadly. I think something that will
- 1:23:07happen a lot more is right now people think of AI as it's like
- 1:23:13a tool there. It's kind of like Google, right?
- 1:23:15You go to Google and you like search something and like, OK, I
- 1:23:17go to my, I go to AI and like, I get something done.
- 1:23:20But I think in, in three to five years, it'll be much more normal
- 1:23:24to just have like AI running continuously and it's just like
- 1:23:27always running on whatever. And it's, it's kind of happening
- 1:23:30in the background, right? Because right now it's kind of
- 1:23:32like you ping it and it does something.
- 1:23:34But the sort of the, the standard practice that we're
- 1:23:38moving towards is just like long running autonomous things.
- 1:23:41And that's, that's cool for a bunch of reasons.
- 1:23:44One, you can just accomplish like way greater tasks.
- 1:23:48And two, you can really increase leverage because like you don't
- 1:23:50have to be involved, like it's just doing things.
- 1:23:53And so this is a, this is a big trend in the coding space.
- 1:23:55If you're following those, like a lot of what?
- 1:23:58You start to sleep, wake up and see what's happened overnight.
- 1:24:00Exactly. Yeah.
- 1:24:01Or what started with like it's, it's kind of tab auto completing
- 1:24:05things Now it's, you know, it moved to OK, you, you give it
- 1:24:08some prompt and it can like write whole things and you
- 1:24:10review it. And then now it's moving towards
- 1:24:12like, OK, it's just like, hey, let me give you some
- 1:24:15instructions. And then like you just kick it
- 1:24:16off and it's like a colleague doing work, right?
- 1:24:19That's the same concept again behind Duet where we've kind of
- 1:24:22created this, this new concept of like a long running agent in
- 1:24:24the background. And I think that will happen
- 1:24:26with consumer agents as well, where you just give it like long
- 1:24:28tasks or it's constantly listening to you and it's just
- 1:24:32like a always there like helper. Great, Jesse, thank you so much
- 1:24:37for coming on the Newcomer Podcast.
- 1:24:38Cool, Eric, thanks for having me and.
- 1:24:39Congratulations for being on the list.
- 1:24:41Thank you. All right, sweet.
- 1:24:42Thank you. Thanks for sticking around to
- 1:24:44the end. Please.
- 1:24:45If you've made it this far, you've got a like comment,
- 1:24:47subscribe. Excited to grow the channel.
- 1:24:50And of course, you can find our writing and reporting on sharps
- 1:24:53and venture capital at newcomer.co.
- 1:24:57We also host events. You can check out what we're
- 1:24:59doing at Newcomer dot events, share other podcasts, help us
- 1:25:03get distribution, support the channel comment support, like
- 1:25:06comment subscribe. You know, you know the deal.
- 1:25:09Thanks. Thanks for being on the journey
- 1:25:10with us. All right, see you next week.