Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:05
Welcome to another weekend bonus episode of the Tech
0:07
Meme Ride Home. I'm Brian McCullough,
0:09
as always. This is a portfolio profile
0:12
episode. And Chris
0:14
is joining us for this one, Chris. Hello,
0:17
hello. The reason being
0:19
because we're going to talk about an investment
0:22
that the Ride Home AI Fund made, which obviously
0:25
the Ride Home Fund made as well. But
0:27
this is one of our AI companies from
0:29
this year. We have with
0:32
us Scott Kvitin, the
0:35
founder of Case Mark AI, which you can
0:37
find out more at as you're listening at
0:40
casemark.ai. Scott, thanks for
0:42
coming on. Thanks for having
0:44
me, Brian. Good to see you both. Good to see
0:46
you, Chris. Yes. So
0:50
let's, let's right off the top. Case
0:53
Mark, as maybe the name implies, is AI in
0:57
the legal space. So give
0:59
us just sort of like the elevator pitch
1:01
for what Case Mark does, Scott. Yeah,
1:04
I think, I think the sort of
1:06
shortest version of this is we've created
1:08
kind of the easy button for attorneys
1:10
for specific discrete tasks that they'd like
1:12
to accomplish with AI. But we
1:15
do it with sort of security and privacy
1:17
in mind. In other words, we deal with
1:20
the implications around data leakage, training, making sure
1:22
that this is safe for
1:24
them to use. But then by boiling
1:26
that down to really simple tasks
1:29
that they try to accomplish, specifically what
1:31
we focused on early on were things
1:34
like deposition summaries, case summaries, trial
1:36
and hearing transcript summaries. And
1:38
that's kind of the early phase of this. But
1:41
what we're realizing is that over time, attorneys are
1:43
realizing, wow, there's actually more you can do with
1:45
AI. And once they see the simple use cases, they
1:47
realize there's a bunch of ways that they can essentially
1:50
take all this unstructured data that they
1:52
have and then leverage AI to transform
1:55
it. Into, you know, interesting structured analysis
1:57
and reporting for what they're trying to
1:59
do. to accomplish. So
2:02
you're talking about things like case
2:04
summaries, like depositions, as you say,
2:06
hearing summaries, contract
2:08
review, discovery response. One
2:12
of the sort of narratives for how
2:14
at the very beginning, in this first
2:17
inning of this era of the AI
2:19
revolution, it's about getting
2:22
rid of the busy work, or at least
2:24
obviating the busy work of
2:26
freeing folks up
2:28
to not have to
2:30
spend so many hours on stuff that allow
2:34
you to focus on other more important things. Is
2:37
that sort of the focus right now, what
2:40
I just described? Let's get rid of spending hours
2:42
going over depositions and things like that? Yeah.
2:45
Let's get rid of the TDM work. It's
2:48
easy to lose the details or miss
2:51
something because you're tired and you're on
2:53
the fifth hour of going through what
2:55
might be a boring deposition. Humans
2:59
are fallible in that sense. Of course, the flip
3:01
side response then we get from some attorneys is,
3:04
well, wait a minute, aren't you going to take
3:06
away my opportunity for a bunch of billables? Our
3:10
response is, well, no. One,
3:12
you still should review what we're producing. But
3:15
a lot of attorneys also have some
3:17
artificial limits in place, whether it's on
3:20
the defense side they can only bill so much
3:22
for what they get from these summaries. It might
3:24
take them six hours to do a summary, but
3:26
they can only bill two. They
3:28
lose that four hours. Or on the plaintiff side,
3:30
they just want to be able to do
3:32
it as inexpensively as possible because they're doing
3:35
it on a contingency basis. Any costs they
3:37
accrue then count against any potential future
3:39
settlements. The other thing that
3:41
I think a lot of folks forget about with respect
3:43
to AI right now is that
3:45
they talk about this doom and gloom around
3:48
how it's going to replace all these paralegals
3:50
and associates. If we were at 100% AI
3:52
usage today right now, I would agree with
3:54
that. However, what people forget
3:56
is that AI is actually going to enable,
3:58
it's going to make get really, really easy
4:00
for law firms to litigate. So we're about
4:03
to go into a hyper growth in terms
4:05
of litigation that's going to happen. So I
4:07
think we're going to see a 5 to
4:09
10x increase in litigation, which means there's
4:11
going to be a demand for not only
4:13
the AI solutions, but you're going to need
4:15
humans to sort of arbitrate and traffic control
4:17
on all this. So what we're doing is
4:20
removing the burdensome, time consuming, tedious
4:22
tasks and allowing attorneys to actually use their
4:24
critical thinking. And we actually think this is
4:26
going to increase their job satisfaction for
4:29
sure. Insert lawyer joke
4:31
here. I don't know if the phrase,
4:34
an explosion in litigation is what some people want
4:36
to hear. But I guess. I'm looking
4:38
forward to that. Yeah. Well, listen. That sounds like
4:40
a party. If you're a lawyer,
4:42
yes, I can see that as well. So
4:44
sue me. Says no one ever. Yeah. Yeah.
4:46
OK. Let's imagine that I'm
4:50
a law firm listening to
4:52
this right now. What does it
4:54
entail for me to start
4:56
using your tool? Does
4:59
it plug into my existing workflows easily? What
5:03
does it take to get running with K-SMARK? Yeah.
5:05
We tried to keep it as simple as possible.
5:08
And we also tried to solve the
5:10
problem in the simplest fashion as
5:13
we could. So you can essentially sign in,
5:16
log in with your email address, your
5:19
law firm's address, whatever it is. You
5:21
upload a file. You choose the workflow you want to run.
5:23
There's a couple options in there if you want to choose.
5:25
Otherwise, you can just do the defaults and click Go. And
5:27
then a couple of minutes later, you're going to have some
5:30
results that you can then download
5:32
and use. And I think what's interesting here,
5:34
too, is the work
5:36
product that we generate is a Word document
5:38
or a PDF. And while that might be
5:41
boring to us as technologists for
5:43
attorneys and law firms, their programming
5:46
language or software of choice happens to be
5:48
a Word document or a PDF. And
5:51
so when we can actually download that
5:53
PDF of a summary of, say, a
5:55
deposition, and then we can actually, what
5:57
we do is we append the transcript.
6:00
the source transcript is appended there. That's
6:02
actually a really powerful encapsulated litigation tool
6:04
that they can then drop into their
6:07
case management solution or
6:09
they can forward it onto their insurance adjuster. And
6:11
it doesn't, we're not completely replacing how they do
6:14
their work. In other words, they can
6:16
jump in, use our tool and then drop it into the
6:18
way that they do things. And
6:20
attorneys really like that because it's a way
6:22
for them to kind of try these things.
6:24
So often I see these really beautiful products
6:26
that are AI powered, but they require that
6:29
the attorneys completely change how they do business.
6:31
And if you think about a law firm that has a hundred
6:34
people, it maybe has 25 attorneys
6:36
and then 75 supporting staff. If
6:39
they all have to then learn some piece of
6:41
software and then that software is continuously
6:43
changing and that will
6:46
completely screw up their daily workflow such
6:48
that if a law firm, if
6:50
all the paralegals lose an hour because
6:53
of some new software, guess what, that's
6:55
75 hours of billable time that you've
6:57
lost and that's a big, big deal.
7:00
And so we try to make our tools as
7:02
simple and as easy as possible and deliver a
7:05
solution that basically mimics what they do by hand
7:07
today. And then in the future, we'll
7:09
evolve these tools to be sort of more advanced and
7:12
sort of ingrain them in their, more of
7:14
their daily process. But again, try to be
7:16
as lightweight as possible today. That
7:19
seems to have the biggest impact for us. I
7:21
think the other piece too is by
7:23
being a lean and mean startup and
7:26
because you're looking at the entire sales and marketing team,
7:29
we don't have time to be able to like follow
7:31
up and reach out to a lot of these folks.
7:33
So we do have a self-service model which
7:36
actually really helps these folks because maybe
7:38
it's the paralegal, the law
7:40
firm administrator, the overworked office
7:42
manager who's in charge of this
7:44
and they can jump into our solution, they can
7:47
try it, they can break it without having to
7:49
talk to a salesperson because they're always worried about
7:51
asking a question that they think might be considered
7:53
stupid which obviously we would never say that and
7:56
we don't think that. But the
7:59
legal industry is full. of people who
8:01
sort of shout down at the people
8:03
who are the paralegals and associates by
8:06
design. It's a weird industry. I can say
8:08
this because my wife's an attorney, so I
8:10
consider myself attorney adjacent. Just
8:13
to hear some of the stories of her when
8:15
she was a younger attorney, she's now been doing
8:17
this for almost 15 years, but
8:20
just to hear how they just get sort of brow beat
8:22
all the time. It's just astounding.
8:25
Anyways. Last one, and then
8:27
I'll let Chris give you some questions too.
8:29
But obviously here, one
8:31
of the concerns would be, I
8:34
can't have these sensitive documents fall
8:37
into the wrong hands, leak, be trained
8:40
on. So how
8:42
are you thinking of and dealing with
8:45
things like privacy, security, stuff
8:47
like that for everyone
8:49
that their business
8:51
is business critical, but you're dealing with legal
8:53
stuff and you could blow up cases if
8:55
you do it wrong. So what's
8:58
your process there? Yeah, for sure. So
9:01
we have relationships with
9:03
Amazon, Microsoft, and Google
9:06
to actually have sort of private cloud infrastructure
9:08
for all of the data that we ingest.
9:11
And so the best way to describe
9:13
it is when we ingest private data from our
9:15
customers, we basically drop it into
9:17
a container along with the LLM of choice that
9:19
we're using. We then do our transformations
9:21
on it and then output some summary, and then we
9:24
tear that whole thing down and it goes away. In
9:26
other words, we don't actually use models that
9:29
are, like we don't connect to OpenAI's
9:32
APIs. We do
9:34
everything inside of Azure with respect to OpenAI,
9:36
but we also are using Gemini and Claude,
9:39
a little bit of llama, some Mistral in
9:41
there. But again, that sort of same theme runs
9:44
true that we're not going to train with our
9:46
data. And we lean really hard into that. And
9:50
we've gone through quite a few, not only sort of
9:52
those third party risk assessments that you have
9:54
to do for some of the larger firms,
9:57
but also having the CIO offices or their
9:59
security teams coming. in and really kicking the
10:01
tires, getting into our source code even to see
10:04
and verify, trust and then verify on
10:06
these things. And so I think that's
10:08
a really critical piece there. And
10:10
I don't know if it's verboten or not, but
10:13
if I can share my screen, I can kind
10:15
of show something really quick if that's OK.
10:17
Sure. Sure. That's sweet. Knock yourself out.
10:20
I mean, if you're listening on the podcast, this
10:22
isn't going to knock your
10:24
doom and spin. Oh, yeah. Yeah. I will
10:27
try to narrate, yes. Yeah. And
10:29
so what I've got here is basically kind
10:31
of a high level architecture diagram of how
10:33
our system works. And kind of at
10:35
the top of this are a series of workflows. And
10:39
the things of that would be like a
10:42
deposition summary, a trial and hearing transfer summary,
10:44
a case summary. And that sits on top
10:46
of what we call our sort of workflow
10:48
engine. And what the workflow engine does is
10:51
it securely ingests this data from our customers
10:53
and then puts it in a variety of different
10:55
places, whether it's
10:57
an elastic search or a Bragg database
11:00
or a vector database, such
11:02
that when we ask the questions that
11:04
are done by these workflows, which are
11:06
essentially sophisticated prompts, prompting chains that we
11:08
do, the workflow engine knows where to retrieve
11:10
that data, how to do it, how to verify that it's
11:12
the right data, all those kinds of things. And then the
11:15
second piece is that sits on top of what
11:17
we call our LLM routing engine. And
11:19
early on, we knew that we kind of wanted
11:21
to be the Switzerland of providers in this space.
11:23
In other words, we want to be cloud
11:26
LLM and then in the very near
11:28
future region agnostic. And
11:30
the reason that's important is our customers don't
11:33
know or care what models we're using. What
11:36
they're doing is signing a deal with us that
11:38
we're going to deliver what they need at a
11:40
decent price performance, that it's going to be accurate,
11:43
secure, private, all those things. And then what
11:45
we do is our LLM routing engine allows
11:47
us to essentially do that across a range
11:49
of different providers. So we'll never
11:51
be holding to any one provider. And like
11:54
we said earlier, we're in the first inning of this
11:56
thing, and we don't
11:58
know who the winner is going to be. And there might
12:00
not be a winner. So this ability to be able
12:02
to pick and choose the different models for specific things
12:04
that they're really good at is absolutely critical. The
12:07
last thing I'll mention, and then I'll sort of stop
12:09
here, but is that all of
12:11
these LLM models have something called
12:13
content filtering built into them. And that
12:16
content filtering is designed as a CYA
12:18
mechanism to make sure that the
12:20
big players are protected against or
12:22
protecting sort of the general public from doing
12:24
untoward things with their models, like making meth
12:26
or building bombs, like that kind of stuff.
12:29
But when it comes to law firms, they
12:31
actually have to deal with some really sensitive
12:33
topics, things like hate
12:36
speech or law-making mesh. Or
12:39
making meth. And we
12:41
actually even had one of our, we just
12:43
posted this story this week, but Lisa Peck
12:45
is a civil rights and employment attorney in
12:48
Northern California. And they just won a $20
12:50
million verdict. And it was an
12:53
employment law that had a bunch
12:55
of racial undertones in it against
12:57
Stanford Health. And it
12:59
was actually really, really powerful the way that
13:01
it worked. But in
13:03
any case, it's exciting
13:06
to see how these solutions can solve
13:08
these problems. And so we've
13:10
been able to figure out how to deal with that content
13:12
filtering in such a way that we deliver the right result
13:14
to folks, even if it is
13:16
sensitive content. And again, those are the kinds of
13:18
things that I think a lot of folks aren't really thinking about quite
13:21
yet. So yeah. So just to jump in
13:23
there, I just want to understand a little bit more
13:25
about the blog post that you previewed there and what
13:27
K-SPARK's participation or involvement was. But I think to make
13:29
the point, I think what you're
13:31
raising is super interesting because there is
13:33
a great deal of conversation about alignment
13:35
and making sure that these things, whether
13:38
it's less so about concerns
13:40
about hallucinations and more about saying, like you
13:43
said, reasonable things and not insulting someone or
13:45
saying stuff that they're not supposed to. Any
13:47
number of times you might have asked ChatTPT for
13:49
instructions to do something that is borderline
13:53
or could result in borderline content, and then it shuts you down.
13:56
So given that part
13:58
and parcel to lawsuits. and
18:00
how this is gonna change litigation,
18:02
especially in real time
18:04
trials are gonna change, I think
18:06
significantly here, so. This
18:09
message is sponsored by Greenlight.
18:12
As your kids get older, some things about parenting
18:14
get easier. You know, they can
18:16
dress themselves, lug their own suitcases
18:19
through the airport, but
18:21
other things don't get easier, like having that
18:23
conversation with your kids about money. The fact
18:25
is, kids won't really know how to manage
18:27
their money until they're actually in charge of
18:29
it. That's where Greenlight can
18:32
help. Greenlight is a debit card and
18:34
money app made for families. Parents
18:36
can send money to their kids and keep an eye
18:38
on kids spending and saving, while kids
18:41
and teens build money confidence and
18:43
lifelong financial literacy skills. With
18:46
the Greenlight app, kids learn how to save,
18:48
invest, and spend wisely thanks to games that
18:50
teach money skills in a fun, accessible way.
18:53
The Greenlight app also includes a
18:55
chores feature where you can set
18:57
up one-time or recurring chores, customize
18:59
to your family's needs, and reward
19:01
kids with allowance for a job
19:03
well done. It's a far
19:05
cry from me getting my allowance in quarters
19:07
for every time I emptied the dishwasher. Millions
19:10
of parents and kids are learning about
19:12
money on Greenlight. It's the easy, convenient
19:15
way for parents to raise financially smart
19:17
kids and families to navigate life together.
19:19
Sign up for Greenlight today and get
19:21
your first month free when
19:23
you go to greenlight.com/ride.
19:26
That's greenlight.com/ride to try
19:29
Greenlight for free, greenlight.com/ride.
19:35
If you are building a SaaS business
19:38
achieving compliance with SOC 2, ISO
19:41
27001 or other in-demand frameworks can
19:44
unlock major growth for your company and
19:46
establish customer trust. However, this
19:49
process is often time intensive and costly.
19:51
Vanta automates your compliance work, getting
19:54
you audit ready quickly and saving you up
19:56
to 85% of associated costs. Vanta
19:59
scales. that
24:00
happen annually and it's a huge, huge business.
24:03
It's a huge market, but these attorneys
24:05
to take a deposition as an attorney is
24:07
actually a process. It takes you several years
24:10
to get good at it, especially a contentious
24:12
deposition. And so one of the things
24:14
that we've seen is folks who can take real
24:16
time depositions and then transcribe those. And
24:18
let's say before you go into the deposition, you
24:20
know there's 15 questions you wanna have asked and
24:22
answered and the way the depositions work is you
24:24
get one bite at the apple. You get to
24:26
ask those questions and then that's it. You don't
24:29
get to go ask later. So if
24:31
you're an attorney that's a junior attorney and it's
24:33
really contentious and you sort of lose your sense
24:37
and you didn't get those 15 questions asked
24:39
and answered, guess what, you're kind of hosed.
24:41
So imagine if you start at 9 a.m.
24:43
it goes till noon and you have
24:45
a lunch break and you could feed in that transcript
24:47
and then you could compare it to the 15 questions
24:49
and the AI could come back and say, oh great,
24:51
you're nine of the 15. So you got
24:53
six more that you need to get answered after
24:55
lunch and so now all of a sudden
24:58
we can up level attorneys that wouldn't normally
25:00
have the skills to be able to do these things.
25:02
And to me that's very exciting. Not only is it
25:04
a job satisfaction thing but it's also helps them kind
25:06
of up level to a certain extent. And
25:09
more importantly, yeah, sorry, go ahead. Well,
25:12
I was gonna say just like one more on that, I guess that topic,
25:15
which is around, I guess like talent and how
25:17
the field itself is changing. One thing that I
25:19
think Brian and I have been witnessing and observing
25:22
as we've been talking to other founders going into
25:24
other spaces, of course, where AI is starting to
25:26
be deployed is just hiring is
25:29
like sort of, I mean, I suppose
25:31
this is like the meme and the reality going
25:33
around, but that people can't hire enough staff,
25:35
talents, they can't keep them, they can't retain them,
25:38
whether it's because now there's a whole new world of influencers
25:40
or whatever it is, it seems
25:42
as though like you said about the court reporters that
25:44
there's going to be deficits in
25:47
certain roles. And so the need for
25:49
AI to actually perform those duties is
25:51
becoming more and more significant. And
25:54
I guess the other thing that I would add to that is that you
25:56
also have in some cases, older
25:59
or more, I
30:00
mean, we probably have one of the original
30:02
relationships. Yes. Yes,
30:04
exactly. Back when you were at the Oregon
30:06
State University running the computer lab, the open
30:09
source lab. Thank you. Back in 2004.
30:13
So, you know, we, we go back 20 years and, you know, we've seen each
30:17
other through a number of different eras and
30:19
waves of technological development
30:23
from open source into, you know, I was a advisor
30:25
to urban airship, which is now just airship, I believe.
30:29
And essentially is focused on the
30:31
productization of push notifications once, you know,
30:33
the iPhone came out and provided a
30:35
solution to replace SMS. The
30:38
reason why that I think is relevant is
30:40
because, you know, Scott and I have been
30:42
in these moments of transformation
30:44
of, you know, basic set
30:47
of, you know, behaviors
30:49
and norms where, you know, to, to sort of
30:51
invoke McLuhan, the idea is to take an existing
30:53
set of, you know, media content and move it
30:55
into a new one.
30:57
And it's just more efficient. And so in a similar
30:59
way, you've got these depositions and it's going to be
31:01
more efficient to process them. And yet the transformation doesn't
31:04
actually happen until you're a few years in. And because
31:07
you've built on the existing
31:09
substrate or set of behaviors that exist, you
31:12
can then start to make subtle tweaks and
31:14
changes to process. So
31:16
my question to you is, you know, as
31:18
you're seeing this, you have this interesting challenge
31:21
to, on the one hand, deliver a product and
31:23
a service and a tool that meets the
31:26
legal profession where it is currently, you know, like
31:28
you said, you're not going to replace the whole case management system, you
31:31
know, one and done. But
31:34
over time you get to redefine
31:36
the way that that work actually is executed.
31:39
So when you think about, can you play out case
31:41
mark over the next several years, you know, one,
31:44
how do you see this moment, you know,
31:46
with AI being different from past technological
31:49
revolutions in terms of you building product?
31:51
And then two, how do you
31:53
see the industry, the
31:55
legal profession sort of
31:58
changing, assuming you're successful? Yeah,
32:01
I mean,
32:03
history doesn't repeat itself, but boy
32:05
does it rhyme. I'll
32:08
tell you what, I just feel like we're going through
32:10
what we saw with the original dot
32:14
com bubble, web 2.0 that you and
32:16
I lived through, Chris, mobile 2.0. To
32:19
me, what's happening is a lot like what happened with
32:21
the iPhone with respect to AI. What
32:24
I mean there is, AI is having what I would
32:26
call an iPhone moment. We
32:28
all had mobile phones in our pockets when the iPhone
32:30
came out, but when we saw the iPhone for the
32:32
first time, we said, oh, this is
32:35
what a mobile device is supposed to be. While
32:38
there were people who had 20 years of experience
32:40
with mobile when the iPhone came out, that was
32:42
all out the window. AI has
32:44
been around for 40 years, but when we all
32:47
saw chat GPT, we said, oh, this is what
32:49
AI is supposed to be. What
32:51
that does, just like with the iPhone, is it creates
32:53
a moment in time where people are saying, oh, my
32:55
gosh, I have to have that. I have to have
32:58
that. That creates an opening for a
33:00
company like ours to create that as a wedge. We
33:02
have an answer for you for your AI solutions because
33:05
that's what we've said we can do.
33:08
Do we have a specific product or feature or point
33:10
solution right now? Yes, we do,
33:12
but really, the way to win in this
33:15
to me is to build a
33:17
partnership with customers over time because you're navigating
33:19
what is a disruptive cycle that will take
33:21
a decade to accomplish. This is exactly what
33:23
we saw with Urban Airship. It's
33:27
been very, very interesting to see
33:29
that as it relates to
33:31
AI and this moment in
33:33
time with legal and, anyways,
33:36
to watch how this is playing out. Again,
33:38
the other thing that's also very interesting is there's
33:41
a lot of companies raising a lot of money
33:43
at valuations that are ridiculous and they don't have
33:45
product market fit and they probably won't get product
33:48
market fit. Me
33:50
having gone through this a gazillion times, I'm
33:52
shouting at the top of my lungs like,
33:54
oh my God, we're really doing this again?
33:56
But again, that's how these cycles go and
33:58
that's okay. But when
34:00
we think about where we wanna be or
34:02
where we think this is gonna go is, I
34:05
see a lot of people saying like, we're gonna be
34:07
this AI assistant for law firms and
34:09
you can ask it questions and it's just gonna do things. And
34:12
what you end up usually with is this sort
34:14
of mile wide and inch deep solution that doesn't
34:16
really solve specific problems. And so what we said
34:18
is, wait a minute, what if we can solve
34:20
from the bottom up? In other words, we can
34:22
solve these specific discrete tasks really, really well and
34:24
we'll get them bulletproof because we're gonna throw thousands
34:26
and thousands of tries at it and we're gonna
34:28
use it and all these people are gonna use
34:30
it. Such that when we do layer
34:32
on that assistant down the road, we
34:34
can then look at a case folder and say, okay,
34:36
well we see some pleadings in there, some transcripts, some
34:39
medical records. Here's our suggestion. We think you should do
34:41
some deposition summaries and medical chronology. And then by the
34:43
way, we'll sum the whole thing up with a
34:45
case summary report that you could forward on to somebody.
34:48
And oh, by the way, we have a couple of
34:50
next best actions you should take because there's a couple
34:52
of filings that are due in here and you should
34:54
check those out. Right, so that's where I think this
34:56
is gonna evolve. I mean, that's what I'm talking about
34:59
there is five years of work, right? And
35:01
it's not gonna happen overnight and people aren't gonna adopt
35:03
or trust it overnight, but that's where we're gonna kind
35:05
of get, I think. And just
35:07
like when the first push notifications got sent,
35:09
I immediately saw, I was like, oh wow.
35:12
And it was companies like Starbucks that said, okay,
35:15
yeah, cool, we have a mobile app and in our
35:17
mobile app, we have our menu. But
35:19
what they saw was, oh wow, this is
35:22
gonna change how we do ordering, how we
35:24
do tipping, how we do stored value. Like
35:26
all those things that have a fundamental change
35:29
to your business that manifest themselves in an
35:31
app, but really mean you have to change
35:34
how your store works and how, like all
35:36
those things. So that's the transformation that Legal's
35:38
about to go through right now and it's gonna be
35:40
painful and it's gonna be hard, but
35:42
those who navigate it and find the right partners to
35:44
help drive that are gonna come out the other side
35:46
way, way stronger and
35:49
way, way more profitable in my opinion, so. You
35:54
mentioned that you're attorney adjacent and you've
35:56
gone into a little bit of your
35:58
background, but can you give me. Any
36:00
sort of the inception of the
36:02
idea of this company, maybe touching
36:04
on where you were when
36:08
you started working on this company and
36:10
where the light bulb moment came from? Yeah,
36:13
I mean, I think so just
36:15
on me, I mean, I'm a serial entrepreneur.
36:17
I cut my teeth at Amazon turn of
36:20
the century. I have the dubious honor of
36:22
being on their Y2K team for
36:24
what was basically the biggest nothing burger
36:26
ever. Was really active in
36:28
a bunch of open source, open technology stuff. Just
36:31
like Chris said, he and I were there when Mozilla
36:33
spun out of AOL and we
36:35
both helped kind of get Firefox 1.0 out
36:37
the door. Then again,
36:39
we partnered up on making sure that
36:41
OpenID and OAuth and pulled Facebook, Microsoft
36:44
and Google together to say, this is it.
36:46
And then we created the OpenID Foundation, which
36:49
I think I was the chairman for a little while and then we
36:51
let that off into the world and that's created a really awesome thing
36:53
there. I started a company called Urban Airship to
36:56
do push notifications. I really moved over to the
36:58
business side to sort of build, scale and sell
37:00
B2B enterprise SaaS companies. After
37:02
that, my co-founder and CTO, who I still
37:04
work with this to this day, Steven Osborne, we
37:07
started a point of sale system for the cannabis industry that we sold
37:09
in 2017. Most
37:12
recently, we sold a transparent ledger for
37:14
physical assets company. So it
37:16
was a blockchain company, except that instead of saying, if
37:18
you build it, they will come. We
37:20
actually had a customer who had sports memorabilia that
37:22
we were doing anyways. We sold that last year,
37:24
ended up being an exit out to Fanatics because
37:27
our customer got acquired. So we got kind of
37:29
swept up into that. And
37:32
so I had a team and this
37:34
is literally June 1st of 2023. So
37:37
just over a year ago, and we
37:39
were kind of looking at what we're going to do. And
37:41
my wife, obviously being an attorney, she runs a firm
37:43
here in Portland. They do Oregon, Washington, Idaho,
37:46
insurance defense. So
37:48
they have big retailers and
37:52
let's just say, driving or vehicle
37:54
related businesses, they support all
37:56
those. And so she kind of
37:58
jokingly said, hey, why don't you do something? something that helps me for
38:00
once. Like a lot of attorneys, she
38:03
had the first experience which was, like I do when
38:05
I'm launching a new company, the first thing I want
38:07
to do is get product into market so we can
38:09
increase our pace of learning, so we can figure out
38:11
what the heck's here. We immediately
38:13
said, let's launch a Word and Chrome extension,
38:16
a Word add-in and a Chrome extension. So
38:18
we found the open source solutions out there,
38:20
we hired the devs, and then
38:22
we launched something within a month, basically about four
38:24
weeks. We learned really quickly that
38:26
attorneys don't want those. They forget
38:29
about the add-in, they hate paying for software
38:31
that they might not use that month. The idea of
38:33
SaaS to them is just, they just
38:35
don't understand it because they bill for the time that they
38:37
work, so they just don't understand it. In
38:40
the fall, we tried some fine-tuning of models
38:42
for firms, but then we
38:45
were left with this whole concept of,
38:47
okay, cool, we fine-tuned your model. Here's
38:49
a prompt window, go ahead and start
38:51
use AI, and they're like,
38:53
what the hell do I do with this? That
38:55
let us down launching what we call legalpromptguide.com,
38:57
which is a really simple solution. It's a
38:59
freebie site there, it helps attorneys figure out
39:02
how they're going to do prompting, and
39:04
really understand, just like I took a generation for
39:06
folks to figure out how to get those Google
39:08
searches right to extract what you want out of
39:10
Google, the same thing is true like 10x for
39:12
chat, GPT or
39:15
just chatting with any generative model.
39:18
That's when we realized, oh, we have to take what
39:21
we learned from all these prompting
39:23
and turn those into easy buttons,
39:26
leveraging these methodologies and the things that we've learned to
39:28
be able to get the most out of the LLMs.
39:31
That's when we really launched in
39:33
earnest in January, and then we had a bunch
39:35
of legal tech players approach us saying, hey, we
39:37
want to license your stuff, which
39:39
then led to, oh, we need an API and
39:41
now we've got folks connecting to the API. It's
39:44
literally just taken on a life of its own
39:46
now, and now we're stamping
39:48
out all kinds of workflows for
39:51
folks to be able to really increase our scope,
39:53
and then anybody who's plumbed up to our API,
39:55
guess what? They get to have access to any
39:57
of the workflows that we have. has
40:00
this multiplicative effect right now.
40:02
And we're kind of have landed on this, you know,
40:04
AI as infrastructure play. And
40:07
then, you know, from a pricing standpoint, we've tried to
40:09
be really aggressive. Instead of saying, here's
40:11
how much it costs for an attorney to do it,
40:15
we're going to charge just a little bit less, we actually
40:17
are going the other way, which is we know what our
40:19
cogs are, and we're going to tack on a margin that
40:21
leaves a lot of room for people to resell our stuff.
40:23
And that's working really, really well right now. And
40:26
so yeah, that's kind of the hook for
40:28
why we landed on it. And then my
40:30
wife's been really instrumental in helping
40:32
us kind of craft the some of the
40:34
initial workflows that we did. Because
40:37
the key with the LLMs is they have
40:39
the answers, you just have to
40:41
know how to ask the right question. And the
40:43
answer can't the prompt can't be pretend you're a
40:45
lawyer. What you have to do is
40:47
ask a question like a lawyer does. And then
40:50
it will respond with a response that a lawyer would
40:52
expect to see. And so those are the
40:54
tricky things that we don't have the experience in. And that's
40:56
why we lean in on my wife and a couple other
40:59
trusted advisors in that front. So
41:01
yeah. It's funny. I'll just comment
41:03
on a question. One
41:06
of the reasons why we invested was along the lines of
41:08
the AI varietals thesis that Brian
41:11
and I work from. And that essentially is
41:13
this concept of bringing together
41:15
some subject matter expert with
41:18
AI engineering or the
41:20
use of generative AI. And the fact that
41:22
obviously your wife is the subject matter expert
41:24
in this case sort of allows you to
41:28
ground in real truth the
41:30
way in which you're bringing generative AI
41:33
into a specific context where there are
41:35
specific requirements. And those requirements come from
41:37
language. They come from a set of
41:40
expectations that people in the field, you
41:43
lose so much credibility if you answer the question in any
41:45
other way that is not sort of legalese. And
41:47
so it's not enough to,
41:49
I think, like you said, like pretend that
41:52
you're a lawyer and then create a prompt.
41:54
There's a whole lot more that goes into
41:56
that, which is around culture and norms and
41:58
why communication happens to certain people. way. And
42:00
I think it's valuable just to
42:02
keep that in mind from a product design perspective is
42:04
that when you're designing something in the generative AI space,
42:06
you have to match the language of the person that
42:09
you're actually interacting with. So the
42:11
question that I have is about maybe the, if
42:16
not the metaphor, the conceptualization
42:19
of bringing the
42:23
output of an LLM into the legal context. And what
42:25
I mean by that, or what I'm getting at is
42:27
the increasing
42:30
interest in agents or having
42:32
kind of like AI employees
42:35
that work alongside someone else versus
42:37
let's say like a copilot, which you treat as
42:39
like a chat bot, which you know is an AI
42:42
and sort of sits alongside, let's say a document versus
42:44
a set of workflows that
42:47
are almost like macros, but lit up, you
42:49
know, for like the 21st century with generative
42:51
AI. And we seem to be at
42:53
this moment, you know, you mentioned AI as infrastructure, but
42:56
there's still a question of how someone
42:58
chooses to invoke these like
43:02
services. And we're, I
43:04
think in some ways struggling with the right interface for
43:06
this. So my
43:08
thought and question is kind of about that, whether
43:11
agents are relevant metaphor for you, whether
43:13
it's you'll have multiple different sort of
43:15
paralegals, but they're all powered by generative
43:17
AI. And that eventually, you
43:19
know, the junior like lawyer goes
43:21
to one that is a specific
43:23
subject matter expert, or where you
43:25
imagine these workflows are the right
43:28
concept and framework for delivering casemark. And
43:30
you're going to stick with that, because that's the one that people
43:32
seem to understand. And you're going to go forward with that. Yeah,
43:37
I think in the near term, what we're finding
43:39
is what's the easiest way that people can kind
43:41
of rock this, such that
43:44
they feel confidence, one that they can test
43:46
it, try it, get a result, and then verify
43:48
that result quickly and easily. And I think that
43:50
sort of manifests itself in the easy button today,
43:54
that idea of a changing paradigm
43:56
where you'd actually trust an assistant to
43:58
do those things or co pilot. even. I
44:01
think that's it's going to be how
44:03
it happens. It's just those things never
44:05
happen overnight. You don't adopt those paradigms
44:09
overnight. I mean, if you think about any new paradigm,
44:11
I mean, the only one that I can
44:13
think of that took hold overnight was
44:16
sort of like the pinch to zoom or like
44:18
any of the sort of touch interfaces, but you
44:20
weren't doing anything new there. You were just doing
44:22
something that was obvious that you would do. Well,
44:25
it was like a digital version of something that kind
44:27
of already existed to some degree. Yeah, exactly. And so
44:29
I think that I think that, you know, for
44:31
us, we just tried to do the lowest, easiest
44:34
way for people to get in and use it.
44:36
I think that will evolve over time,
44:38
but we always have to continuously check
44:40
in with our customers on this. The other
44:42
thing is that we've learned too, as we
44:44
sort of track the sessions and watch users
44:46
and their behavior of our solution. I'm
44:49
always astounded at how easy
44:52
we are, how simple we have to make the
44:54
interface so that they understand it because
44:57
these attorneys, you know, a lot of the times can
44:59
be sort of neophytes. And
45:01
that's a problem. That's our problem.
45:03
It's not their problem. And so
45:05
often I just keep seeing so
45:08
many companies that are building software that
45:10
I know attorneys will never use, or at least, you
45:12
know, even this current generation, the
45:15
sort of youngest generation, they'll probably be able
45:17
to make it work. But those aren't the
45:19
folks who are making the buying decisions or,
45:21
you know, have the budget to purchase, you
45:23
know, the line items that would require for these things.
45:25
So yeah, it's very,
45:27
very interesting to watch how this
45:29
whole thing is shaking out for
45:32
sure. Yeah. During
45:34
design sprints, many tools like Jira can
45:36
be restrictive in the way they organize
45:38
information. Your ability to see all tasks
45:40
is restricted by the size of your
45:42
screen, which can make it harder to
45:45
understand the relationships between tasks, especially between
45:47
different teams or epics. That's why Miro
45:49
was so useful when we were doing
45:51
the AI resume project, the entire project
45:53
could be laid out in one Miro
45:55
board. Miro helps ensure your team
45:57
has the contacts they need before devoting time
45:59
and resources. resources to get the work
46:01
done. With Miro, planning team tasks is
46:03
smoother and gives everyone a clean sense
46:05
of mission for every sprint. Plan
46:08
sprints with ease using Miro's planning widget.
46:11
Connect your team's Jira or Azure
46:13
instance to your Miro board to
46:15
visualize and filter tasks by sprint
46:17
week, status, epic, and team. That
46:19
two-way sync to Jira Azure DevOps
46:21
is key. Miro your Jira Azure
46:24
instance in Miro and any change
46:26
in tasks between sprints, processes,
46:28
or updates are instantly synced in
46:30
Jira. Whether you work in product
46:32
design, engineering, UX, agile, or marketing,
46:35
bring your team together on Miro.
46:37
Your first three Miro boards are
46:39
free when you sign up today
46:41
at miro.com. That's three free boards
46:43
at miro.com. This
46:48
podcast is sponsored by The Washington Post. If you're
46:51
a fan of the show, you're a fan of
46:53
the reporting from The Washington Post because I quote
46:55
from them all the time, as you know. Their
46:57
journalists bring you the facts and provide
46:59
clarity about what's happening in the tech
47:02
industry, revealing the role tech giants and
47:04
regulators play in our lives, the dangers
47:06
and wonders of breakthrough technological developments, and
47:08
the national conversation around artificial intelligence.
47:10
If you're trying to keep up with the ever-changing
47:12
world of AI, check out
47:14
Drew Harwell's unmatched reporting on all
47:16
things AI. Do you live for
47:18
all things tech and online culture? Give
47:20
Taylor Lawrence a read. Think the Post
47:22
only covers politics? Not true at all.
47:24
You name it, they cover it. From
47:26
climate and culture to crosswords and cooking,
47:29
The Washington Post helps you discover a
47:31
world of surprising stories, important insights, and
47:33
actionable advice. Now is the
47:35
time to sign up for The Washington
47:37
Post. Go to washingtonpost.com/ride to subscribe for
47:39
just 50 cents per week for your first
47:41
year. That's 80% off their
47:43
typical offer, so this is truly
47:45
a steal. Once again, that's washingtonpost.com
47:47
backslash ride to subscribe for just
47:50
50 cents per week for your first
47:52
year. that
48:01
a lot of the people listening to this
48:03
episode right now are listening for, if
48:06
I was in his shoes, what
48:08
would I do? Would I make
48:10
decisions this way, that way? As
48:15
I'm looking at the tech meme back end right now,
48:18
there literally seven hours ago was
48:20
another legal AI company
48:23
that announced arrays. This
48:25
is a hugely competitive space. In
48:27
fact, by the way, I got
48:29
out of that that legal tech
48:32
startups have pulled in $356 million so
48:35
far this year. Now that's down slightly
48:37
compared to last year. That's
48:40
from GeekWire, by the way. But so, OK, number
48:42
one. Oh, and it's twice that
48:44
now with Harvey closing that
48:46
round, right? OK, well, so number
48:48
one, hugely competitive space. Number
48:51
two, a lot of times VCs
48:54
will encourage startups to raise
48:57
big rounds because it sort of freezes
48:59
the market. Like if somebody
49:01
has hit a billion dollar valuation, then
49:04
that stops other people from being able to
49:07
raise. So in
49:10
this broadway or
49:12
as narrow a way as I can
49:14
ask this, how are you
49:16
thinking about being in a super competitive space
49:18
where, A, you've got a competitor
49:20
that people would look at the market
49:22
and say, further ahead, bigger valuation? You've
49:25
made the argument also that this
49:28
is early innings. But number two, everybody,
49:31
again, I'm looking at the tech name back end, the
49:33
amount of legal AI startups.
49:37
How many lawyers are there in the United States? That's
49:40
a good question. I want to say it's somewhere in the
49:42
neighborhood of like 800,000 attorneys,
49:45
and that's going to grow by like 18% in
49:47
the next 10 years or something like that. So
49:49
a big market, but also that's
49:51
why everyone's going after it. So as
49:54
a founder
49:56
of a company in a super
49:58
competitive space, How
50:01
are you thinking about that? Is it
50:03
just, we got to
50:05
focus on the product and everything
50:07
will work out? Or is
50:09
there strategically as you're building
50:12
this company, how big is the team right now? We're
50:15
10 people right now. Yeah. As you're
50:17
building the company, do you make decisions
50:19
based on that that you wouldn't make
50:21
if you had a space to yourself?
50:23
You know what I mean? Well,
50:27
I think if you have a
50:29
space to yourself, you don't have a business. Right?
50:32
Or you're doing something so crazy that no one
50:34
knows it's a business yet. That's true. That's true
50:36
too. But that's so rare because even
50:38
the idea of sort of monkeys, infinite
50:42
number of monkeys in front of an infinite number
50:44
of typewriters, the
50:46
odds are there's going to be a lot that are really, really close to
50:48
each other. And so I feel like that's
50:50
kind of where we're at right now. And the answer
50:53
to this is, you have to
50:55
kind of play the market a little bit, but also make
50:58
sure that you're focusing on the fundamentals. I mean, we did
51:00
this at Urban Airship, which was let's make sure we get
51:02
a sales motion in place such that we
51:05
know how and who we're selling to. And we
51:07
have a strong relationship with those folks. And let's
51:09
lock that up. And we
51:11
did raise, I don't know, all of us there, we
51:14
raised about 50 million, which is like chump change now.
51:16
Our Series A was $1.1 million, which
51:19
is like a lack of investment. That's
51:21
not even a pre-seed now. And
51:23
so, but for us at
51:26
Urban Airship, what we did is we said,
51:28
let's get that sales motion right. Let's make
51:30
sure we have a really solid product market
51:32
fit. And then we're going to
51:34
watch this market. And what we're going to do is
51:36
we're going to identify those players that either overraised or
51:39
couldn't find product market fit or ran out of gas. And
51:41
then what we did is we acquired them for pennies on
51:43
the dollar. And we got great teams in tech, and
51:46
we were able to fold that into our sales motion.
51:48
And the uplift was anywhere from 15 to 30% in
51:52
net new growth from a
51:54
business perspective. And I think the same thing
51:56
is going to happen here. And that's how we're architecting our
51:58
business. It's one of the reasons we're- we only
52:00
raised the 1.7 million is we're in
52:03
this, we wanted to do a small amount to prove
52:05
out some things that we think we're proving out right
52:07
now. And then we'll
52:09
likely raise again, but
52:11
we don't have to because we're now throwing
52:13
off a bunch of cash and building an
52:15
interesting compelling business. But as we
52:17
hit 18, 24 months, my gut says that
52:19
this is gonna go faster than the original
52:21
sort of SaaS cycle because AI
52:23
is moving so fast that we'll be able
52:25
to look out at the landscape and say,
52:27
well, where do we have gaps? What could
52:30
we fold into our sales motion such that
52:32
it'll, we'll have some
52:34
uplift there that allows us to continue
52:36
to grow and scale and actually turn us into
52:38
not just an interesting
52:40
company, but a brand that people
52:42
will depend on for legal in
52:44
general. And that to me is where
52:47
it gets really interesting because AI is gonna become a
52:49
feature. Right. Without
52:51
putting words in your mouth, what it sounds
52:53
like you're saying to me is let other
52:55
people get headlines with big raises. Let
52:59
20 other people raise rounds.
53:01
We'll make the headlines down the road
53:03
if we've executed on the sales, because
53:06
then we'll make the headlines because our raise
53:09
will be based on the revenue that because
53:11
we've executed on the plan. Yeah,
53:14
I think that's about right. But I think there's
53:16
also some element of, you have to play the
53:18
market a little bit too. Obviously,
53:21
after we close the round and now
53:23
it's, there's so much interest in the
53:25
space that we get investors constantly pinging
53:27
us. And I take those
53:29
calls. I have those conversations with those
53:31
folks and there's always interest, which is
53:33
great. That's great. But again, we have
53:35
to focus on execution. And I know
53:37
it sounds boring, but that's the critical
53:39
pieces right now that I think is
53:41
really important. Now, I think the
53:44
unfair advantage that we have is as a team, we've
53:46
all worked together. It's a bunch of airship folks that
53:48
have come together, urban airship
53:50
folks that have are putting this together.
53:52
And we've all scaled companies like this
53:55
before. So a lot of
53:57
these startups have the challenge of market
53:59
headwinds. and implementation and all those
54:01
things, as well as learning as they go
54:03
on how to scale a company. Team dynamics
54:05
too. The team dynamics are
54:08
really helpful. We can shout at each other or get
54:10
angry and have a disagreement, and guess what? The next
54:12
day we wake up and go, okay, cool, we solved
54:14
that problem. Because
54:16
we have that relationship, I think that sets us
54:18
apart as well. I
54:21
think it's one of the reasons we've been able to accomplish so much in just
54:23
a year. Especially
54:26
when I look at some of the
54:28
folks that are in our same space or
54:30
even in the same portfolio companies like a
54:32
Gradient and others, the
54:35
amount we've accomplished and the challenges they have
54:37
around just what I consider simple stuff
54:39
around scaling, they're struggling with. But
54:41
I always give feedback on those things and say, hey,
54:43
maybe you want to think about this. That
54:46
to me is also really fun to help
54:48
those other companies. Cool.
54:51
You mentioned the pre-seed round,
54:54
which was led by Gradient, which is Google's
54:56
seed fund. We are
54:58
honored to be a part of it as well
55:00
with the Ride Home AI Fund and the Ride
55:02
Home. We're excited. So happy you all are involved
55:05
without a doubt. Very excited. Again, if
55:07
you want to learn more as you're listening
55:10
right now, it's casemark.ai.
55:13
But also, if people are listening and
55:17
want to learn more or want to get
55:19
involved, are you hiring?
55:23
Do you have an ask for this audience
55:25
that you never know who's listening that might
55:27
have something that they can deliver for you? I
55:30
mean, we're definitely hiring. We're
55:32
looking for folks on
55:34
the engineering side of things without
55:37
a doubt, especially DevOps, Site
55:39
Reliability. We're having some
55:41
what I call wonderful problems around scaling
55:43
in the sense that this
55:45
is going really fast and so we have
55:47
to figure out how we're going to scale
55:49
those things. So always looking for folks who
55:52
are interested in a serious going
55:54
concern that has the fun problems of a
55:56
startup which are scaling and those
55:59
kinds of things. If you
56:01
have folks that are running, whether it's an
56:03
insurance, friends that are running insurance defense firm
56:05
or doing transactional work like
56:07
personal injury stuff, then hey, they should point them
56:09
out in our direction, and have them check out
56:11
case mark.ai. We offer
56:13
up a little free plan where you can test it
56:16
out with a couple of different free summaries.
56:19
We find that when people see the free
56:21
version and they try it against it, especially
56:23
a deposition they've taken themselves, and they see
56:25
the summary, they're actually really blown away. That
56:28
would be my ask, was just check it out and spread the word.
56:32
By the way, according to Claude, there's
56:34
1.3 million lawyers in the US. Oh,
56:36
so I was wrong. I was going
56:38
off of what I'd seen in my
56:40
last Gartner report, which I probably mis-voted.
56:42
Great. Claude must
56:44
have it right. I was going to say AI to the rescue.
56:48
They have references, so I believe
56:50
it. Yeah. That's true. That's fair.
56:52
Fair enough. Again, case mark,
56:54
case mark.ai. Scott,
56:56
thanks for coming on and telling
56:59
us all about that. Chris,
57:01
thank you for joining me as well. Yeah.
57:04
Thank you, Brian. Thank you, Chris. Thanks so
57:06
much for having us. Really appreciate it. The
57:14
travel experiences you'll have once you
57:17
arrive. That's why you need Viator.
57:19
Book guided tours, activities, excursions, and
57:21
more in one place to make
57:24
your trip truly unforgettable. Viator has
57:26
over 300,000 travel experiences to choose
57:28
from. Everything from simple tours to
57:30
extreme adventures, and all the niche,
57:33
interesting stuff in between. So you
57:35
can plan something that everyone you're
57:37
traveling with will enjoy. Real traveler
57:39
reviews give the inside scoop from
57:42
people who've already been on the
57:44
experiences you're considering. So you can
57:46
plan with confidence. Free cancellation helps
57:49
you plan for the unexpected, and
57:51
24-7 customer support means you can
57:53
travel worry-free. Download the Viator app
57:56
now and use code Viator10 for
57:58
10% off. off your
58:00
first booking in the app. Find travel
58:02
experiences for you. Do more
58:04
with Viator.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More