Checking In on Legal’s AI Transformation: A Matter of Perspective

1
00:00:02,254 --> 00:00:06,144
Today I'm joined by Findley Penn Hughes, who is counsel at Mayer Brown.
2
00:00:06,844 --> 00:00:13,554
For over a year now, Findley and I have been on separate but parallel paths in our AI journey.
3
00:00:14,364 --> 00:00:22,254.001
He as a compliance and white-collar attorney at a white shoe law firm, me as an antitrust specialist in-house at a Global five hundred.
4
00:00:23,184.001 --> 00:00:32,684
Today's episode is our attempt to check in on where we are individually in our legal practice and more generally across the legal services sector.
5
00:00:33,414 --> 00:00:40,354
What we really want to look at is how are we doing with injecting AI into day-to-day legal practice.
6
00:00:40,993.998 --> 00:00:46,424
This conversation plays out much like it does when Findley and I meet up in real life and talk legal tech.
7
00:00:46,484 --> 00:00:55,824
So we trade war stories, share our frustrations and triumphs, and try to unpack why what works does and what doesn't doesn't.
8
00:00:56,823.996 --> 00:01:11,273.994
In the process, we end up exploring the limitations of generative AI and the various business and practical barriers that help explain the methodical approach to AI deployment that we're seeing among lawyers industry-wide.
9
00:01:12,353.995 --> 00:01:29,583.998
It was a conversation that helped me ground my assessment of how well or not AI is doing and helped me calibrate my expectations for how quickly we can expect to see legal go to the frontier of the large language models with things like agentic AI.
10
00:01:29,943.992 --> 00:01:34,3.998
I always enjoy vibing with Findley, and I hope you enjoy the episode.
11
00:01:58,415.99848525 --> 00:01:58,885.99848525
Hi, Findley.
12
00:01:58,905.99848525 --> 00:01:59,895.99848525
Welcome to Version Up.
13
00:02:00,725.99848525 --> 00:02:01,185.99848525
Happy to be here.
14
00:02:01,895.99848525 --> 00:02:21,975.99648525
You are a fellow lawyer, practicing lawyer on the law firm side, and a good friend, and I'm really looking forward to just talking about where we are with AI and legal practice, and that sounds open-ended, but I actually think, we need to stop and pause as lawyers- Yeah and assess where we're at, how we got here, what's working, what's not.
15
00:02:22,15.99948525 --> 00:02:26,205.99848525
Because if we don't, I'm a little worried, people are gonna get lost in the forest.
16
00:02:26,215.99648525 --> 00:02:27,956.00148525
So, thank you for joining.
17
00:02:28,956.00248525 --> 00:02:31,895.99848525
No, as I say, very happy to be here, and I feel like it's been a long time coming.
18
00:02:31,896.00148525 --> 00:02:52,45.99848525
You and I have been discussing this kind of stuff from, you know, months, I mean, a year at least now, right? Um, and as you say, like, reflecting on, on, on, on my use of it before this, before this call,, it's not actually that long ago that I s- that we started using this stuff in the grand scheme of things, you know? And, like, especially in the terms of our whole career and how far we've come.
19
00:02:52,495.99648525 --> 00:02:52,845.99648525
Yeah.
20
00:02:53,415.99448525 --> 00:02:57,236.00148525
That's the same exact reaction I have in looking back at my timeline.
21
00:02:57,665.99748525 --> 00:03:02,685.99248525
Just to kind of l- give us some context, I, I looked up the date for GPT 3.5.
22
00:03:02,685.99248525 --> 00:03:02,705.98948525
Yeah.
23
00:03:03,445.98648525 --> 00:03:06,485.99048525
November 30th, 2022.
24
00:03:07,921.98948525 --> 00:03:16,11.98948525
Harvey's Series A, which is really sort of when they, you know, that's when they really raised the money and, started, really landing serious accounts, landing law firm accounts.
25
00:03:16,481.98848525 --> 00:03:19,421.98848525
That was only f- like four months later.
26
00:03:19,431.98948525 --> 00:03:19,741.98848525
Mm-hmm.
27
00:03:19,741.98948525 --> 00:03:20,921.98948525
April 2023.
28
00:03:21,431.98948525 --> 00:03:24,301.98948525
April 2023, Harvey had their Series A.
29
00:03:24,811.98848525 --> 00:03:38,901.98948525
And then GPT Last Date, GPT-4, which launched, uh, and they also, that was when it scored in the 90th percentile on our uniform bar exam, was March 2023.
30
00:03:38,901.98948525 --> 00:03:39,421.99048525
Yeah.
31
00:03:39,421.99048525 --> 00:03:44,351.98848525
So, so spring of 2023, we have AI legal tools in the market.
32
00:03:45,61.98948525 --> 00:03:51,381.98748525
And then I'll give you my personal timeline, which was, um, it's embarrassing 'cause I have a legal tech podcast, but let's put it out there.
33
00:03:52,611.98948525 --> 00:04:03,201.98848525
I, when I think about it critically, I would say AI was not a part of my legal practice at all, like my actual legal practice, until early 2025.
34
00:04:03,551.98748525 --> 00:04:03,961.98948525
Mm-hmm.
35
00:04:04,111.98948525 --> 00:04:13,911.99048525
And if I were to be really self-critical and say, when was it a daily part of my practice, I would say summer or fall of 2025.
36
00:04:14,11.99048525 --> 00:04:24,841.98848525
So we're talking, I'm only about a year and change, year and a half if we're being generous, into really having AI embedded in my practice in some capacity.
37
00:04:25,481.98748525 --> 00:04:31,911.98748525
And so to read back something you just said, essentially, um, it hasn't been that long.
38
00:04:32,761.98648525 --> 00:04:32,861.98648525
No.
39
00:04:33,691.98648525 --> 00:04:41,51.99148525
No, exactly, and I think for me, I was again, I was trying to, I had, it's something I hadn't really done until you and I spoke about doing this, about reflecting on when I started using it.
40
00:04:41,801.98748525 --> 00:04:49,271.99048525
I mean, the, you, the date that you said the release of, of, of GPT to like the general, the general consumer is probably the...
41
00:04:49,701.98848525 --> 00:04:52,841.98848525
I started using it as, you know, personally from the day it was released.
42
00:04:53,411.98748525 --> 00:04:58,841.99048525
And then I don't think it was that long after that that I had tried to start using it in, in my private practice.
43
00:04:59,131.98948525 --> 00:05:00,171.99048525
I don't, I can't remember.
44
00:05:00,211.99048525 --> 00:05:02,91.98548525
I can, I should have, like I could probably look it up.
45
00:05:02,91.98848525 --> 00:05:10,851.99048525
I can't remember when we got Harvey, but we were certainly encouraged to use ChatGPT as long as it wasn't client information from the early stages.
46
00:05:11,551.98748525 --> 00:05:20,521.98448525
And I think the problem, like, you know, you, you can say there's been lots of discussion about this, about whether or not GPT released too early, that, you know, that, that first model.
47
00:05:21,111.98448525 --> 00:05:39,557.99148525
'Cause I re- 'cause I think it might have for certain people, and maybe lawyers in particular, might have caused more harm than good to release, perhaps to have released it too early, right? I remember the first time I tried to use it, it was writing, this is when I was still in the UK, so that would probably ear- it was probably early 2023 And I had to write, like, one of these client alerts.
48
00:05:39,587.99148525 --> 00:05:44,287.99148525
I think it was something, some Bank of England thing that was relevant to our clients.
49
00:05:44,557.99148525 --> 00:05:57,957.99148525
And so I asked And so the Bank of England had released some, I don't know, 96-page report, and I just, I sort of thought, "Well, maybe it can distill this down and save me having to read the 96 pages, and I can just have the important bits and then start my draft from there."
50
00:05:59,197.99148525 --> 00:06:05,337.99048525
So I get it to do that, and then you, you read the, you read the summary it produces, and you go, "Well, that's interesting.
51
00:06:06,247.99048525 --> 00:06:10,297.99048525
Can you, can you give me I would like to read more about what, you know, the, the, the detailed bit about that.
52
00:06:10,517.99048525 --> 00:06:15,777.98948525
Can you give me a page reference of where you say this?" I, and then very confidently it says, "Page 96."
53
00:06:16,757.98848525 --> 00:06:24,817.98848525
So you go to page 96, and you can't find it anywhere, and then you say, "Well, can you give me an exact quote so I can, you know, Control F and find the, the sentence?" Gives you an exact quote very confidently.
54
00:06:25,327.99248525 --> 00:06:25,917.98948525
Can't find it.
55
00:06:26,447.99148525 --> 00:06:36,867.99048525
And so then, so then you obviously realize that this hallucination problem is very, very real, and I think for most lawyers And all the bigger problems for hallucinations, case hallucinations and so on, I think came out a lot later.
56
00:06:37,237.98848525 --> 00:06:57,727.99148525
But certainly anecdotally the people that I was working with in the office would say, you know, immediately didn't like to use it because it, it ends up, it ended up creating more work for you Yeah, by the time you've done that, you've put all this work into, into the prompts and so on, and there's follow-up prompts, and you may as well just have read the, read the damn document initially, right? Um- I, I, I totally agree.
57
00:06:57,757.99148525 --> 00:07:01,667.99048525
And I think it's a good way to split this discussion up a little bit is between those two time periods.
58
00:07:01,687.99048525 --> 00:07:05,147.99148525
Because for me, there's, from the timeline, there's two interesting takeaways.
59
00:07:05,187.99148525 --> 00:07:13,304.65781859
The first is that there is this, like, two-year gap from, let's call it ChatGPT 3.5,
60
00:07:13,304.65781859 --> 00:07:13,347.99148525
3.-
61
00:07:13,347.99148525 --> 00:07:14,707.98948525
Yeah uh, 4.0,
62
00:07:14,957.99048525 --> 00:07:16,247.99148525
Harvey's in there somewhere.
63
00:07:16,907.99148525 --> 00:07:22,567.99248525
Between that and, my daily practice, not to mention the average lawyer's, my daily practice, it's a two-year gap.
64
00:07:22,567.99248525 --> 00:07:23,247.98948525
That's a long time.
65
00:07:23,737.99148525 --> 00:07:48,827.99248525
And I think you're getting at exactly at that topic, which is why was there that gap? I, I think part of it was there was some goodwill and trust lost early on with an audience, legal users, lawyers, an audience that, uh, is very kind of, for good reasons, I think, risk-averse and careful- Mm-hmm methodical, and they went and played with it and, and they were like, "Oh, I can't, I can't put client data in this.
66
00:07:48,827.99248525 --> 00:07:51,877.98948525
I can't trust this to give me a, a good output."
67
00:07:52,347.98748525 --> 00:07:53,227.99248525
And so they put it down.
68
00:07:53,657.99348525 --> 00:07:57,517.99248525
And they probably put it down for too long, but they did put it down- Agreed I think for good reasons.
69
00:07:57,517.99348525 --> 00:08:03,627.98748525
And I, I will say that I did, and I kept checking back in and, and really, and this goes to the second part of the timeline.
70
00:08:04,737.98748525 --> 00:08:09,887.99248525
Okay, what happened at that point when I really started using it? Well, that's when I have Copilot at work.
71
00:08:10,177.98948525 --> 00:08:15,577.99148525
That's when ChatGPT/Copilot, um, for the Microsoft 365 Suite.
72
00:08:16,57.98748525 --> 00:08:19,787.99148525
That's when I was like, "Okay, this feels like hallucinations."
73
00:08:19,797.99248525 --> 00:08:19,817.99248525
Yeah.
74
00:08:20,287.99048525 --> 00:08:21,577.99248525
Feel like they're fully come down.
75
00:08:22,177.98948525 --> 00:08:27,547.99348525
The traceability has gotten better, and my trust in it is getting stronger.
76
00:08:28,137.98848525 --> 00:08:29,377.99148525
That's kind of when it clicked for me.
77
00:08:29,987.98748525 --> 00:08:30,497.98748525
Yeah.
78
00:08:31,77.99148525 --> 00:08:33,577.99048525
And I think for me, again, it was about maybe 18 months.
79
00:08:33,587.99348525 --> 00:08:38,877.98948525
So this was, so my, that experience I j- I just described was, um, early 2023, I think.
80
00:08:39,267.99348525 --> 00:08:39,497.99348525
Yeah.
81
00:08:39,597.98748525 --> 00:08:50,147.98748525
And then I moved to New York in November '24, and I think j- ba- basically just prior to that, I had started using Harvey on a regular basis.
82
00:08:50,497.98748525 --> 00:08:53,307.98848525
So in between that time, I kind of picked it up for work now and then.
83
00:08:53,377.98648525 --> 00:08:55,927.99348525
I hadn't really liked it and had basically stopped using it.
84
00:08:55,927.99348525 --> 00:09:01,279.99248525
I used it in my- Used it in my personal life and whatever, and it kind of is a bit of a gimmick certainly in the early days.
85
00:09:01,279.99248525 --> 00:09:10,49.99248525
You know, you could write you a sonnet in, about an avocado in the style of William Shakespeare or something, but you didn't really feel like it was gonna be something that you could trust to do, do the work.
86
00:09:10,49.99248525 --> 00:09:18,469.99248525
And then Harvey came along and, you know, people have their problems with Harvey, but f- certainly at my firm, I think it's pretty good.
87
00:09:18,579.99148525 --> 00:09:23,589.99148525
Um, I think the fact that you can select models, um, is definitely really good.
88
00:09:23,589.99148525 --> 00:09:33,349.99248525
We can probably talk about that a bit later, but, um, that was, that's the only model I'm able to use at work other than now, and that c- came a bit later, the Copilot integration.
89
00:09:33,889.99248525 --> 00:09:48,389.99048525
Um, so and then now I use it every day, every single day for various different tasks and in various different ways, both for like, you know, client work and for st- you know, to help me sort out administrative tasks and billing and so on at work as well.
90
00:09:48,879.99048525 --> 00:09:59,69.98948525
And that's, I think, the average lawyer's experience now- Mm which is pretty incredible if you, if you give it that actually, like, year-and-a-half timeline, it's, it makes it seem more remarkable to me.
91
00:09:59,569.98948525 --> 00:10:04,226.65681859
When I first started thinking about this topic, I was kind of more pessimistic how long it took lawyers to use AI.
92
00:10:04,226.65681859 --> 00:10:08,746.65481859
But then I looked at my own usage, and I looked at when has the ramp-up really happened.
93
00:10:08,746.65481859 --> 00:10:14,556.65581859
It's really been about a year and a half, and that's pretty- Mm-hmm to me, it's pretty incredible that, that much has happened.
94
00:10:15,36.65481859 --> 00:10:22,796.64981859
Now everyone still is at a different stage, and I feel like I'm somewhere in the middle to upper middle of the pack.
95
00:10:23,486.65181859 --> 00:10:26,526.65181859
Uh, I feel like from talking to you, you're a little more on the frontier.
96
00:10:27,696.65381859 --> 00:10:34,356.64781859
I would love to hear your journey, um, with OpenClaw and just sort of really diving into the agentic a- applications.
97
00:10:35,466.65081859 --> 00:10:40,576.65381859
I know you weren't doing that for work, but just for- Yeah your own personal kind of, like, toying around, understanding the product.
98
00:10:40,966.65081859 --> 00:10:49,26.64881859
Can you walk me through that experience and what you've learned from it, and where it leaves you thinking about agentic AI for, for legal? Yeah.
99
00:10:49,256.65081859 --> 00:11:05,106.64881859
And so, yeah, so the, I, as you and I have discussed before, I can, we both are interested in this stuff, like to play around with it, and then I don't, I don't know when it was, maybe whenever Open- OpenClaw came out, I decided I would run an instance of it in a little PC that I have in, uh, my closet here.
100
00:11:05,606.64981859 --> 00:11:13,876.65681859
Um, so I set that up about six months ago, and what I wanted it to do, and it d- did actually, was designed initially to kind of help me from a professional perspective.
101
00:11:13,876.65681859 --> 00:11:24,996.64781859
So I had, like, a sort of assistant that I'd communicate through Telegram, and then I also built, like, a little tr- like, stocks trading bot that I, that I traded on pa- just on, like, fake money.
102
00:11:25,506.64381859 --> 00:11:30,476.64581859
Um, but, like, to try and just to play around with and just to try and see what, what, what it could do.
103
00:11:30,726.64481859 --> 00:11:38,374.65281859
So that, um, so on the, on the assistant side, what I tried to get it to do was- It's a bit like this idea, I don't know, that you, you...
104
00:11:38,434.65281859 --> 00:11:45,984.65281859
The sci-fi films that we grew up with, you know, whether that's, like, as old as, like, 2000 Space, uh, 2001 Space Odyssey or Her or all these kind of things.
105
00:11:45,984.65281859 --> 00:11:50,224.65281859
I had this, like, really idea that we'd have this, like, little thing in my pocket that would do all my tasks for me.
106
00:11:50,654.65281859 --> 00:12:07,504.65181859
Um, but yeah, I set it up to scan, to basically scan the web for any interesting developments, so whether that's like DOJ, SEC, uh, FINRA, UK regulator, like, developments that I might write a, a, like, a client alert on, right? And there's always the first mover advantage for this stuff.
107
00:12:07,504.65181859 --> 00:12:10,374.65281859
The partners alw- always wanna see that we're doing it su- super quick.
108
00:12:10,884.65281859 --> 00:12:18,24.65081859
Um, so it would send me these alerts, and then I could, in theory, respond and say, "Okay, well, like, development two sounds really good.
109
00:12:18,24.65281859 --> 00:12:18,764.65181859
Like, why don't you..."
110
00:12:18,844.65181859 --> 00:12:28,444.64981859
And it, it had in its, like, uh, context a bunch of example client report, client alerts, client articles I'd written before, so it knew what the style was.
111
00:12:29,74.65181859 --> 00:12:30,34.65381859
So that was the theory.
112
00:12:30,894.65181859 --> 00:12:37,224.65081859
It took me about, I don't know, about a month to get all of the programming right with Claude, with Claude Code, and even then it was never quite right.
113
00:12:37,754.65181859 --> 00:12:40,414.65281859
Um, and then for, for some reason it wouldn't...
114
00:12:41,814.65381859 --> 00:12:51,814.65381859
The developments it was sending me, even though it was scanning daily, were always a few days old, even though in theory it should have been scanning just, like, the, the, the news.
115
00:12:52,324.64781859 --> 00:12:58,194.64981859
So I never really got to the bottom of why they were always a few days old, so that didn't really work that well in the end.
116
00:12:58,344.65181859 --> 00:13:00,294.65181859
Although the drafting wasn't bad, and it would then...
117
00:13:00,894.64781859 --> 00:13:02,644.64881859
I'd set it up so it would, it would put a...
118
00:13:02,744.65381859 --> 00:13:11,84.65081859
it would draft a, uh, an article and then email it to my personal account, and I could then forward it on to my work account if I wanted and edit, edit it from there.
119
00:13:11,514.65081859 --> 00:13:23,584.65081859
Um, so the drafting was pretty good, and I think we found that with, certainly with, like, the, the later versions of Claude, particularly if you give it an example of, or a few examples of the writing that you like or your style.
120
00:13:24,24.65281859 --> 00:13:34,410.65281859
It kind of avoids all that Claude-ese, um, you know, like, very, very, I don't know- wording that's, like, is very i- illustrative of what, of, of Claude.
121
00:13:34,410.65281859 --> 00:13:37,720.65281859
I find it, like, it speaks in a certain way, right? I don't know if you- It's a tell I don't know if you've noticed it.
122
00:13:37,850.65281859 --> 00:13:38,150.65281859
Yeah.
123
00:13:38,150.65281859 --> 00:13:38,640.65281859
It is a tell.
124
00:13:38,830.65281859 --> 00:13:38,860.65281859
Yeah.
125
00:13:38,860.65281859 --> 00:13:40,570.65281859
It, it likes to use the word quietly.
126
00:13:40,590.65281859 --> 00:13:43,920.65281859
Have you ever not- have you noticed this? It always says like- Quietly? Quietly.
127
00:13:43,920.65281859 --> 00:13:45,960.65181859
So- Like, not loud, but quietly? Yeah.
128
00:13:46,0.65281859 --> 00:13:46,630.65281859
Yeah, quietly.
129
00:13:46,630.65281859 --> 00:13:51,910.65281859
But as in, like, it'll go like, "Oh, this such and such it is, like, that is quietly failing," or like, it just seems to...
130
00:13:52,190.65281859 --> 00:14:00,660.65181859
I don't know whether, 'cause I think it sounds like, I think it think it sounds, it thinks it sounds, uh, uh, authoritative or impressive or something to say.
131
00:14:00,700.65181859 --> 00:14:02,860.65181859
So I always have to try and avoid, uh, avoid it doing that.
132
00:14:02,860.65181859 --> 00:14:06,280.65181859
And then obviously it- No, this is, this is 'cause you're at a white shoe law firm.
133
00:14:06,290.65381859 --> 00:14:09,140.65181859
It's trying to impress you and bring- come up to your level.
134
00:14:09,520.65281859 --> 00:14:11,490.65081859
I'm here in the trenches, in-house lawyer.
135
00:14:11,490.65081859 --> 00:14:14,180.65681859
It talks to me like I'm just another business person.
136
00:14:15,650.65381859 --> 00:14:23,450.65581859
Um, so yeah, so that was the theory, and I've actually pulled the plug on, on the, the OpenClaude thing recently because it was just wasn't really working for me.
137
00:14:23,450.65681859 --> 00:14:30,570.65481859
It had, wasn't, had- I hadn't quite got it right, and I couldn't be bothered to waste another six hours of the weekend poking around with Claude code to get it right.
138
00:14:30,910.65581859 --> 00:14:38,730.65781859
So I'm k- trying to reset it up, but just as a, um, but just powered by Claude rather than by the OpenClaude.
139
00:14:38,780.65881859 --> 00:14:43,60.65781859
Um- Did you, did you make any fake money, by the way, on the stock trading? The fake stock trading.
140
00:14:43,60.65781859 --> 00:14:49,380.65681859
I think it, I think I had, I, I made, on a $500 fake pot, I made $13 in three months.
141
00:14:50,110.65281859 --> 00:14:50,630.65281859
But that's all right.
142
00:14:50,630.65781859 --> 00:14:51,460.65781859
You know, it's not a loss.
143
00:14:51,510.65481859 --> 00:14:54,330.65381859
You take that, a couple of coffees, or maybe one coffee in New York.
144
00:14:54,330.65481859 --> 00:14:55,160.65181859
I don't know.
145
00:14:55,160.65581859 --> 00:14:59,600.65681859
Imagine if that, imagine if, if the agent had some skin in the game, if it was real money.
146
00:14:59,600.65681859 --> 00:14:59,670.65381859
Mm-hmm.
147
00:14:59,670.65381859 --> 00:14:59,680.65381859
Mm-hmm.
148
00:14:59,680.65381859 --> 00:15:01,550.65881859
Who knows? Maybe it didn't have the right incentive.
149
00:15:02,460.65881859 --> 00:15:07,360.65381859
So, so you moved over- Well, it was super- Yeah interestingly, it was super, super, um, cautious.
150
00:15:07,400.65381859 --> 00:15:08,840.65481859
It w- wouldn't deploy money.
151
00:15:08,850.65781859 --> 00:15:10,20.65581859
It kept being like, "I just wanna..."
152
00:15:10,20.65681859 --> 00:15:14,980.65481859
It kept feeling that it, I just wanna hold onto cash, and I kept telling it, "No, no, let's change the parameters.
153
00:15:15,0.65981859 --> 00:15:15,820.65481859
Don't hold onto the cash.
154
00:15:15,820.65981859 --> 00:15:17,140.65281859
Just deploy, deploy the money."
155
00:15:17,140.65881859 --> 00:15:17,840.65781859
It was very strange.
156
00:15:17,910.65581859 --> 00:15:18,470.65581859
I don't, I don't know.
157
00:15:18,530.65581859 --> 00:15:19,640.65281859
I never got to the bottom of why that was.
158
00:15:19,640.65881859 --> 00:15:23,370.65581859
So actually, inflation adjusted, I think your return might have been negative.
159
00:15:23,370.65581859 --> 00:15:23,460.65581859
Pro- probably.
160
00:15:23,600.65781859 --> 00:15:24,235.65431859
I hate to say.
161
00:15:24,235.65431859 --> 00:15:24,580.65381859
Probably.
162
00:15:24,580.65381859 --> 00:15:25,490.65881859
Yeah, probably.
163
00:15:33,140.54542701 --> 00:15:39,970.54442701
So you switched over from OpenClaw to Claude, and Claude is a more sort of comprehensive, full ecosystem.
164
00:15:40,10.54442701 --> 00:15:44,60.54442701
I use it- Mm-hmm in my everything version up and personal, I, I, I use Claude.
165
00:15:44,590.54442701 --> 00:15:52,570.54342701
How's that transition gone for you? Well, it was only like, I don't know, 10 days ago that I, I sort of started this to, to try and set it up again.
166
00:15:52,570.54342701 --> 00:15:58,770.54442701
So I copied over all of the like, essentially the logs and all the files I'd used for the OpenClaude instance, and I wanted it to...
167
00:15:58,880.54342701 --> 00:16:03,600.54142701
And actually, the OpenClaude instance was ultimately, uh, you know, powered by Claude.
168
00:16:03,740.54342701 --> 00:16:05,330.54442701
That was the API plugin I had under...
169
00:16:05,370.54442701 --> 00:16:09,180.54342701
Well, I had Claude and ChatGPT actually with different, different models for different tasks.
170
00:16:09,770.54242701 --> 00:16:35,536.54342701
Um, but yeah, so I tried to, I think I was setting, I tried to set it up- Last week, um, and one of the things, one of the reasons I decided not to use OpenClaw was 'cause you, I mean, you'd been reading about all of these, uh, potential security issues with OpenClaw, right? And I had some, you know, I was careful to, to not have, um, you know, exposed API keys and so on, but I don't know ultimately enough about that security aspect of it to be comfortable doing it myself.
171
00:16:35,606.54442701 --> 00:16:38,286.54442701
So I thought, well, let's shut it down and let's start again with Claude and see if we can...
172
00:16:38,616.54442701 --> 00:16:44,186.543427
And one of the things I was emphasizing to Claude Code was, well, let's make sure it's very secure.
173
00:16:44,546.544427 --> 00:16:53,156.542427
So I tried to set it up, and then, uh, it stop, tried to restart itself, um, but the entire computer was just broken.
174
00:16:53,996.543427 --> 00:16:56,66.544427
Um, and I now have to re...
175
00:16:56,116.542427 --> 00:17:04,666.542427
I don't really know h- what has happened to it, um, but it's missing a registry key, which is obviously a kind of, like a sort of fundamental underlying thing in your Windows.
176
00:17:04,666.542427 --> 00:17:17,656.543427
And then now, and now I don't actually own an external hard drive, so I've had to go and had to go out and buy one so I can load Windows 11, like ISO file onto it, and then like mount it onto to reinstall Windows.
177
00:17:17,676.543427 --> 00:17:29,196.546427
So yeah, just I guess a bit of a warning that, you know, these things can end in complete disaster, particularly when you're someone like me who's probably more of a enthusiastic amateur than a, like, expert in this kind of stuff.
178
00:17:29,746.543427 --> 00:17:38,16.544427
the amount of work that is c- created- Yes in trying to wrangle these tools to do what we want them to do is astonishing.
179
00:17:38,236.541427 --> 00:17:45,546.543427
Now, I appreciate that we're sort of the users that wanna play with this, and we enjoy doing it, and we wanna be some of the early adopters.
180
00:17:45,786.544427 --> 00:17:51,726.542427
So this is the pain that you and I accept- Mm for the reward that we enjoy uniquely.
181
00:17:52,236.541427 --> 00:17:56,346.543427
But the average user does not wanna go through this, does not care about this.
182
00:17:57,6.546427 --> 00:18:07,956.545427
Um, but at the same time, I feel like the way these solutions are marketed and the way that, uh, experts talk about them, they're always thinking about it on the frontier.
183
00:18:07,956.545427 --> 00:18:22,126.537427
So they, they've been talking about agentic AI now- Yeah for six months, and I've yet to see agentic AI deployed anywhere near the way that the experts were telling us it could be done in, in legal, in my day-to-day practice.
184
00:18:22,396.548427 --> 00:18:28,726.551427
do you share that view, or have you seen some success on that front in legal practice at least? I haven't seen it, um, any...
185
00:18:28,896.551427 --> 00:18:43,896.539427
I haven't seen it really being used successfully yet, but that's not to say I'm sure that there are some firms and there are some who are more at the frontier of this, and you have some of the, the, like, in-house AI, proper AI specialists, not just lawyers who've rebranded themselves as AI enthusiasts.
186
00:18:44,156.545427 --> 00:18:47,596.539427
Um, you know, n- n- nothing, not to knock those people, but I think there is a very...
187
00:18:47,836.538427 --> 00:18:57,26.549427
'Cause that would probably be someone like me, but there are some ve- you know- I think particularly for this agentic stuff, it's gonna be a lot more than just, oh, here's a consumer facing package where you can type a prompt in.
188
00:18:57,236.549427 --> 00:18:57,906.549427
It's, it, it...
189
00:18:57,946.549427 --> 00:19:06,266.549427
As you and I have found with our just basic experience of, of, of, of setting up, um, OpenClaw or whatever, it, it's complicated this stuff.
190
00:19:06,726.549427 --> 00:19:22,846.548427
Now, like obviously, um, Harvey has these workflows that you can set up, and the idea is that they're supposed to be fairly s- well, not, not easy, but they're supposed to be fairly intuitive to set up just within the Harvey interface on, on your PC, on your work PC.
191
00:19:22,846.548427 --> 00:19:30,586.548427
But then, um, and then there's some stuff at the back end I think you can get your Harvey, um, client contact or whatever to, to sort out.
192
00:19:30,586.548427 --> 00:19:34,186.549427
But we, we are increasingly looking into trying to do some.
193
00:19:34,496.547427 --> 00:19:51,956.547427
And every firm will have their own basics that there are-- the IT people at every firm will have in- like interacted with Harvey, um, to set up their own instance, and they'll have some basic, like, this is our firm's workflows and this is maybe some transactional stuff that you can like do data rooms and all this.
194
00:19:52,446.548427 --> 00:19:52,736.549427
So there's all...
195
00:19:52,956.546427 --> 00:20:08,172.549427
I think most people are pretty familiar with that, but we're now looking on the kind of litigation and investigation side to try and set up Essentially like self-executing repeatable work.
196
00:20:08,682.549427 --> 00:20:28,602.549427
So I mean, this has always been a thing for law firms, right? For right from probably the beginning of law firms in a sense that every firm or, and lawyers will have the precedents that they like, right? Which they may rightly or wrongly take with them between firms and so on, right? and they'll have a, a database of preferred clauses and so on.
197
00:20:28,602.549427 --> 00:20:55,402.549427
And I think what we're gonna start seeing with law firms, we're already seeing with law firms, is that there's gonna be these kind of custom workflows, particularly where I think some of the work, some of the work is gonna become a bit more commoditized, right? So you're looking at like-- so some of the compliance work that I do, if you're looking at a suite of compliance documents and they need to be updated from, for the, like, latest, um, legal developments, but it's a suite of thirty-five policies and procedures.
198
00:20:55,912.549427 --> 00:21:09,952.551427
The expectation from clients, if they-- even if, if they don't take it in-house, which is another risk, is that it's gonna be done by your external lawyers at pretty low cost because it shouldn't be the-- that you're not reading and digesting thirty-five policy and procedure documents anymore.
199
00:21:09,952.551427 --> 00:21:13,672.547427
You're using AI to kind of see where, how they all link together and where they need to be updated.
200
00:21:14,12.552427 --> 00:21:19,312.546427
And there's obviously custom agentic workflows that you can develop in your firm or in even in-house to do that.
201
00:21:19,752.549427 --> 00:21:31,52.550427
Um, I think there's some space in some of the investigations work that lots of, that lots of law firms do, um, to make it, um, you know, quicker and cheaper.
202
00:21:31,172.548427 --> 00:21:34,212.551427
And that, again, I think that will be the expectation from the client side.
203
00:21:35,72.549427 --> 00:21:51,82.549427
Um, so yeah, I think we're, we're increasingly looking at this and, and, um, I guess if you're an enthusiastic non-partner, I suspect it's gonna fall on some of those more senior associates to try and work out how to do that, and you'll probably get credit for it if you do.
204
00:21:51,192.552427 --> 00:22:00,12.551427
So, um- I, I, yeah, so I, I'm going to try and w- develop one of these for specifically in the investigation space in the weeks coming up.
205
00:22:00,412.551427 --> 00:22:03,292.551427
Um, but I haven't actually put my head down to think about it.
206
00:22:03,332.551427 --> 00:22:12,722.550427
This is, this is the problem, right? Like, uh, the perennial problem in private practice is that, that, that in and of itself won't be billable work, but you're likely to get a lot of results from it and therefore credit.
207
00:22:13,672.551427 --> 00:22:14,322.550427
Yeah, it's funny.
208
00:22:14,322.551427 --> 00:22:21,242.551427
I'm in a similar place in my practice in that I've kind of mocked up and, and imagined up the solution that I want to develop.
209
00:22:21,862.552427 --> 00:22:43,172.553427
I've written a little bit about this, but it's essentially creating some sort of agentic process to support the business advisory that I do day to day- Mm-hmm take advantage of prior advisory that I've done, take advantage of these guidance notes that we've created internally, and automate a lot of the process of business asks question, I provide answer.
210
00:22:43,912.551427 --> 00:22:50,242.551427
But like you, I've framed it out and just now I, am I really getting to the point where we're about to start building this thing.
211
00:22:50,302.550427 --> 00:23:07,762.555427
So I think timing-wise, I feel like we're actually in a similar place and probably not a coincidence because going back to the, the sort of previous topic, y- yes, there's been a lot of progress, and it's been very quick in the last year and a half, but still feels like we're well short of some significant potential that's out there.
212
00:23:08,232.555427 --> 00:23:08,412.555427
Mm.
213
00:23:08,802.555427 --> 00:23:17,532.554427
And so that makes me wonder, okay, well, why? What's holding us back as an industry? Um, I think we've already addressed sort of the tech is tough to deal with.
214
00:23:17,542.552427 --> 00:23:19,72.550427
It's not quite there yet.
215
00:23:19,132.549427 --> 00:23:20,262.549427
It's not plug and play.
216
00:23:20,262.553427 --> 00:23:22,902.551427
It's not as easy as, as people make it out to be.
217
00:23:23,192.554427 --> 00:23:23,942.549427
So there's that.
218
00:23:24,52.554427 --> 00:23:26,262.550427
We can, we can sort of say that we've covered that topic.
219
00:23:26,692.554427 --> 00:23:31,32.553427
Um, but there's a couple other ones you've already hinted at, I think, in, in your last response.
220
00:23:31,42.560427 --> 00:23:35,222.559427
One of them is this data question, which is kind of garbage in, garbage out.
221
00:23:35,272.555427 --> 00:23:49,542.560427
Um- Yeah do law firms, do corporate legal departments have the data they need to, uh, automate their work? Do they have it in the right format? Is it clean? Is it usable? So let me, let me pose this to you.
222
00:23:50,824.6702259 --> 00:23:51,814.6702259
You're a law firm lawyer.
223
00:23:52,524.6692259 --> 00:24:04,364.6672259
how much secret sauce do you think there is that's unique to a particular law firm or a particular practice or a particular lawyer at the practice? 'Cause I do think right now the topic is data.
224
00:24:04,384.6692259 --> 00:24:05,554.6692259
It's knowledge.
225
00:24:05,574.6672259 --> 00:24:21,574.6682259
It's- Yeah what is this, uh, proprietary thing that lawyers have that's gonna give them an advantage with the help of AI? But there's this huge assumption there, which I continue to sort of test out in my practice, and I, I like to ask lawyers.
226
00:24:22,64.6692259 --> 00:24:27,474.6672259
There's a huge assumption there, which is that there is something there, that there's a real secret sauce there to be exploited.
227
00:24:28,904.6692259 --> 00:24:33,854.6672259
So I think it, I think it falls into two buck- two buckets, right? I think the first is, is data, as you say.
228
00:24:34,244.6652259 --> 00:24:45,624.6692259
Which for law firms is primarily huge amounts of mainly precedent documents, former deal documents and so on, sitting on, probably on iManage, right? iManage Cloud, whatever it is.
229
00:24:45,654.6692259 --> 00:24:52,674.6672259
Massively unstructured, um, huge amounts of duplicates and different drafts and stuff which aren't necessarily that, that useful.
230
00:24:53,154.6692259 --> 00:25:12,734.6672259
Now, do I think that there is that much differentiation between these different law firms' precedents and so on at, at, at the, at a particular level? You know, once you get to Am Law 100 or, or top 20 firms or whatever it might be, um, I, I'm, I don't think there's gonna be a huge amount of differential between m- many firms there, despite what firms will want to tell you.
231
00:25:12,744.6652259 --> 00:25:18,534.6682259
And I, that's not to say that there will be certain firms who are probably better at certain things.
232
00:25:18,594.6642259 --> 00:25:25,134.6672259
You know, pick any of the big names, you might be a specialist in private equity or, um, specialist in white collar or whatever it might be.
233
00:25:25,134.6672259 --> 00:25:26,974.6652259
Perhaps there is some differentiation there.
234
00:25:27,814.6652259 --> 00:25:37,14.6692259
Um, but I think as you say, the, the problem there Is just going to be using that data in a, in an effective way.
235
00:25:38,14.6692259 --> 00:25:42,14.6692259
And then I think the second point is obviously the, the, the individuals.
236
00:25:42,484.6692259 --> 00:25:54,54.6692259
And I didn't And so the, a lot of what people are saying now is that, you know, so you might have a f- you know, if you're Kirkland Ellis, you might have a $500 million in-house AI system, which I think they've said that they're gonna spend recently.
237
00:25:54,54.6692259 --> 00:26:03,974.6672259
I, um, I think was it, was it Latham as well have recently said that they're gonna bring, they're gonna spend some money on some Nvidia kit and run an open model of their own in-house and stuff.
238
00:26:04,594.6682259 --> 00:26:10,674.6682259
Um, but the, the, the whole idea recently has been, hasn't it, is that, well, you have all these systems, but then you'll have the judgment.
239
00:26:10,954.6692259 --> 00:26:23,634.6682259
The, the lawyer, the, the partner who has been at the, the coal face of these top, top deals for the last 30 years, it's his judgment at the, or her judgment at the end of the day which is gonna drive, um, the, the real differentiat- differentiator.
240
00:26:23,894.6682259 --> 00:26:29,244.6672259
And I, I think I was reading something, I don't know whether you saw this story, but I didn't actually, uh, read the full story.
241
00:26:29,244.6672259 --> 00:26:37,514.6662259
It was about the, that they're essentially gonna cr- try and create like a, like a twin, an AI twin of, of individual partners.
242
00:26:37,544.6652259 --> 00:26:38,704.6662259
Did you see this? I d- No.
243
00:26:38,874.6672259 --> 00:26:39,394.6712259
I don't know.
244
00:26:39,444.6672259 --> 00:26:50,884.6702259
I, and I, I, I, I probably would need to read the story more, but I think the idea is that they would l- rather than just focus on just the data as a whole, they would say, "Well, this lawyer, let's look at their emails.
245
00:26:50,884.6702259 --> 00:26:53,634.6672259
Let- let's look at their, the judgment that they do.
246
00:26:53,654.6662259 --> 00:27:00,914.6672259
Maybe let's have the trains in training by this lawyer," and it's go- to create kind of an AI twin of, of each, of individual lawyers.
247
00:27:01,334.6662259 --> 00:27:51,26.6672259
Now- That there's obvious- I think w- where there might not be pushback on the data side from lawyers, that's something that might, you might see some pushback on, right? Because it starts to, that starts to create an industry where, well, what If I, a big, a top partner at a top law firm want to move to another firm, does that then create a problem for me if, if my, my existing firm has created an AI twin of me and then how do I do, how do I differentiate myself? So yeah, I mean, I th- all that's a very long way of saying I think the data question has yet to be solved, and then I think it remains to be seen about just in terms of individual partners and what their judgment might be and how, and how in the future we're gonna see, uh, you know, much more agentic AI and AI decisions that are then supplemented at the sort of higher end by, by individual partners or senior lawyers.
248
00:27:51,876.6682259 --> 00:27:52,706.6682259
That's so interesting.
249
00:27:52,716.6682259 --> 00:27:54,676.6672259
Two takeaways that I have from that.
250
00:27:54,676.6692259 --> 00:28:01,56.6652259
One is that it's human, human judgment that cannot be templified.
251
00:28:01,296.6682259 --> 00:28:01,806.6682259
It's not a word.
252
00:28:01,816.6682259 --> 00:28:02,476.6652259
I'm gonna make it a word.
253
00:28:02,506.6672259 --> 00:28:03,166.6672259
Templified.
254
00:28:03,716.6662259 --> 00:28:10,656.6692259
Human judgment that cannot be templified, um, is something that AI's going to struggle with.
255
00:28:11,496.6682259 --> 00:28:24,936.6672259
The human judgment that cannot be templified is probably the thing that will drive differentiation between lawyers- Yeah just like it always has, and that that is the case in an AI world or, or a non-AI world.
256
00:28:26,486.6652259 --> 00:28:44,666.6702259
But the other takeaway is a kind of more technical one, which is maybe the real value is going to come not from playbooks, templates, these things that we've been talking about up until now, but really more on sort of this digital brain concept, the twin that you just described.
257
00:28:45,296.6712259 --> 00:28:53,626.6712259
Maybe that's sort of really where, um, we're headed to extract that human judgment that cannot be templified.
258
00:28:54,256.6702259 --> 00:28:54,776.6702259
Yeah.
259
00:28:54,776.6702259 --> 00:29:13,616.6712259
I, I mean, people have said this about AI since it started in the sense that I, I don't think I can see a time in the near future where that judgment is gonna be, that human judgment is gonna replace or clients are gonna be happy with that human judgment- I agree fully being fully replaced.
260
00:29:14,26.6712259 --> 00:29:17,406.6652259
E- even if we, you know, we talk about AGI in the next two years.
261
00:29:17,406.6712259 --> 00:29:18,296.6612259
I mean, maybe we won't be here.
262
00:29:18,296.6712259 --> 00:29:21,516.6612259
May- maybe we'll all, maybe, maybe it'll destroy us all, and it won't even matter, Guy.
263
00:29:21,566.6602259 --> 00:29:26,846.6612259
But, um, I, yeah, I just don't, I don't see it happening i- in, uh, in the near future.
264
00:29:27,226.6632259 --> 00:29:42,976.6642259
What I, I don't know, worry about or is I, I feel like it might start to become a winner-takes-all market, and then you might see people at the top end of the market who can afford, firms at the top end of the market who can afford these very expensive systems to start to pull away and capture more market share.
265
00:29:43,506.6622259 --> 00:29:56,928.6702259
I've heard it argued both ways because equally it might reduce the barriers to entry, and if you've got- But big firms that formerly relied on having, you know, 30 junior associates and, and during any litigation and that sort of powering their bottom line.
266
00:29:56,998.6702259 --> 00:30:07,618.6702259
Now you can have smaller boutique firms where AI can do that work and it, and so, uh, it's difficult to say the way it's gonna go, and I think litigation and transactional are gonna be di- are gonna have slightly different outcomes.
267
00:30:08,118.6692259 --> 00:30:14,258.6692259
And there's a huge middle of the market that could consolidate and- Yeah and it's not like accounting, you got the big four.
268
00:30:14,318.6692259 --> 00:30:22,48.6682259
Yes, there are these super premium firms, but there is a huge next tier that could easily consolidate to at least achieve the resources.
269
00:30:22,508.6672259 --> 00:30:25,758.6652259
And match the big guns- Yeah um, of the big players.
270
00:30:25,768.6672259 --> 00:30:29,908.6642259
So maybe that's another counterbalancing kind of effect.
271
00:30:30,8.6662259 --> 00:30:32,418.6682259
So that's sort of the data and expertise side.
272
00:30:32,918.6662259 --> 00:30:36,838.6632259
I wanna try another topic that came out in your earlier response.
273
00:30:36,848.6652259 --> 00:30:52,428.6622259
You, talked about the challenges of, uh, getting associates and lawyers to put time into building things that- Mm will make legal more efficient when that comes at the expense of billable hours in the law firm context, let's say.
274
00:30:53,48.6642259 --> 00:31:14,682.6652259
And then now we also can add to that, this challenge of how do you align, how do you compensate a partner- Um, if they're building this digital brain or twin that essentially becomes a perpetuating kind of IP of the law firm- Mm-hmm uh, that they can continue to extract value from even the partner who's left.
275
00:31:15,272.6652259 --> 00:31:20,102.6652259
Al- tho- both those things for me go in the lane of incentives.
276
00:31:20,442.6642259 --> 00:31:51,612.6642259
How do you align incentives so that lawyers have a reason to maximally deploy AI to push us to the frontier and, and really push us to the point where we are using AI in a way that automates legal tasks, makes us more efficient, not just on the high volume, low value stuff, but also kind of the higher judgment work? If, if let's just pretend for a second that's sort of the end goal for the industry is to sort of get closer to that frontier, um, there's a incentive alignment issue.
277
00:31:51,972.6632259 --> 00:31:57,452.6642259
Um- Yeah I, I would just wanna hear your thoughts on that how do you think about that? Yeah.
278
00:31:57,462.6642259 --> 00:31:58,942.6652259
I mean, I think it's a funny one.
279
00:31:59,752.6652259 --> 00:32:15,112.6622259
Law firms have always s- spent money on hiring people who, who, um, you know, business support people, right? And e- e- when I started as a trainee in 2017 in London, you'd have all these people who were actually mainly tasked with building up the knowledge banks, you know.
280
00:32:15,112.6622259 --> 00:32:24,512.6632259
So the precedents, making sure that we were being careful with how, you know, pulling out, pulling out precedents and putting them in the correct place so lawyers can find them and use them and all that kind of stuff.
281
00:32:24,512.6632259 --> 00:32:27,872.6612259
So in a sense, I don't think that's necessarily gonna change.
282
00:32:27,922.6672259 --> 00:32:32,992.6622259
Unfortunately for law firms, and I think there's been some legal press on this recently, is that, that they...
283
00:32:33,212.6602259 --> 00:32:51,392.6662259
people are gonna have to start hiring AI specialists whose jo- whose sole job it is to do this, right? To work with, to work with, um, attorneys, um, and in order to kind of extract this knowledge and turn it into AI effectively to, rather than just putting it on a precedent bank is now gonna go into AI.
284
00:32:51,782.6602259 --> 00:33:09,892.6642259
Now previously, what the, you know, these people wouldn't have necessarily been, um, on the kind of, uh, salaries that attorneys were being paid, but now because of the extreme expense of AI specialists, firms are gonna find themselves having, I think, to spend a, a lot more money on, on those kind of people.
285
00:33:09,912.6642259 --> 00:33:22,602.6662259
And law firms are inherently conservative places, right? Some firms more than others, and I think some management, management of some firms are gonna find it, uh, difficult to see, uh, where the benefit to that lies.
286
00:33:23,570.6652259 --> 00:33:24,480.6652259
So I think that's one thing.
287
00:33:25,460.6652259 --> 00:33:38,230.6642259
Now, how you incentivize actual fee owners to put down their pens and spend their, what is effectively their free time working to improve AI within a firm.
288
00:33:38,230.6652259 --> 00:33:41,430.6642259
I just, is, I think, I think that's been again, a perennial, a perennial problem.
289
00:33:41,760.6662259 --> 00:33:46,790.6642259
Do you know firms, firms will give, uh, billable credit for, for pro bo- for pro bono.
290
00:33:47,420.6652259 --> 00:34:01,750.6642259
Is there a, is- can we foresee a time when, when firms will give credit for working on these AI systems? I know lots of firms already do give- Yes billable, billable credit for, not necessarily AI specifically, but for these kind of projects that are, you know, not necessarily cl- clearly billable.
291
00:34:02,730.6652259 --> 00:34:25,990.6652259
So, but I think the firm- firms are gonna find themselves in a kind of like in between limit, like kind of liminal state, right? I don't know, I couldn't tell you, but between now and sometime in the near future, are we gonna move to a, uh, a system whereby firms charge their clients for tokens? It's possible, right? I, I, I, I don't know enough about it to speculate any more than that.
292
00:34:25,990.6652259 --> 00:34:30,220.6652259
But you can foresee it, particularly if you're a firm where you're bringing all of this compute in-house.
293
00:34:30,790.6642259 --> 00:34:45,260.6642259
Um, is, is the, is the long-term plan to charge your, your clients for compute? If so, how do you do that? And how, you know, how do you say, "Oh, well, we used a million tokens on," or whatever, "10 million tokens on this project."
294
00:34:45,390.6642259 --> 00:34:48,284.6642259
But again, the client will want to see efficiency from that.
295
00:34:48,284.6642259 --> 00:34:50,380.6652259
It's like this is the whole problem with this token maxing.
296
00:34:50,400.6642259 --> 00:34:58,860.6642259
You know? Uh, will the, will law firms try and do try and go the token maxing route if, if, if they do end up charging for tokens? Um, so I think it's all to play for.
297
00:34:58,900.6642259 --> 00:35:08,990.6652259
And I, I There will prob- a b- a business model will emerge either from an existing firm, um, or from, from a, a, a newcomer that maybe become the preferred, the norm.
298
00:35:09,40.6632259 --> 00:35:09,500.6632259
I don't know.
299
00:35:17,250.8120898 --> 00:35:21,560.8130897
Yeah, I mean, uh, things that I could rattle off just thinking about it now.
300
00:35:22,250.8120898 --> 00:35:22,970.8120897
Benchmarking.
301
00:35:23,740.8140898 --> 00:35:24,30.8130898
Mm-hmm.
302
00:35:24,60.8140898 --> 00:35:29,810.8150898
The ability to tell which tool is better than the other tool, which model to use for which use case.
303
00:35:30,570.8120898 --> 00:35:32,70.8140898
All that falls under benchmarking.
304
00:35:33,230.8140897 --> 00:35:36,160.8160897
The law firm is gonna need an expert who can do that.
305
00:35:36,740.8120897 --> 00:35:36,970.8120897
Yeah.
306
00:35:36,970.8160897 --> 00:35:39,980.8130897
Um, ROI, they're gonna need an expert who can do that.
307
00:35:40,140.8140897 --> 00:35:41,960.8120897
You mentioned token economics.
308
00:35:42,460.8130897 --> 00:35:43,130.8130898
Absolutely.
309
00:35:43,670.8130898 --> 00:35:44,940.8170897
Um, client pricing.
310
00:35:45,650.8160897 --> 00:36:05,630.8110897
How do you calculate fee- fees as you move away from billable hours that are fixed instead of billable? How do you incorporate the cost of AI into the bill? Is it pass-through? Is it a technology fee? So client pricing, I, I think you might need a PhD economist on staff- Yeah if, if you're a big enough law firm.
311
00:36:05,640.8140897 --> 00:36:13,980.8050897
Like, really, I agree that the sort of the kind of roles that exist at law firms are gonna shift quite a bit, um, when it comes to, to law and lawyers.
312
00:36:14,390.8060897 --> 00:36:14,790.8060897
Yeah.
313
00:36:15,120.8090897 --> 00:36:28,510.8050897
And I don't know if you saw, there was a story, was it this I think it was this week, um, that some of the big banks have come out and said, um, publicly that they want to start see cost savings arising out of AI passed, passed directly onto them.
314
00:36:28,990.8040897 --> 00:36:40,590.8060897
Because I think the, the, the, what, what these, these banks were saying was that, "You told us you're using AI to, um, make work more efficient, but we, we need to see this pass through in terms of reduced fees."
315
00:36:41,10.8010897 --> 00:36:42,870.8020897
How firms are gonna do that, I, I don't know.
316
00:36:42,870.8020897 --> 00:36:55,186.8100897
I mean, law firms are al- have always been under pressure from their clients for- For, you know, decades and decades to, to reduce their fees and certain things and, you know But I think law firms change and adapt and the models change.
317
00:36:55,676.8100897 --> 00:37:08,126.8100897
Uh, we sort, you know, we need You document review used to be done manually, um, and now, now it's done on the, on the, on the, the document review, um, you know, e- online systems.
318
00:37:08,156.8100897 --> 00:37:17,556.8100897
And so I just think probably that the model will change again, and I'm sure there will be some cost savings for, for clients, and I hope there will be cost, sort of, cost savings for clients.
319
00:37:17,556.8100897 --> 00:37:25,336.8100897
But again, I just think the economics are gonna, are gonna change, and firms will find another way to Especially if they've got these propriety systems to monetize those.
320
00:37:26,166.8110897 --> 00:37:27,96.8080897
Yeah, for sure.
321
00:37:27,106.8090897 --> 00:37:27,586.8100897
For sure.
322
00:37:27,616.8080897 --> 00:37:36,386.8090897
And so we, we've added, um, incentives to the list of, things that are slowing down AI transformation.
323
00:37:36,976.8110897 --> 00:37:44,826.8110897
Now I could, I guess I could add also this sort of, let's call it personnel resources, just the things you have to build out internally to do this.
324
00:37:45,96.8100897 --> 00:37:45,256.8100897
Mm.
325
00:37:45,266.8080897 --> 00:37:49,186.8110897
It's gonna take time, and it's gonna take resources, and, um, all that.
326
00:37:49,666.8090897 --> 00:37:52,676.8150897
Uh, let's keep moving down the list.
327
00:37:52,686.8140897 --> 00:38:00,336.8150897
Uh, the, another one that I think of is, is you've actually already alluded to, which is in, in your own use of OpenClaw, it's, is security.
328
00:38:00,876.8170897 --> 00:38:01,246.8170897
Yeah.
329
00:38:01,316.8120897 --> 00:38:06,16.8130897
Um, and I think security was the first thing law firms thought of when it came to this.
330
00:38:06,536.8130897 --> 00:38:23,916.8120897
Um, and in some ways I feel like it still continues to be a top concern, and a legitimate one if, a legitimate one if you are following the news and you're seeing about, you know, OpenAI hugging face and all these- Yeah developments around, you know, agents going rogue.
331
00:38:23,916.8120897 --> 00:38:33,926.8170897
And I, I, I personally, I, I find that if I go too down that rabbit hole, I can find five people who sound incredibly smart who tell me there's nothing to worry about.
332
00:38:34,336.8160897 --> 00:38:43,486.8140897
You know? There's, there's six ways to explain exactly why the hugging face hack happened and why it isn't a concern for, you know, AI deployed in the enterprise.
333
00:38:43,946.8170897 --> 00:38:44,166.8170897
Yeah.
334
00:38:44,196.8170897 --> 00:38:57,106.8160897
Um, but then I can find five people who are very concerned, very, very concerned, and then I just sort of using whatever judgment I have, just sort of being at 30,000 foot level, I get anxious, right? I get anxious because of the uncertainty and, and the lack of clarity around this issue.
335
00:38:57,106.8200897 --> 00:39:06,366.8160897
So, um, how do you think about security and, and, and, um, the, the How do lawyers, how are lawyers gonna work through that barrier? Yeah.
336
00:39:06,706.8110897 --> 00:39:13,652.8160897
Well, I think the security, the security issue boils down Boils down to two, two things, right? One is cybersecurity, i.e.,
337
00:39:13,652.8160897 --> 00:39:19,72.8160897
the, the sec- the, the security of law firms' own, uh, own IT systems and so on.
338
00:39:19,452.8160897 --> 00:39:21,912.8160897
And then it's also the security of client data.
339
00:39:22,542.8160897 --> 00:39:33,482.8160897
And just taking the second one first, I don't know if you saw this story this week about, or th- yesterday, uh, that, that OpenAI claims to have solved or nearly solved one of these millennium prize problems.
340
00:39:33,562.8150897 --> 00:39:34,642.8150897
Did you see this? Yes.
341
00:39:34,652.8150897 --> 00:39:49,282.8140897
And then, and then this, um, NYU mathematician came out and said, "Actually, the, the solution that has been alluded to in the press release from OpenAI sounds suspiciously like, um, my approach."
342
00:39:49,842.8160897 --> 00:39:51,792.8140897
And this is not an approach that has just come out of thin air.
343
00:39:51,792.8170897 --> 00:39:53,772.8150897
This is approach that he's generated over years and years and years.
344
00:39:53,772.8160897 --> 00:39:57,422.8160897
And he said, "Well, I actually ran some of these through OpenAI."
345
00:39:57,762.8140897 --> 00:40:00,182.8140897
And OpenAI came back and said, "Well, we..."
346
00:40:00,232.8150897 --> 00:40:14,322.8160897
And I think Sam Altman tweeted, you know, "We can't, we can't rule out that some of the data that he put through may have inadvertently been accessed by these 10,000 agents that were using some, you know, uh, some frontier model that's even better than A- Astra 6."
347
00:40:14,322.8160897 --> 00:40:30,946.8150897
And I think that then points to, you know, you- The, the reason that law firms are u- are u- using Harvey and not using, uh, ChatGPT as much, although there are firms of course that are using enterprise systems like that, and there's all this guarantee that they're gonna be air locked or whatever the term is, you know, they're not taking your data.
348
00:40:31,296.8160897 --> 00:40:52,816.8140897
But I think firm leaders and IT specialists and so on can increasingly see those kind of, uh, stories and think, "Well, maybe then putting aside the economics, maybe then it is a reason for us to bring in our own, um, system r- and run our own closed models rather than rely on these, on, on these o- on these, um, you know, frontier model providers or whatever."
349
00:40:53,76.8170897 --> 00:40:55,326.8140897
So I think that's, that's, that's one point.
350
00:40:55,866.8140897 --> 00:41:05,766.8150897
And then obviously on the, on the cybersecurity side, I just, I mean, most firms are relying primarily on like th- third party providers for these kind of things.
351
00:41:05,766.8150897 --> 00:41:12,666.8140897
We've seen again in recent There's apparently been several law firms hacked this, this week and last.
352
00:41:12,716.8160897 --> 00:41:26,486.8130897
Um, so I think, I, you know, uh, LLM and, and, and, uh, AI models have made it much easier for these bad actors to kind of access firm data and there's And, but it's always, it's always a cat and mouse game.
353
00:41:26,486.8170897 --> 00:41:33,506.8160897
I don't th- I don't think necessarily 'Cause the whole idea about cybersecurity, right, is that AI works both ways.
354
00:41:33,506.8160897 --> 00:41:40,106.8130897
It, it benefits the attackers, but it also benefits the defenders, albeit attackers at the moment are gonna have slight advantage.
355
00:41:40,526.8180897 --> 00:41:46,286.8160897
Um, so I, I, I don't really think on the cybersecurity side we're gonna see much of a change.
356
00:41:46,286.8160897 --> 00:41:56,56.8170897
I know it's, it's a huge, um, priority for firms, and you think about, and also a huge target for these bad actors 'cause you think about the number, the amount of very sensitive data that these firms hold.
357
00:41:56,626.8160897 --> 00:42:02,546.8170897
Um- Although what is telling about those hacks so far is it's mostly just like I mean, I think it's all just social engineering.
358
00:42:03,146.8130897 --> 00:42:03,486.8130897
Yeah.
359
00:42:03,496.8110897 --> 00:42:05,476.8150897
It's just like, it's, it's a hacking 101.
360
00:42:05,476.8150897 --> 00:42:12,366.8090897
They still have to email or call a person- Yeah and convince them somehow to give access to a system that they shouldn't have gotten access to.
361
00:42:12,366.8140897 --> 00:42:20,706.8130897
So, so I guess that puts me, that's one argument in the camp of, look, like it hasn't changed as much as we think it's changed.
362
00:42:21,76.8130897 --> 00:42:25,36.8100897
Um, we haven't yet seen an agent go rogue and steal client data yet.
363
00:42:25,46.8140897 --> 00:42:25,196.8090897
Mm-hmm.
364
00:42:25,196.8090897 --> 00:42:36,894.8140897
But, but- there's obviously good reason for law firms and corporate legal departments to be cautious, and concerned about all this.
365
00:42:37,394.8130897 --> 00:42:45,194.8150897
Um, now at the same time, I think you've given some great examples of, like, how there probably will be workarounds, and fixes,, but that's gonna take time.
366
00:42:45,194.8150897 --> 00:42:46,74.8150897
That's gonna take money.
367
00:42:46,454.8140897 --> 00:42:53,164.8130897
And maybe most importantly, we're still so early in this that it feels like everything changes e- every month.
368
00:42:53,174.8140897 --> 00:42:53,194.8140897
Mm-hmm.
369
00:42:53,574.8130897 --> 00:42:58,24.8140897
And that is not an environment in which you can come up with a solution and feel confident that your solution is gonna be secure.
370
00:42:58,34.8140897 --> 00:42:58,54.8140897
Yeah.
371
00:42:58,604.8140897 --> 00:43:16,644.8130897
And so I think that's just all to say the security topic is, I think, a huge driver for why we're not deploying, to go back to, where we started, why we're not deploying agents in legal- Yeah the way that maybe some venture capitalist six months ago thought we would be.
372
00:43:17,144.8110897 --> 00:43:17,404.8110897
Hmm.
373
00:43:18,744.8130897 --> 00:43:32,554.8150897
No, agreed, and that was the kind of, I guess, that's the element of security that I wasn't really thinking, thinking about, the kind of internal, you know, what happens if your AI systems go rogue, 'cause that's another, a- another, um, sort of risk that, that I...
374
00:43:32,914.8150897 --> 00:43:53,564.8120897
You know, and, and it's only gonna be a matter of time before that happens to a law firm, right? Now, whether it, whe- that's probably gonna, it's gonna be, when we hear the story about that happening to a law firm, I think it's gonna be a lot further down than it happening to other types of companies, as you say, because the law firms are too, uh, conservative, I think, to deploy some of this stuff before other industries.
375
00:43:54,674.8090897 --> 00:44:11,564.8130897
Yeah, and, and maybe this is a good way to come back to where we started with the technology being a limiting factor in its- Hmm in itself, which is, I think what I've noticed that's really important is we are nowhere near seeing any clear winners in this space.
376
00:44:11,984.8070897 --> 00:44:15,774.8080897
Sure, Harvey, Legora, they've made great progress signing clients.
377
00:44:15,774.8080897 --> 00:44:16,724.8110897
They have contracts.
378
00:44:16,724.8120897 --> 00:44:17,404.8110897
They're renewing them.
379
00:44:17,424.8090897 --> 00:44:18,94.8140897
That's awesome.
380
00:44:18,414.8070897 --> 00:44:22,724.8120897
That's some sort of first mover advantage in terms of at least distributing these legal services.
381
00:44:23,204.8080897 --> 00:44:30,994.8090897
Um, but there's really no reason that someone couldn't come along in a year and wow us with something that causes half those firms to switch.
382
00:44:31,434.8040897 --> 00:44:31,674.8040897
Yeah.
383
00:44:31,684.8040897 --> 00:44:32,164.8040897
It could happen, right? Agreed.
384
00:44:32,924.8090897 --> 00:44:34,414.8130897
Um, you go up the chain.
385
00:44:34,424.8090897 --> 00:44:46,564.8160897
You go to the Frontier labs, right? Clearly, there's no winner on the l- on, on, on the large language model side, right? E- every month it feels like someone's maybe going, moving ahead for some particular use, but then ev- the others catch up.
386
00:44:46,574.8030897 --> 00:45:01,504.8060897
So in a space where it's so fragmented the user, whether it's a law firm, whether it's a corporate legal department, a legal user really has no reason to, like, make a bet and, and pick a system, which then means you have to pick multiple systems, multiple tools, which is duplicative.
387
00:45:01,574.8060897 --> 00:45:05,864.8070897
It costs more money, it's more upkeep, it's confusing to the, to your lawyers.
388
00:45:05,864.8070897 --> 00:45:08,134.8060897
You're, you're telling them you have to use three different things.
389
00:45:08,674.8060897 --> 00:45:16,384.8060897
I think the fact that we're nowhere near a consolidation, we're nowhere near a point where we have like LexisNexis and Westlaw to use a- Mm a legal analogy.
390
00:45:16,844.8070897 --> 00:45:28,904.8050897
Um, the, the fact that we don't have one or two options that basically you're choosing from means that I think it's very rational right now to sort of sit and wait, try things, dabble, and kind of see how it shakes out.
391
00:45:28,954.8050897 --> 00:45:35,354.8030897
We have to acknowledge that as a driver for why law firms are not sort of like diving into one tool and being, "Okay, we've got this figured out.
392
00:45:35,354.8030897 --> 00:45:40,14.8050897
We're gonna go ahead and invest a ton of money building out an agentic solution on the following platform."
393
00:45:40,514.8030897 --> 00:45:41,294.8070897
I, I agree.
394
00:45:41,314.8060897 --> 00:45:51,494.8070897
Look, and look, anybody who tells you that the economics of all this is settled, right, is, is just, is lying, right? It's not, that's not u- that's not unique to the, the legal industry.
395
00:45:51,494.8070897 --> 00:45:52,4.8080897
That's just a...
396
00:45:52,74.8010897 --> 00:45:53,864.8060897
That's true across AI as a whole.
397
00:45:53,974.8020897 --> 00:46:06,984.8030897
Like the, um, you know, whether or not the frontier labs are gonna win out and whether all this huge amount of money that's being spent on, uh, on all of the infrastructure is, is gonna, is gonna crash remains to be seen.
398
00:46:07,4.8010897 --> 00:46:19,714.8010897
And like, and on the flip side of, of that, on the frontier labs and all the, and the huge data center infrastructure maybe, maybe for most firms just running like 90, 90% of queries on your own compute in-house is the way to do it.
399
00:46:19,714.8010897 --> 00:46:20,444.8050897
I- Yeah.
400
00:46:20,444.8050897 --> 00:46:20,904.8030897
Nobody knows.
401
00:46:20,904.8030897 --> 00:46:46,26.8060897
And I, I think so law firms, sp- some of the law firms who are spending big on, on this stuff now are the law firms who have experience in the past of, um, great benefits of being first movers On, in o- in other areas, you know, whether that's a big bet on private equity in the nine- in the early 2000s or, you know, big bets on certain o- other types, other, other types of work or ways of working or big hires.
402
00:46:46,566.8060897 --> 00:46:54,156.8060897
But I just don't think that we know enough about what is coming down the line f- for AI just in general for...
403
00:46:54,626.8050897 --> 00:47:00,456.8050897
And may- may- maybe those big bets will pay off, but I think it's far from clear that that will be the case.
404
00:47:00,846.8060897 --> 00:47:02,646.8050897
Oh, I think some of the big bets will pay off.
405
00:47:02,646.8050897 --> 00:47:02,741.8050897
Yeah.
406
00:47:02,741.8050897 --> 00:47:06,96.8050897
I mean, that's sort of the nature of a big bet is some of them will pay off and others won't.
407
00:47:06,486.8040897 --> 00:47:11,906.8040897
Um, but like you, I think, alludes earlier, not all enterprises can make the same size bet.
408
00:47:11,916.8040897 --> 00:47:11,936.8040897
Yeah.
409
00:47:12,6.8040897 --> 00:47:16,626.8060897
So, uh, does that perpetuate a situation that's winner takes all? I don't know.
410
00:47:16,956.8050897 --> 00:47:23,506.8060897
I should have a consultant on who can really talk about sort of competition between law firms and- Yeah and how the legal market competes.
411
00:47:23,536.8030897 --> 00:47:39,446.8120897
Um, but it certainly feels, very unsettled, and it feels like a moment of great opportunity, uh, for, for anyone, um, who's thinking about how to differentiate whether that's an individual lawyer, right, who just wants to stand out within their enterprise or whether it's, uh, a law firm competing against others.
412
00:47:39,946.8050897 --> 00:47:52,186.8020897
To wrap it up, Findley, how are you feeling right now about AI and legal? What are you kind of looking forward to, let's say, for the next six months? Um, you know, g- give me your vibes.
413
00:47:53,286.8010897 --> 00:48:02,796.8080897
There was a time late last year where I was feeling very pessimistic, I mean, P doom or whatever of, of my job, um, where I was...
414
00:48:02,826.8020897 --> 00:48:14,776.8030897
You know, these mo- these really good models were starting to come out and I was seeing, you know, f- the fir- just a reasonably simple prompt and, um, very limited context that these models were able to do some pretty serious drafting.
415
00:48:15,456.8070897 --> 00:48:21,666.8000897
And I was thinking, "Well, I, I don't see why lawyers are gonna exist even in two years' time."
416
00:48:21,736.8060897 --> 00:48:25,186.8070897
That was in the sort of the, my most cynical, my most cynical moods.
417
00:48:25,726.8040897 --> 00:48:27,676.8000897
I'm feeling a bit more optimistic now.
418
00:48:27,686.8030897 --> 00:48:37,426.7970897
I think there's no, there is no doubt that the legal industry is going to change more fundamentally than it's ever changed, and I include in that the introduction of the computer.
419
00:48:37,426.8080897 --> 00:48:43,986.8050897
I mean, the legal industry used to exist as a sort of, you know, armies of typists and clerks and so on, and it, and it changed, but, but it changed slowly.
420
00:48:44,636.8000897 --> 00:48:55,976.8080897
You know, whe- when I fir- when I was first a, a trainee in London in, in 2017, it was kind of the norm, I think both, both sides of the Atlantic that almost all attorneys would have their own assistants, their own secretaries.
421
00:48:56,296.8040897 --> 00:48:57,306.8060897
That doesn't exist anymore.
422
00:48:57,306.8060897 --> 00:49:01,266.8080897
That's been sort of slowly eroded 'cause the economics changes and technology changes.
423
00:49:01,856.8010897 --> 00:49:05,406.8080897
So I, I, I think the legal industry is on the cusp of profound change.
424
00:49:06,126.8050897 --> 00:49:11,626.8040897
But do I think it's gonna be, it's gonna look that different in 10 years' time? I think there's gonna be a smaller number of lawyers.
425
00:49:11,656.8040897 --> 00:49:19,486.8020897
I think unfortunately, particularly in the context of, as I understand it, very high levels of law, law school intakes.
426
00:49:20,46.8040897 --> 00:49:26,810.8060897
I think there's gonna be a lot of pressure on hiring at the lower end in terms of numbers Um, of, of juniors.
427
00:49:27,0.8060897 --> 00:49:31,980.8060897
And I think there's gonna be a very d- there's gonna have to be a very different approach to legal training.
428
00:49:32,590.8060897 --> 00:49:41,790.8060897
Because I think fundamentally, certainly at the moment, a lot of the tasks that AI is very good at is the kind of stuff that junior lawyers would have done for decades and decades.
429
00:49:42,420.8050897 --> 00:49:48,910.8050897
Um, and so we're gonna have, as an industry, to think a little bit more about how we train juniors.
430
00:49:49,140.8060897 --> 00:49:50,700.8060897
And it, it could be a good thing, you know.
431
00:49:50,770.8050897 --> 00:49:58,460.8040897
I think it frees up, it frees up a lot of juniors' time from the, some of the drudgery, some of the assembling of documents, some of the document review and so on.
432
00:49:58,950.8040897 --> 00:50:06,110.8060897
But then you start to get into the economics of it in terms of numbers of associates, in terms of salaries and so on, and in terms of what you get them to do instead.
433
00:50:06,740.8050897 --> 00:50:27,290.8040897
And I think it's good, it's good because it means that they can focus on more interesting things, the kind of, uh, like high-end, uh, transactions or cases that where, where th- th- this legal judgment that comes from the, from the very experienced partners comes, and perhaps we can start to pass down more of that more directly rather than get them to do doc review for 15 hours a day.
434
00:50:27,920.8050897 --> 00:50:39,330.8040897
Um, so, so I'm feeling optimistic, but I think we're gonna have to change much quicker than we ever have as an industry, and I don't think anybody has really a handle on h- what that's gonna look like.
435
00:50:39,330.8040897 --> 00:50:45,220.8020897
So it's a space, it's a Yeah, optimistic but, but uncertain, I think is the, the headline.
436
00:50:46,60.8040897 --> 00:50:46,250.8040897
Yeah.
437
00:50:46,250.8070897 --> 00:50:58,294.8110897
I think it's such, so hard to- Speak in generalities because there's different kinds of practices, there's different kinds of firms, uh, there's lawyers working in public interest, there's lawyers working for-profit.
438
00:50:58,304.8110897 --> 00:50:58,324.8110897
Mm-hmm.
439
00:50:58,334.8100897 --> 00:51:00,104.8110897
There's ones in government, there's ones in-house.
440
00:51:00,554.8110897 --> 00:51:06,114.8110897
So I think it's always gonna be very much defined by, uh, different lawyers' specific context.
441
00:51:06,114.8110897 --> 00:51:11,244.8110897
But I think if I were to think about my kind of practice, your kind of practice, which we would call corporate legal practice, right? Mm-hmm.
442
00:51:11,744.8110897 --> 00:51:13,444.8110897
Whether I'm in-house, you're at a law firm.
443
00:51:14,24.8100897 --> 00:51:20,464.8100897
I do think that for those that survive, it's gonna be higher value work and higher paid.
444
00:51:21,304.8090897 --> 00:51:30,404.8150897
I think for those that make it, who make the cut, um, they're gonna get to do better work, more interesting work quicker, and, top law firms are gonna get paid more, money for it.
445
00:51:30,664.8140897 --> 00:51:33,64.8140897
Now, there might be fewer people that survive.
446
00:51:33,74.8140897 --> 00:51:33,94.8120897
Mm-hmm.
447
00:51:33,144.8140897 --> 00:51:38,604.8150897
There might be fewer people needed in that system, but I think the ones that survive will probably be better off for it.
448
00:51:39,4.8140897 --> 00:51:52,144.8160897
Um, now, the ones that don't make that cut, I wonder how much self-selection there will be into other roles at law firms or, uh, all these other roles that we're talking about that are sort of legal adjacent, maybe in legal tech, et cetera.
449
00:51:52,504.8140897 --> 00:52:02,194.8050897
Um, and, and it might be that, after two, three years of cutting their teeth in legal practice, a lot of people move into these adjacent roles and are perfectly happy and well-paid to do those roles.
450
00:52:02,714.8050897 --> 00:52:07,704.8060897
So that's, another way that I think it's hard to sort of net-net say what the impact would be.
451
00:52:08,44.8090897 --> 00:52:10,694.8090897
Um, I think a lot of work will be generated by AI.
452
00:52:10,704.8060897 --> 00:52:12,414.8100897
I think already we're seeing that.
453
00:52:12,414.8120897 --> 00:52:15,64.8060897
We're seeing more work because AI's around.
454
00:52:15,94.8080897 --> 00:52:17,74.8120897
We're seeing more demand for client services.
455
00:52:17,484.8070897 --> 00:52:22,384.8080897
We're seeing, um, um, inefficiencies created by AI that create more work.
456
00:52:22,384.8080897 --> 00:52:22,674.8090897
Yeah.
457
00:52:22,674.8090897 --> 00:52:26,654.8070897
And, you know, we're seeing, pro se litigants can file lawsuits with the click of a button.
458
00:52:26,664.8100897 --> 00:52:31,214.8120897
So if you're, a litigator, for example, you, you actually might see an increase in work.
459
00:52:31,214.8120897 --> 00:52:34,384.8080897
So I think net-net it's hard to predict how it shakes out.
460
00:52:34,634.8070897 --> 00:52:40,584.8050897
But I think certainly there will be to, for lack of a better word, there will be winners and losers when you look at specific silos.
461
00:52:41,334.8040897 --> 00:52:41,664.8040897
Agreed.
462
00:52:42,534.8040897 --> 00:52:42,714.8040897
Yeah.
463
00:52:42,794.8100897 --> 00:52:43,64.8100897
Yeah.
464
00:52:44,194.8040897 --> 00:52:47,384.8050897
Um, well, Findley, thanks so much for coming on and talking to me.
465
00:52:47,574.8090897 --> 00:52:53,234.8210897
Uh, we should do this again soon because I do think every three months, uh, my mood, uh, shifts dramatically.
466
00:52:54,434.8100897 --> 00:52:54,784.8100897
Agreed.
467
00:52:54,834.8190897 --> 00:52:56,814.8190897
And, um, yeah, thank you very much for having me on.
468
00:52:57,74.8090897 --> 00:53:01,714.8080897
Hopefully, um, when I see you again in the next, in the next few months, we won't have I still have a job.
469
00:53:04,514.8120897 --> 00:53:06,94.8150897
Um, I have no doubt you will.
470
00:53:06,114.8180897 --> 00:53:07,74.8170897
I have no doubt you will.
471
00:53:07,304.8200897 --> 00:53:08,414.8190897
Um, yeah.
472
00:53:08,534.8160897 --> 00:53:09,354.8130897
It's been great chatting.
473
00:53:10,84.8190897 --> 00:53:10,294.8190897
Thanks.