1
00:00:00,000 --> 00:00:07,280
Welcome to Reinventing Professionals, a podcast hosted by industry analyst Ari Kaplan,
2
00:00:07,280 --> 00:00:12,680
which shares ideas, guidance and perspectives from market leaders shaping the next generation
3
00:00:12,680 --> 00:00:16,680
of legal and professional services.
4
00:00:16,680 --> 00:00:22,040
This is Ari Kaplan, and I'm speaking today with Lori Erlich, the Chief Legal Officer
5
00:00:22,040 --> 00:00:29,160
and Farragasmi, co-founder and chief product officer at Dioptra, an AI-powered contract
6
00:00:29,160 --> 00:00:31,960
redlining and playbook generation tool.
7
00:00:31,960 --> 00:00:33,960
Lori, Farragasmi, great to see you both.
8
00:00:33,960 --> 00:00:36,120
Likewise, thanks for having us.
9
00:00:36,120 --> 00:00:40,040
I'm looking forward to this conversation, so Farrag, tell us about your background and
10
00:00:40,040 --> 00:00:42,320
the genesis of Dioptra.
11
00:00:42,320 --> 00:00:46,960
I've been building AI products for the past 10 years in an above different industries,
12
00:00:46,960 --> 00:00:51,280
some more regulated than others, and then most recently I was building the internal
13
00:00:51,280 --> 00:00:56,560
AI infrastructure to have secure and safe AI tool sets, body-fying, and I've been teaching
14
00:00:56,560 --> 00:01:01,120
at Columbia at the Business School for an above years and AI product management class.
15
00:01:01,120 --> 00:01:07,800
To take a little bit about Dioptra, we actually started doing a lot of consulting work for
16
00:01:07,800 --> 00:01:14,280
an above different enterprises and helping them with a lot of different AI applications.
17
00:01:14,280 --> 00:01:19,920
By doing that, we were negotiating contracts with all those large enterprises, and none
18
00:01:19,920 --> 00:01:24,760
of us founders are lawyers, and so we were constantly trying to understand what every
19
00:01:24,760 --> 00:01:28,240
one of those terms or contracts were saying and not doing.
20
00:01:28,240 --> 00:01:32,240
Because we're part of the white combinator community of founders, we're constantly chatting
21
00:01:32,240 --> 00:01:36,680
with other founders, and we realized this was bigger problem than just us.
22
00:01:36,680 --> 00:01:42,040
A lot of the small, medium, or growth and companies were having the same challenge, the business
23
00:01:42,040 --> 00:01:46,800
owners were having those challenges when they negotiated contracts, and so we realized
24
00:01:46,800 --> 00:01:50,400
that they had to be a better way to do that, and that was coming at about the same time
25
00:01:50,400 --> 00:01:56,400
as TGBT came out, and we also had a very strong understanding of what the technology was
26
00:01:56,400 --> 00:02:00,400
capable of doing, the new technology was capable of doing, and that's how we decided to
27
00:02:00,400 --> 00:02:02,440
get started in the contracts space.
28
00:02:02,440 --> 00:02:07,080
Laurie, tell us about your background and how you leverage your extensive legal experience
29
00:02:07,080 --> 00:02:11,480
to enhance the contract redlining and playbook generation process.
30
00:02:11,480 --> 00:02:19,120
I came out of top law school, went to big law, like everybody else does, as an IP associate,
31
00:02:19,120 --> 00:02:25,040
realized that did not want a partner life, and so I looked to go in-house, and I went to
32
00:02:25,040 --> 00:02:29,480
a large outsourcing company and did contract after contract after contract to the point
33
00:02:29,480 --> 00:02:33,440
where I was so bored, I stopped reading contracts before I got on the phone to negotiate
34
00:02:33,440 --> 00:02:38,960
them because they already knew what all the arguments were going to be, and I was bored,
35
00:02:38,960 --> 00:02:42,720
and I needed to figure out what I wanted to do.
36
00:02:42,720 --> 00:02:46,840
And I went on a self-discovery journey and realized that what I'm really passionate about
37
00:02:46,840 --> 00:02:48,600
is transformation.
38
00:02:48,600 --> 00:02:54,000
And so I pivoted while I was still a cognizant from doing, as the chief counsel for our insurance
39
00:02:54,000 --> 00:02:59,280
vertical, to heading our, like, really, founding and heading our legal ops department, and
40
00:02:59,280 --> 00:03:07,080
started working on data, metrics, and centralizing intake and knowledge management, and I really
41
00:03:07,080 --> 00:03:11,680
loved that, but I didn't love not being close to the business.
42
00:03:11,680 --> 00:03:16,720
And so I went to data dog, and I got to do commercial, as the chief commercial counsel,
43
00:03:16,720 --> 00:03:20,320
and I got to run it from a kind of operational standpoint.
44
00:03:20,320 --> 00:03:25,200
So it's not, like, just doing the contracts and law, but how do we do contracts really
45
00:03:25,200 --> 00:03:26,200
efficiently?
46
00:03:26,200 --> 00:03:27,200
How do we make it scalable?
47
00:03:27,200 --> 00:03:31,360
How do we spend our time on what matters and don't spend time on things where we're not
48
00:03:31,360 --> 00:03:34,040
providing really legal value?
49
00:03:34,040 --> 00:03:39,480
And loved it, and it was, like, totally fulfilling and satisfying, but I wasn't as close to
50
00:03:39,480 --> 00:03:40,640
the businesses I wanted to be.
51
00:03:40,640 --> 00:03:41,640
And I was close to the business.
52
00:03:41,640 --> 00:03:44,000
I talked to sales every day.
53
00:03:44,000 --> 00:03:46,280
I talked to our partnership team every day.
54
00:03:46,280 --> 00:03:52,360
I talked to all the vendor teams, but I didn't feel like I was at the core of the decision-making
55
00:03:52,360 --> 00:03:53,960
for the company.
56
00:03:53,960 --> 00:03:58,120
So I came into Diopter Room, part of every decision, which is wonderful.
57
00:03:58,120 --> 00:04:05,240
But I really leveraged that combination of contracting expertise and ops expertise to help
58
00:04:05,240 --> 00:04:10,040
both Diopter's development and our customers in their journey, because I know why we need
59
00:04:10,040 --> 00:04:11,040
playbox.
60
00:04:11,040 --> 00:04:14,400
I know the benefit of standardization on how we negotiate contracts.
61
00:04:14,400 --> 00:04:19,080
I know how to understand when I'm looking at a red line, what matters and what doesn't
62
00:04:19,080 --> 00:04:24,920
matter and help our customers to the way I think of it is when you use a tool like Diopter
63
00:04:24,920 --> 00:04:30,000
Room, you can spend a lot more time on the strategy of contracting and a lot less time
64
00:04:30,000 --> 00:04:31,760
on the grunt work of contracting.
65
00:04:31,760 --> 00:04:36,800
So when you don't have to spend all of your time taking, you're fine, that precedent that
66
00:04:36,800 --> 00:04:40,880
has that clause that you really like to one time and you spend two hours looking for
67
00:04:40,880 --> 00:04:44,080
it and then you find it and you're like, oh, that Ashley isn't quite right.
68
00:04:44,080 --> 00:04:46,000
That's not a good use of our time.
69
00:04:46,000 --> 00:04:49,920
But if you can take your precedence, like we have people be able to do them, no, let me
70
00:04:49,920 --> 00:04:55,480
just need a couple and synthesize them into a playbook, then you spend where I think the
71
00:04:55,480 --> 00:04:59,640
real value of the capital, what lawyers bring to the table, which is looking at those
72
00:04:59,640 --> 00:05:04,120
rules that come out of your precedent saying, I do want that in my contracts or I don't
73
00:05:04,120 --> 00:05:07,480
want them in my contracts or I want this in my contracts, but actually I want it with
74
00:05:07,480 --> 00:05:08,960
this variation.
75
00:05:08,960 --> 00:05:10,480
And that's, to me, that's the value.
76
00:05:10,480 --> 00:05:12,720
The value isn't the red line.
77
00:05:12,720 --> 00:05:18,200
The value is what you decide matters to be redlined in the contract.
78
00:05:18,200 --> 00:05:23,360
And so I'm able to help our customers on that journey of the self discovery of what you want
79
00:05:23,360 --> 00:05:25,080
your contracts to be.
80
00:05:25,080 --> 00:05:28,320
That's part of what you do when you say, okay, I'm not going to do the red line.
81
00:05:28,320 --> 00:05:32,360
I'm going to have someone else do the red line in the AI, do the red lining, but I need
82
00:05:32,360 --> 00:05:33,960
to tell the AI what to do.
83
00:05:33,960 --> 00:05:39,240
>> What's the difference between rag and agents when we're talking about AI?
84
00:05:39,240 --> 00:05:46,200
But lawyer was describing is a very good example of what agents do well, right?
85
00:05:46,200 --> 00:05:55,240
So our agents are able to go through those precedents, just still the intent, and then
86
00:05:55,240 --> 00:06:03,320
come up with that plan, if you will, to actually be able to build that playbook and implement
87
00:06:03,320 --> 00:06:05,800
it consistently and accurately, right?
88
00:06:05,800 --> 00:06:09,000
And so agents, and I know there are multiple definitions out there, but
89
00:06:09,000 --> 00:06:15,600
agents are, or AI agents are those entities that we be able to take more than one step.
90
00:06:15,600 --> 00:06:21,000
We'll be able to take multiple steps, usually we use different models to achieve every
91
00:06:21,000 --> 00:06:22,760
single one of those steps.
92
00:06:22,760 --> 00:06:29,400
As opposed to rag, rag is more, you have a database and, or you have some amount of data,
93
00:06:29,400 --> 00:06:32,080
even if it is contracts or database and so on and so forth.
94
00:06:32,080 --> 00:06:37,800
And you just want to search and find something from that collection of information.
95
00:06:37,800 --> 00:06:40,360
And so the approach is fundamentally different.
96
00:06:40,360 --> 00:06:45,680
In one step, you're focusing on a process that you're trying to achieve on the other step,
97
00:06:45,680 --> 00:06:49,560
it's more like a search or like retrieval of information, which is what rag stands for,
98
00:06:49,560 --> 00:06:50,560
right?
99
00:06:50,560 --> 00:06:55,760
And so being able to search and retrieve information from a collection of data that you already
100
00:06:55,760 --> 00:06:57,680
have somewhere.
101
00:06:57,680 --> 00:07:02,840
>> When you want to stop having legal be directory assistance in your in-house, instead you want
102
00:07:02,840 --> 00:07:07,560
people to just go to a source for that information that they're looking for, rag is fantastic
103
00:07:07,560 --> 00:07:12,440
for that because it's taking the information that you've created and it's a library and you
104
00:07:12,440 --> 00:07:15,320
really just want people to get the right answer from the library.
105
00:07:15,320 --> 00:07:19,640
Whereas for RedLine, in contract, it's an action, not information.
106
00:07:19,640 --> 00:07:25,320
You need that, you want a red line that is like a lawyer with red line.
107
00:07:25,320 --> 00:07:28,000
And rag alone can't do that.
108
00:07:28,000 --> 00:07:34,560
>> I'm always amazed by seeing all the excitement about AIH and
109
00:07:34,560 --> 00:07:40,880
today when we started doing this, we launched our product, it was May last year.
110
00:07:40,880 --> 00:07:46,040
And we launched it with, we sent Sincini back then and when we started talking about agents,
111
00:07:46,040 --> 00:07:48,120
everybody was like, what are you talking about?
112
00:07:48,120 --> 00:07:49,120
What are agents?
113
00:07:49,120 --> 00:07:52,320
And it's like nobody, it wasn't a thing.
114
00:07:52,320 --> 00:07:54,760
It was like, everybody was like into rag and everything.
115
00:07:54,760 --> 00:07:58,720
And it was like that agents is so much more powerful because it allows you to do more complex
116
00:07:58,720 --> 00:07:59,720
things.
117
00:07:59,720 --> 00:08:04,520
And now everybody is asking about agents and these agents in every single operational
118
00:08:04,520 --> 00:08:06,760
and enterprise and industry, right?
119
00:08:06,760 --> 00:08:11,960
It's just in the legal space, they are extremely powerful when dealing with complex,
120
00:08:11,960 --> 00:08:14,040
multi-step processes.
121
00:08:14,040 --> 00:08:19,960
>> Lori, what challenges are you seeing around the adoption of AI?
122
00:08:19,960 --> 00:08:22,720
>> I think there are two big challenges.
123
00:08:22,720 --> 00:08:23,920
One is trust.
124
00:08:23,920 --> 00:08:27,880
When nobody really knew how to use GenNRVI when it first came out.
125
00:08:27,880 --> 00:08:31,920
And so people made a lot of mistakes because they thought it would be like Google search,
126
00:08:31,920 --> 00:08:36,920
it's pulling things that exist, but no, it's generating, that's why it's called GenNRVI,
127
00:08:36,920 --> 00:08:37,920
right?
128
00:08:37,920 --> 00:08:39,280
And so we didn't use it right.
129
00:08:39,280 --> 00:08:43,640
And a lot of people got hung out to drive because of errors that they made.
130
00:08:43,640 --> 00:08:48,440
And so there's fear around thinking AI can't be trusted.
131
00:08:48,440 --> 00:08:51,000
I think that's not the case anymore.
132
00:08:51,000 --> 00:08:54,800
Most of the tools that are out there have conquered the hallucination issue to a large
133
00:08:54,800 --> 00:08:55,800
degree.
134
00:08:55,800 --> 00:08:58,240
It still happens, but it's not the Wild West anymore.
135
00:08:58,240 --> 00:09:05,080
And I think the other big issue is there's just so much noise, right?
136
00:09:05,080 --> 00:09:06,080
You can't.
137
00:09:06,080 --> 00:09:08,680
No one's going to the market without saying they have AI today.
138
00:09:08,680 --> 00:09:10,640
Everybody has AI.
139
00:09:10,640 --> 00:09:13,200
And in some cases it's really good AI.
140
00:09:13,200 --> 00:09:16,360
And in other cases it's maybe not such great AI.
141
00:09:16,360 --> 00:09:22,000
And it's finding out, figuring out, like, and I'm sorry, I'll just, a lot of the in-house
142
00:09:22,000 --> 00:09:24,160
departments and firms have this mandate adopt AI.
143
00:09:24,160 --> 00:09:27,360
And we're seeing a huge sea change in that before it was experimented.
144
00:09:27,360 --> 00:09:28,920
And now it's adopt.
145
00:09:28,920 --> 00:09:31,560
But that's not how you solve problems.
146
00:09:31,560 --> 00:09:37,320
On the lunch day, Meredith talked about how you need to figure out what your problem is.
147
00:09:37,320 --> 00:09:38,720
And then go look for the solution.
148
00:09:38,720 --> 00:09:41,320
And a lot of people are coming to the table saying, "I need AI."
149
00:09:41,320 --> 00:09:43,600
And they haven't figured out what their problem is.
150
00:09:43,600 --> 00:09:47,440
And so they're not necessarily selecting tools that are the right tool to solve their problems
151
00:09:47,440 --> 00:09:49,560
because they're not starting from what's my problem.
152
00:09:49,560 --> 00:09:50,560
And I think that's true.
153
00:09:50,560 --> 00:09:52,320
It's not new to Jeremy AI, right?
154
00:09:52,320 --> 00:09:56,200
That is a problem with technology adoption that it's timeless.
155
00:09:56,200 --> 00:10:01,000
It's just that there's more ways, I think, with AI than there was with other tools.
156
00:10:01,000 --> 00:10:06,000
>> Parallel, what type of AI are buyers most interested in?
157
00:10:06,000 --> 00:10:10,440
>> It's only that I was actually reading and listening to a podcast, another podcast from
158
00:10:10,440 --> 00:10:13,040
the OIC community.
159
00:10:13,040 --> 00:10:19,040
And there was a lot of talk about AI agents and specifically in the enterprise space,
160
00:10:19,040 --> 00:10:24,120
like the B2B, not SAS, B2B agents software in the enterprises.
161
00:10:24,120 --> 00:10:28,560
And that is the range of applications that we hear about is tremendous.
162
00:10:28,560 --> 00:10:36,400
It is any type of operational task or process, which is exactly the sweet spot of agents, right?
163
00:10:36,400 --> 00:10:44,400
Any operational process that enterprises have to do again and again is now being automated
164
00:10:44,400 --> 00:10:45,400
or semi-automated.
165
00:10:45,400 --> 00:10:52,040
There's always a human interloop that is supervising what the agent is doing.
166
00:10:52,040 --> 00:10:55,920
Then imagine every single process that enterprises have to do, right?
167
00:10:55,920 --> 00:10:59,120
And every single one of those are now getting agents for.
168
00:10:59,120 --> 00:11:05,240
And there's a huge amount of adoption and demand for those tools if the tool is actually working.
169
00:11:05,240 --> 00:11:06,480
And I think that's a challenge.
170
00:11:06,480 --> 00:11:09,760
Like Neuys said, there's a lot of noise out there today.
171
00:11:09,760 --> 00:11:12,840
And it is difficult, often here customers that come to me.
172
00:11:12,840 --> 00:11:16,920
And they're like, I tried XYZ and I tried ABC and I tried, and it's not working.
173
00:11:16,920 --> 00:11:18,960
And why would you be able to do that differently?
174
00:11:18,960 --> 00:11:19,960
And I get it, right?
175
00:11:19,960 --> 00:11:23,160
I get it, everybody is playing with AI.
176
00:11:23,160 --> 00:11:28,400
It's hard to understand what is actually going to be working from a biased perspective.
177
00:11:28,400 --> 00:11:32,280
But if the tool works, I've seen a significant adoption.
178
00:11:32,280 --> 00:11:36,440
Neuys is what comes, like what you're doing is has nothing to do with what I've seen.
179
00:11:36,440 --> 00:11:38,920
And it's extremely different and the quality is different.
180
00:11:38,920 --> 00:11:41,320
And I think that is true for not just the legal space.
181
00:11:41,320 --> 00:11:46,440
I think that's true for all the different verticals where agents are being used.
182
00:11:46,440 --> 00:11:51,520
Laurie, what's the challenge for lawyers to secure buy-in for new AI tools?
183
00:11:51,520 --> 00:11:57,480
>> I think this is different from any other legal technology because AI is not a legal tool.
184
00:11:57,480 --> 00:12:00,280
AI is everything tool.
185
00:12:00,280 --> 00:12:01,600
It's like computers, right?
186
00:12:01,600 --> 00:12:05,200
When companies adopt computers, it was only like lawyers had to make a business case for
187
00:12:05,200 --> 00:12:06,000
computers.
188
00:12:06,000 --> 00:12:09,080
It was the company saying to lawyers, and now you have to use computers.
189
00:12:09,080 --> 00:12:13,600
And I think that's the same thing happening with AI that lawyers are being told you have to use AI.
190
00:12:13,600 --> 00:12:19,560
And I think the problem with adoption is and buy-in is really a mix of any time there's change.
191
00:12:19,560 --> 00:12:27,320
It's scary and AI specifically taking on work is scary because it's is this my job?
192
00:12:27,320 --> 00:12:33,200
And then some tools really require lawyers to become prompt engineers, which I think lawyers
193
00:12:33,200 --> 00:12:39,360
can be very good prompt engineers because it's manipulation of language, which is what we all do as lawyers.
194
00:12:39,360 --> 00:12:42,840
But I compare it to the first time I tried to write a patent claim.
195
00:12:42,840 --> 00:12:46,960
I met who are laughed at me because it was so bad it was written in English.
196
00:12:46,960 --> 00:12:49,040
It just wasn't right for a pen.
197
00:12:49,040 --> 00:12:50,040
And so it's the same thing.
198
00:12:50,040 --> 00:12:53,920
If you could write a contract that's and then you try to write a rule for a contract that's
199
00:12:53,920 --> 00:12:56,120
a prompt, it's not going to be quite right.
200
00:12:56,120 --> 00:12:57,720
You have to learn how to do it.
201
00:12:57,720 --> 00:13:03,920
And so the tools that require their customers to master that art without giving them assistance,
202
00:13:03,920 --> 00:13:09,840
I think is asking lawyers to learn a brand new tool set.
203
00:13:09,840 --> 00:13:14,360
How does diopter differ from other contracting tools in the market?
204
00:13:14,360 --> 00:13:20,800
I've been building AI products for the past 10 years and the challenge is always being
205
00:13:20,800 --> 00:13:25,920
the trade-off between how to navigate your tool is so how much can it handle right versus
206
00:13:25,920 --> 00:13:29,600
what is the accuracy that you can achieve right?
207
00:13:29,600 --> 00:13:34,280
And there's a trade-off because the more analyst your AI is less accurate it is at doing every
208
00:13:34,280 --> 00:13:39,960
single thing which is understandable right.
209
00:13:39,960 --> 00:13:43,360
We have been focusing on that trade-off since day one because we knew we needed to be able
210
00:13:43,360 --> 00:13:47,640
to tackle a number of different types of contracts right.
211
00:13:47,640 --> 00:13:52,480
And we knew also that accuracy matter like the quality of the red lines that land on that
212
00:13:52,480 --> 00:13:56,720
I'm not going to say piece of paper but on that contract is important.
213
00:13:56,720 --> 00:14:02,720
It's because we have started from that we have been able to build an agent in a way that
214
00:14:02,720 --> 00:14:08,720
is extremely accurate as compared to a lot of tools out there when building the red lines
215
00:14:08,720 --> 00:14:13,720
or when writing the red lines in a way that it is surgical in a way that it is elegant
216
00:14:13,720 --> 00:14:17,360
and then similar to what humans would do right.
217
00:14:17,360 --> 00:14:26,120
And then we use all those playbooks to guide and make our agent generalizable right.
218
00:14:26,120 --> 00:14:30,240
And that's how we are able to tackle different types of contracts by helping our customers
219
00:14:30,240 --> 00:14:35,040
build those are people from their presidents and distilling that knowledge from those presidents.
220
00:14:35,040 --> 00:14:39,640
And I'll say our agent is smart enough that it keeps schooling me.
221
00:14:39,640 --> 00:14:41,640
Laurie how do you see contracting evolving?
222
00:14:41,640 --> 00:14:43,920
Oh it's going to be so different.
223
00:14:43,920 --> 00:14:50,680
So right now I would say the majority of companies contracts go through every contract goes
224
00:14:50,680 --> 00:14:57,680
to the legal team or contracting team and it's a lot of tasks switching.
225
00:14:57,680 --> 00:15:01,120
So you have to look at the playbook or look at the guidelines or look at the checklist and
226
00:15:01,120 --> 00:15:03,800
look at the document and go back and forth and back and forth and send it to the team and
227
00:15:03,800 --> 00:15:09,480
send it to that team and there's just a lot of administration around the contracting process.
228
00:15:09,480 --> 00:15:17,180
And I think that AI is going to allow us our CTO of the other days that the gift of AI
229
00:15:17,180 --> 00:15:19,080
allows us to elevate our thinking.
230
00:15:19,080 --> 00:15:21,000
We don't have to be in the weeds anymore.
231
00:15:21,000 --> 00:15:24,160
We can think strategically spend more time thinking strategically.
232
00:15:24,160 --> 00:15:26,120
And I think that's what we're going to be able to do.
233
00:15:26,120 --> 00:15:30,520
We're going to be able to spend a lot more time thinking about the strategic aspects of
234
00:15:30,520 --> 00:15:36,040
contracting and a lot less time doing the mundane administrative aspects of contracting.
235
00:15:36,040 --> 00:15:42,720
This is Ari Kaplan speaking with Laurie Erlich, the chief legal officer and Farragasmi,
236
00:15:42,720 --> 00:15:47,840
the co-founder and chief product officer at Deoctra, an AI powered contract redlining
237
00:15:47,840 --> 00:15:50,080
and playbook generation tool.
238
00:15:50,080 --> 00:15:52,440
Laurie Farrah thank you so very much.
239
00:15:52,440 --> 00:15:53,440
Thank you Ari.
240
00:15:53,440 --> 00:15:54,440
Thank you.
241
00:15:54,440 --> 00:15:59,000
Thank you for listening to the reinventing professionals podcast.
242
00:15:59,000 --> 00:16:05,400
Visit reinventingprofessionals.com or Ari Kaplan advisors.com to learn more.
00:00:00,000 --> 00:00:07,280
Welcome to Reinventing Professionals, a podcast hosted by industry analyst Ari Kaplan,
2
00:00:07,280 --> 00:00:12,680
which shares ideas, guidance and perspectives from market leaders shaping the next generation
3
00:00:12,680 --> 00:00:16,680
of legal and professional services.
4
00:00:16,680 --> 00:00:22,040
This is Ari Kaplan, and I'm speaking today with Lori Erlich, the Chief Legal Officer
5
00:00:22,040 --> 00:00:29,160
and Farragasmi, co-founder and chief product officer at Dioptra, an AI-powered contract
6
00:00:29,160 --> 00:00:31,960
redlining and playbook generation tool.
7
00:00:31,960 --> 00:00:33,960
Lori, Farragasmi, great to see you both.
8
00:00:33,960 --> 00:00:36,120
Likewise, thanks for having us.
9
00:00:36,120 --> 00:00:40,040
I'm looking forward to this conversation, so Farrag, tell us about your background and
10
00:00:40,040 --> 00:00:42,320
the genesis of Dioptra.
11
00:00:42,320 --> 00:00:46,960
I've been building AI products for the past 10 years in an above different industries,
12
00:00:46,960 --> 00:00:51,280
some more regulated than others, and then most recently I was building the internal
13
00:00:51,280 --> 00:00:56,560
AI infrastructure to have secure and safe AI tool sets, body-fying, and I've been teaching
14
00:00:56,560 --> 00:01:01,120
at Columbia at the Business School for an above years and AI product management class.
15
00:01:01,120 --> 00:01:07,800
To take a little bit about Dioptra, we actually started doing a lot of consulting work for
16
00:01:07,800 --> 00:01:14,280
an above different enterprises and helping them with a lot of different AI applications.
17
00:01:14,280 --> 00:01:19,920
By doing that, we were negotiating contracts with all those large enterprises, and none
18
00:01:19,920 --> 00:01:24,760
of us founders are lawyers, and so we were constantly trying to understand what every
19
00:01:24,760 --> 00:01:28,240
one of those terms or contracts were saying and not doing.
20
00:01:28,240 --> 00:01:32,240
Because we're part of the white combinator community of founders, we're constantly chatting
21
00:01:32,240 --> 00:01:36,680
with other founders, and we realized this was bigger problem than just us.
22
00:01:36,680 --> 00:01:42,040
A lot of the small, medium, or growth and companies were having the same challenge, the business
23
00:01:42,040 --> 00:01:46,800
owners were having those challenges when they negotiated contracts, and so we realized
24
00:01:46,800 --> 00:01:50,400
that they had to be a better way to do that, and that was coming at about the same time
25
00:01:50,400 --> 00:01:56,400
as TGBT came out, and we also had a very strong understanding of what the technology was
26
00:01:56,400 --> 00:02:00,400
capable of doing, the new technology was capable of doing, and that's how we decided to
27
00:02:00,400 --> 00:02:02,440
get started in the contracts space.
28
00:02:02,440 --> 00:02:07,080
Laurie, tell us about your background and how you leverage your extensive legal experience
29
00:02:07,080 --> 00:02:11,480
to enhance the contract redlining and playbook generation process.
30
00:02:11,480 --> 00:02:19,120
I came out of top law school, went to big law, like everybody else does, as an IP associate,
31
00:02:19,120 --> 00:02:25,040
realized that did not want a partner life, and so I looked to go in-house, and I went to
32
00:02:25,040 --> 00:02:29,480
a large outsourcing company and did contract after contract after contract to the point
33
00:02:29,480 --> 00:02:33,440
where I was so bored, I stopped reading contracts before I got on the phone to negotiate
34
00:02:33,440 --> 00:02:38,960
them because they already knew what all the arguments were going to be, and I was bored,
35
00:02:38,960 --> 00:02:42,720
and I needed to figure out what I wanted to do.
36
00:02:42,720 --> 00:02:46,840
And I went on a self-discovery journey and realized that what I'm really passionate about
37
00:02:46,840 --> 00:02:48,600
is transformation.
38
00:02:48,600 --> 00:02:54,000
And so I pivoted while I was still a cognizant from doing, as the chief counsel for our insurance
39
00:02:54,000 --> 00:02:59,280
vertical, to heading our, like, really, founding and heading our legal ops department, and
40
00:02:59,280 --> 00:03:07,080
started working on data, metrics, and centralizing intake and knowledge management, and I really
41
00:03:07,080 --> 00:03:11,680
loved that, but I didn't love not being close to the business.
42
00:03:11,680 --> 00:03:16,720
And so I went to data dog, and I got to do commercial, as the chief commercial counsel,
43
00:03:16,720 --> 00:03:20,320
and I got to run it from a kind of operational standpoint.
44
00:03:20,320 --> 00:03:25,200
So it's not, like, just doing the contracts and law, but how do we do contracts really
45
00:03:25,200 --> 00:03:26,200
efficiently?
46
00:03:26,200 --> 00:03:27,200
How do we make it scalable?
47
00:03:27,200 --> 00:03:31,360
How do we spend our time on what matters and don't spend time on things where we're not
48
00:03:31,360 --> 00:03:34,040
providing really legal value?
49
00:03:34,040 --> 00:03:39,480
And loved it, and it was, like, totally fulfilling and satisfying, but I wasn't as close to
50
00:03:39,480 --> 00:03:40,640
the businesses I wanted to be.
51
00:03:40,640 --> 00:03:41,640
And I was close to the business.
52
00:03:41,640 --> 00:03:44,000
I talked to sales every day.
53
00:03:44,000 --> 00:03:46,280
I talked to our partnership team every day.
54
00:03:46,280 --> 00:03:52,360
I talked to all the vendor teams, but I didn't feel like I was at the core of the decision-making
55
00:03:52,360 --> 00:03:53,960
for the company.
56
00:03:53,960 --> 00:03:58,120
So I came into Diopter Room, part of every decision, which is wonderful.
57
00:03:58,120 --> 00:04:05,240
But I really leveraged that combination of contracting expertise and ops expertise to help
58
00:04:05,240 --> 00:04:10,040
both Diopter's development and our customers in their journey, because I know why we need
59
00:04:10,040 --> 00:04:11,040
playbox.
60
00:04:11,040 --> 00:04:14,400
I know the benefit of standardization on how we negotiate contracts.
61
00:04:14,400 --> 00:04:19,080
I know how to understand when I'm looking at a red line, what matters and what doesn't
62
00:04:19,080 --> 00:04:24,920
matter and help our customers to the way I think of it is when you use a tool like Diopter
63
00:04:24,920 --> 00:04:30,000
Room, you can spend a lot more time on the strategy of contracting and a lot less time
64
00:04:30,000 --> 00:04:31,760
on the grunt work of contracting.
65
00:04:31,760 --> 00:04:36,800
So when you don't have to spend all of your time taking, you're fine, that precedent that
66
00:04:36,800 --> 00:04:40,880
has that clause that you really like to one time and you spend two hours looking for
67
00:04:40,880 --> 00:04:44,080
it and then you find it and you're like, oh, that Ashley isn't quite right.
68
00:04:44,080 --> 00:04:46,000
That's not a good use of our time.
69
00:04:46,000 --> 00:04:49,920
But if you can take your precedence, like we have people be able to do them, no, let me
70
00:04:49,920 --> 00:04:55,480
just need a couple and synthesize them into a playbook, then you spend where I think the
71
00:04:55,480 --> 00:04:59,640
real value of the capital, what lawyers bring to the table, which is looking at those
72
00:04:59,640 --> 00:05:04,120
rules that come out of your precedent saying, I do want that in my contracts or I don't
73
00:05:04,120 --> 00:05:07,480
want them in my contracts or I want this in my contracts, but actually I want it with
74
00:05:07,480 --> 00:05:08,960
this variation.
75
00:05:08,960 --> 00:05:10,480
And that's, to me, that's the value.
76
00:05:10,480 --> 00:05:12,720
The value isn't the red line.
77
00:05:12,720 --> 00:05:18,200
The value is what you decide matters to be redlined in the contract.
78
00:05:18,200 --> 00:05:23,360
And so I'm able to help our customers on that journey of the self discovery of what you want
79
00:05:23,360 --> 00:05:25,080
your contracts to be.
80
00:05:25,080 --> 00:05:28,320
That's part of what you do when you say, okay, I'm not going to do the red line.
81
00:05:28,320 --> 00:05:32,360
I'm going to have someone else do the red line in the AI, do the red lining, but I need
82
00:05:32,360 --> 00:05:33,960
to tell the AI what to do.
83
00:05:33,960 --> 00:05:39,240
>> What's the difference between rag and agents when we're talking about AI?
84
00:05:39,240 --> 00:05:46,200
But lawyer was describing is a very good example of what agents do well, right?
85
00:05:46,200 --> 00:05:55,240
So our agents are able to go through those precedents, just still the intent, and then
86
00:05:55,240 --> 00:06:03,320
come up with that plan, if you will, to actually be able to build that playbook and implement
87
00:06:03,320 --> 00:06:05,800
it consistently and accurately, right?
88
00:06:05,800 --> 00:06:09,000
And so agents, and I know there are multiple definitions out there, but
89
00:06:09,000 --> 00:06:15,600
agents are, or AI agents are those entities that we be able to take more than one step.
90
00:06:15,600 --> 00:06:21,000
We'll be able to take multiple steps, usually we use different models to achieve every
91
00:06:21,000 --> 00:06:22,760
single one of those steps.
92
00:06:22,760 --> 00:06:29,400
As opposed to rag, rag is more, you have a database and, or you have some amount of data,
93
00:06:29,400 --> 00:06:32,080
even if it is contracts or database and so on and so forth.
94
00:06:32,080 --> 00:06:37,800
And you just want to search and find something from that collection of information.
95
00:06:37,800 --> 00:06:40,360
And so the approach is fundamentally different.
96
00:06:40,360 --> 00:06:45,680
In one step, you're focusing on a process that you're trying to achieve on the other step,
97
00:06:45,680 --> 00:06:49,560
it's more like a search or like retrieval of information, which is what rag stands for,
98
00:06:49,560 --> 00:06:50,560
right?
99
00:06:50,560 --> 00:06:55,760
And so being able to search and retrieve information from a collection of data that you already
100
00:06:55,760 --> 00:06:57,680
have somewhere.
101
00:06:57,680 --> 00:07:02,840
>> When you want to stop having legal be directory assistance in your in-house, instead you want
102
00:07:02,840 --> 00:07:07,560
people to just go to a source for that information that they're looking for, rag is fantastic
103
00:07:07,560 --> 00:07:12,440
for that because it's taking the information that you've created and it's a library and you
104
00:07:12,440 --> 00:07:15,320
really just want people to get the right answer from the library.
105
00:07:15,320 --> 00:07:19,640
Whereas for RedLine, in contract, it's an action, not information.
106
00:07:19,640 --> 00:07:25,320
You need that, you want a red line that is like a lawyer with red line.
107
00:07:25,320 --> 00:07:28,000
And rag alone can't do that.
108
00:07:28,000 --> 00:07:34,560
>> I'm always amazed by seeing all the excitement about AIH and
109
00:07:34,560 --> 00:07:40,880
today when we started doing this, we launched our product, it was May last year.
110
00:07:40,880 --> 00:07:46,040
And we launched it with, we sent Sincini back then and when we started talking about agents,
111
00:07:46,040 --> 00:07:48,120
everybody was like, what are you talking about?
112
00:07:48,120 --> 00:07:49,120
What are agents?
113
00:07:49,120 --> 00:07:52,320
And it's like nobody, it wasn't a thing.
114
00:07:52,320 --> 00:07:54,760
It was like, everybody was like into rag and everything.
115
00:07:54,760 --> 00:07:58,720
And it was like that agents is so much more powerful because it allows you to do more complex
116
00:07:58,720 --> 00:07:59,720
things.
117
00:07:59,720 --> 00:08:04,520
And now everybody is asking about agents and these agents in every single operational
118
00:08:04,520 --> 00:08:06,760
and enterprise and industry, right?
119
00:08:06,760 --> 00:08:11,960
It's just in the legal space, they are extremely powerful when dealing with complex,
120
00:08:11,960 --> 00:08:14,040
multi-step processes.
121
00:08:14,040 --> 00:08:19,960
>> Lori, what challenges are you seeing around the adoption of AI?
122
00:08:19,960 --> 00:08:22,720
>> I think there are two big challenges.
123
00:08:22,720 --> 00:08:23,920
One is trust.
124
00:08:23,920 --> 00:08:27,880
When nobody really knew how to use GenNRVI when it first came out.
125
00:08:27,880 --> 00:08:31,920
And so people made a lot of mistakes because they thought it would be like Google search,
126
00:08:31,920 --> 00:08:36,920
it's pulling things that exist, but no, it's generating, that's why it's called GenNRVI,
127
00:08:36,920 --> 00:08:37,920
right?
128
00:08:37,920 --> 00:08:39,280
And so we didn't use it right.
129
00:08:39,280 --> 00:08:43,640
And a lot of people got hung out to drive because of errors that they made.
130
00:08:43,640 --> 00:08:48,440
And so there's fear around thinking AI can't be trusted.
131
00:08:48,440 --> 00:08:51,000
I think that's not the case anymore.
132
00:08:51,000 --> 00:08:54,800
Most of the tools that are out there have conquered the hallucination issue to a large
133
00:08:54,800 --> 00:08:55,800
degree.
134
00:08:55,800 --> 00:08:58,240
It still happens, but it's not the Wild West anymore.
135
00:08:58,240 --> 00:09:05,080
And I think the other big issue is there's just so much noise, right?
136
00:09:05,080 --> 00:09:06,080
You can't.
137
00:09:06,080 --> 00:09:08,680
No one's going to the market without saying they have AI today.
138
00:09:08,680 --> 00:09:10,640
Everybody has AI.
139
00:09:10,640 --> 00:09:13,200
And in some cases it's really good AI.
140
00:09:13,200 --> 00:09:16,360
And in other cases it's maybe not such great AI.
141
00:09:16,360 --> 00:09:22,000
And it's finding out, figuring out, like, and I'm sorry, I'll just, a lot of the in-house
142
00:09:22,000 --> 00:09:24,160
departments and firms have this mandate adopt AI.
143
00:09:24,160 --> 00:09:27,360
And we're seeing a huge sea change in that before it was experimented.
144
00:09:27,360 --> 00:09:28,920
And now it's adopt.
145
00:09:28,920 --> 00:09:31,560
But that's not how you solve problems.
146
00:09:31,560 --> 00:09:37,320
On the lunch day, Meredith talked about how you need to figure out what your problem is.
147
00:09:37,320 --> 00:09:38,720
And then go look for the solution.
148
00:09:38,720 --> 00:09:41,320
And a lot of people are coming to the table saying, "I need AI."
149
00:09:41,320 --> 00:09:43,600
And they haven't figured out what their problem is.
150
00:09:43,600 --> 00:09:47,440
And so they're not necessarily selecting tools that are the right tool to solve their problems
151
00:09:47,440 --> 00:09:49,560
because they're not starting from what's my problem.
152
00:09:49,560 --> 00:09:50,560
And I think that's true.
153
00:09:50,560 --> 00:09:52,320
It's not new to Jeremy AI, right?
154
00:09:52,320 --> 00:09:56,200
That is a problem with technology adoption that it's timeless.
155
00:09:56,200 --> 00:10:01,000
It's just that there's more ways, I think, with AI than there was with other tools.
156
00:10:01,000 --> 00:10:06,000
>> Parallel, what type of AI are buyers most interested in?
157
00:10:06,000 --> 00:10:10,440
>> It's only that I was actually reading and listening to a podcast, another podcast from
158
00:10:10,440 --> 00:10:13,040
the OIC community.
159
00:10:13,040 --> 00:10:19,040
And there was a lot of talk about AI agents and specifically in the enterprise space,
160
00:10:19,040 --> 00:10:24,120
like the B2B, not SAS, B2B agents software in the enterprises.
161
00:10:24,120 --> 00:10:28,560
And that is the range of applications that we hear about is tremendous.
162
00:10:28,560 --> 00:10:36,400
It is any type of operational task or process, which is exactly the sweet spot of agents, right?
163
00:10:36,400 --> 00:10:44,400
Any operational process that enterprises have to do again and again is now being automated
164
00:10:44,400 --> 00:10:45,400
or semi-automated.
165
00:10:45,400 --> 00:10:52,040
There's always a human interloop that is supervising what the agent is doing.
166
00:10:52,040 --> 00:10:55,920
Then imagine every single process that enterprises have to do, right?
167
00:10:55,920 --> 00:10:59,120
And every single one of those are now getting agents for.
168
00:10:59,120 --> 00:11:05,240
And there's a huge amount of adoption and demand for those tools if the tool is actually working.
169
00:11:05,240 --> 00:11:06,480
And I think that's a challenge.
170
00:11:06,480 --> 00:11:09,760
Like Neuys said, there's a lot of noise out there today.
171
00:11:09,760 --> 00:11:12,840
And it is difficult, often here customers that come to me.
172
00:11:12,840 --> 00:11:16,920
And they're like, I tried XYZ and I tried ABC and I tried, and it's not working.
173
00:11:16,920 --> 00:11:18,960
And why would you be able to do that differently?
174
00:11:18,960 --> 00:11:19,960
And I get it, right?
175
00:11:19,960 --> 00:11:23,160
I get it, everybody is playing with AI.
176
00:11:23,160 --> 00:11:28,400
It's hard to understand what is actually going to be working from a biased perspective.
177
00:11:28,400 --> 00:11:32,280
But if the tool works, I've seen a significant adoption.
178
00:11:32,280 --> 00:11:36,440
Neuys is what comes, like what you're doing is has nothing to do with what I've seen.
179
00:11:36,440 --> 00:11:38,920
And it's extremely different and the quality is different.
180
00:11:38,920 --> 00:11:41,320
And I think that is true for not just the legal space.
181
00:11:41,320 --> 00:11:46,440
I think that's true for all the different verticals where agents are being used.
182
00:11:46,440 --> 00:11:51,520
Laurie, what's the challenge for lawyers to secure buy-in for new AI tools?
183
00:11:51,520 --> 00:11:57,480
>> I think this is different from any other legal technology because AI is not a legal tool.
184
00:11:57,480 --> 00:12:00,280
AI is everything tool.
185
00:12:00,280 --> 00:12:01,600
It's like computers, right?
186
00:12:01,600 --> 00:12:05,200
When companies adopt computers, it was only like lawyers had to make a business case for
187
00:12:05,200 --> 00:12:06,000
computers.
188
00:12:06,000 --> 00:12:09,080
It was the company saying to lawyers, and now you have to use computers.
189
00:12:09,080 --> 00:12:13,600
And I think that's the same thing happening with AI that lawyers are being told you have to use AI.
190
00:12:13,600 --> 00:12:19,560
And I think the problem with adoption is and buy-in is really a mix of any time there's change.
191
00:12:19,560 --> 00:12:27,320
It's scary and AI specifically taking on work is scary because it's is this my job?
192
00:12:27,320 --> 00:12:33,200
And then some tools really require lawyers to become prompt engineers, which I think lawyers
193
00:12:33,200 --> 00:12:39,360
can be very good prompt engineers because it's manipulation of language, which is what we all do as lawyers.
194
00:12:39,360 --> 00:12:42,840
But I compare it to the first time I tried to write a patent claim.
195
00:12:42,840 --> 00:12:46,960
I met who are laughed at me because it was so bad it was written in English.
196
00:12:46,960 --> 00:12:49,040
It just wasn't right for a pen.
197
00:12:49,040 --> 00:12:50,040
And so it's the same thing.
198
00:12:50,040 --> 00:12:53,920
If you could write a contract that's and then you try to write a rule for a contract that's
199
00:12:53,920 --> 00:12:56,120
a prompt, it's not going to be quite right.
200
00:12:56,120 --> 00:12:57,720
You have to learn how to do it.
201
00:12:57,720 --> 00:13:03,920
And so the tools that require their customers to master that art without giving them assistance,
202
00:13:03,920 --> 00:13:09,840
I think is asking lawyers to learn a brand new tool set.
203
00:13:09,840 --> 00:13:14,360
How does diopter differ from other contracting tools in the market?
204
00:13:14,360 --> 00:13:20,800
I've been building AI products for the past 10 years and the challenge is always being
205
00:13:20,800 --> 00:13:25,920
the trade-off between how to navigate your tool is so how much can it handle right versus
206
00:13:25,920 --> 00:13:29,600
what is the accuracy that you can achieve right?
207
00:13:29,600 --> 00:13:34,280
And there's a trade-off because the more analyst your AI is less accurate it is at doing every
208
00:13:34,280 --> 00:13:39,960
single thing which is understandable right.
209
00:13:39,960 --> 00:13:43,360
We have been focusing on that trade-off since day one because we knew we needed to be able
210
00:13:43,360 --> 00:13:47,640
to tackle a number of different types of contracts right.
211
00:13:47,640 --> 00:13:52,480
And we knew also that accuracy matter like the quality of the red lines that land on that
212
00:13:52,480 --> 00:13:56,720
I'm not going to say piece of paper but on that contract is important.
213
00:13:56,720 --> 00:14:02,720
It's because we have started from that we have been able to build an agent in a way that
214
00:14:02,720 --> 00:14:08,720
is extremely accurate as compared to a lot of tools out there when building the red lines
215
00:14:08,720 --> 00:14:13,720
or when writing the red lines in a way that it is surgical in a way that it is elegant
216
00:14:13,720 --> 00:14:17,360
and then similar to what humans would do right.
217
00:14:17,360 --> 00:14:26,120
And then we use all those playbooks to guide and make our agent generalizable right.
218
00:14:26,120 --> 00:14:30,240
And that's how we are able to tackle different types of contracts by helping our customers
219
00:14:30,240 --> 00:14:35,040
build those are people from their presidents and distilling that knowledge from those presidents.
220
00:14:35,040 --> 00:14:39,640
And I'll say our agent is smart enough that it keeps schooling me.
221
00:14:39,640 --> 00:14:41,640
Laurie how do you see contracting evolving?
222
00:14:41,640 --> 00:14:43,920
Oh it's going to be so different.
223
00:14:43,920 --> 00:14:50,680
So right now I would say the majority of companies contracts go through every contract goes
224
00:14:50,680 --> 00:14:57,680
to the legal team or contracting team and it's a lot of tasks switching.
225
00:14:57,680 --> 00:15:01,120
So you have to look at the playbook or look at the guidelines or look at the checklist and
226
00:15:01,120 --> 00:15:03,800
look at the document and go back and forth and back and forth and send it to the team and
227
00:15:03,800 --> 00:15:09,480
send it to that team and there's just a lot of administration around the contracting process.
228
00:15:09,480 --> 00:15:17,180
And I think that AI is going to allow us our CTO of the other days that the gift of AI
229
00:15:17,180 --> 00:15:19,080
allows us to elevate our thinking.
230
00:15:19,080 --> 00:15:21,000
We don't have to be in the weeds anymore.
231
00:15:21,000 --> 00:15:24,160
We can think strategically spend more time thinking strategically.
232
00:15:24,160 --> 00:15:26,120
And I think that's what we're going to be able to do.
233
00:15:26,120 --> 00:15:30,520
We're going to be able to spend a lot more time thinking about the strategic aspects of
234
00:15:30,520 --> 00:15:36,040
contracting and a lot less time doing the mundane administrative aspects of contracting.
235
00:15:36,040 --> 00:15:42,720
This is Ari Kaplan speaking with Laurie Erlich, the chief legal officer and Farragasmi,
236
00:15:42,720 --> 00:15:47,840
the co-founder and chief product officer at Deoctra, an AI powered contract redlining
237
00:15:47,840 --> 00:15:50,080
and playbook generation tool.
238
00:15:50,080 --> 00:15:52,440
Laurie Farrah thank you so very much.
239
00:15:52,440 --> 00:15:53,440
Thank you Ari.
240
00:15:53,440 --> 00:15:54,440
Thank you.
241
00:15:54,440 --> 00:15:59,000
Thank you for listening to the reinventing professionals podcast.
242
00:15:59,000 --> 00:16:05,400
Visit reinventingprofessionals.com or Ari Kaplan advisors.com to learn more.
