A fresh Anthropic announcement set off a week of market jitters and existential questions: what happens when the big model shops ship “legal productivity” features and the public markets flinch. This week, we bring Otto von Zastrow back for a rapid-response conversation, with a front-row view from New York and a blunt take: software grows cheaper to reproduce, so value migrates. The discussion lands on a key distinction, interface versus data, and why the old guard still holds leverage even as new entrants sprint.
From there, the conversation zooms in on “systems of record” and the uneasy truth that the safest vault often loses mindshare when a new interface sits on top. Otto points to email, calendar, SharePoint, DMS platforms, and the growing power of a single chat workspace to become the place where work happens. The hosts press on a critical nuance for lawyers: legal research data is not flat, and “good law” demands hierarchy, treatment, and reliable citation context, not a pile of cases plus vibes.
Otto frames Midpage.ai as a data company first, built on continuous court ingestion plus normalization that used to demand armies of editors. He argues AI turns messy inputs into structured repositories at a scale that favors speed and breadth, yet accuracy still requires process design and verification loops. Greg sharpens the point for litigators: the bar is not clever answers, the bar is defensible citations, negative treatment, and confidence that the record matches reality. Otto agrees on the need for trust, then flips the lens: many annotation tasks look like grind work where modern models, paired with strong QA, start to outperform large manual pipelines.
The headline feature is integration via Model Context Protocol, described as a USB-C style connector for tools and models. Midpage chose distribution inside Claude and ChatGPT rather than forcing lawyers into yet another standalone site. Otto explains the wager: lawyers want fewer surfaces, and general chat platforms ship features at a pace no niche vendor matches alone, so the smart move is to meet users where daily work already lives. The demo story centers on research inside chat, with Midpage returning real case links and citations, then letting the user push deeper with uploads and follow-on tasks, while keeping verification one click away.
The back half turns to second-order effects: pricing, agent spend, and the rise of “vibe” work where professionals act more like managers of agent teams than sole authors of first drafts. Marlene raises governance and liability when internal DIY tools pop up outside formal review, and Otto predicts a pendulum toward professionalized deployment plus change management. The conversation closes on Midpage’s “holy grail” topic, citators and the case relationship graph, plus a clear-eyed forecast: standalone research websites shrink as a primary workspace, while research becomes groundwork performed by agents, with lawyers spending more time interrogating results than running searches.
Listen on mobile platforms: Apple Podcasts | Spotify | YouTube | Substack
[Special Thanks to Legal Technology Hub for their sponsoring this episode.]
Email: geekinreviewpodcast@gmail.com
Music: Jerry David DeCicca
Transcript
Greg Lambert (00:00)
I am Greg Lambert with the Geek in Review and I am here with our friend Stephanie Wilkins from Legal Technology Hub. Stephanie, how you doing? Good. So I hear that LTH is putting out some new geographic coverage. So it’s not just the North America and Europe ⁓ anymore, is it?
Stephanie Wilkins (00:07)
I’m good, how are you Greg?
No, it’s not. mean, a lot of it started, a lot of our coverage started in the US and the UK, because that’s where things sort of sprang from. But as we all know, the legal tech market is global and it’s becoming more so by the day. luckily, as we’ve grown our content team in the last six months, we’ve also grown that globally.
So we have a writer on the ground in Sweden currently, and we have a writer on the ground in India, which are both hot legal tech markets right now. I suppose we’re having a lot of great conversations and going to local conferences and really getting the lay of the land and a much deeper understanding of what’s going on in those legal tech ecosystems specifically. And we figured we shouldn’t let all that Intel go to waste, so we’re going to capitalize it and start some coverage around it, some landscape overviews of the region.
Greg Lambert (00:45)
For sure.
Stephanie Wilkins (01:06)
I mean those are the two we’re going to start with because we have the work already underway and our writers locally, but you should expect to see some overviews geographically, some deeper dives into different parts of the world. If there is an area somebody is really curious about, you can let us know and we’ll add it to the list because we do want to do more of this as we go along. In the meantime, if you are playing around in our directory, there is a way to sort it by the region served. So if you’re curious in a particular geography, that’s a good starting point.
point, but soon enough you will have deeper looks at these different areas and I’m particularly excited about the Nordics in India because there’s so much interesting stuff coming out there. And so stay tuned on LegalTechnologyHub.com for more to come.
Greg Lambert (01:49)
All right, thanks Stephanie. Thanks for catching us up.
Stephanie Wilkins (01:51)
Thanks.
Marlene Gebauer (01:59)
Welcome to The Geek in Review, the podcast focused on innovative and creative ideas in the legal industry. I’m Marlene Gabauer.
Greg Lambert (02:06)
And I’m Greg Lambert and this week we have a returning guest. Actually, we like spun this one up really quick because the topic is so hot. have Otto Van Zastrow, the founder and CEO of MidPage AI. Otto’s been on the podcast before and we’re glad to have him back, especially giving some timely news here with MidPage.
Marlene Gebauer (02:28)
Exactly. Midpage just announced a new integration that allowed Anthropix Claude to connect Midpage’s legal research tools and case law. And so I think it’s, it’s a good moment to step back and talk about what this kind of connection really means in practice.
Greg Lambert (02:42)
All right, so Otto, let’s spin this up and get started. Welcome back to the Geek in Review.
Otto (02:48)
Great to be here.
Greg Lambert (02:50)
All right, so let’s just jump right in to the deep end here. You’re now based in New York. You’re right in the thick of it. And when Anthropic released a legal productivity plug-in just on February 3rd, we saw all the legal tool markets kind of crash a little bit, TR and Lexus stocks.
know, drop 15 to 20%. So what was the conversation like inside your organization there at MidPage? Do you view this plugin as like a validation that AI agents are the future or do you see it as a threat that something like software is dead?
Otto (03:38)
⁓ That’s a big question. I mean, I was, coming from a, you know, software and robotics background. And I have for the last two years been telling all my friends that I think that software is, you know, going to become less and less important because it’s easier and easier to create. And of course this means that we’ll have much more software than ever before. And we’ll do many more things and it will be growing GDP in many interesting ways. But if…
Greg Lambert (03:40)
You
Otto (04:00)
you know, if you just look at a fixed piece of software, the value of that will go down because it’d be so cheap and easy to recreate from scratch within days.
I think that the big publishers are of course not just software, right? For them it’s much more about data. And then I think it depends a lot on, know, every data has some kind of half-life when it loses value. Sometimes this data is proprietary, sometimes not. And if some of this data can be gotten by software, ⁓ then it feels more like software that can change value quite quickly. I think it’s very clear that there’s not just, you know, generated software from nothing that is compatible.
competing with the old guard players like Reuters here. It’s also the big legal tech companies like Harvey and DeGora and smaller ones like MidPage that are growing really fast where it’s just less clear where do the users spend their time. so software is obviously not going away, but it’s the value shifts to other parts.
Marlene Gebauer (04:52)
So, you know, a lot of the legal tech startups, they’ve built their business model on being like the interface for, for legal AI. So, you know, now Anthropic is offering these, these open source templates, you know, for contract review and NDA triage for free. ⁓ does the wrapper business model exist or will it exist? you know, are we seeing kind of a bifurcation of either where, you know, you own the data like mid page.
Or you own the model like Anthropic and everyone else in the middle is getting squeezed.
Otto (05:24)
So I think it’s.
there are some companies that are a system of record. This means that all the really important critical data is in those companies. And this could be your document management system or your case management system, sometimes your CRM. And of course, it’s really hard to switch away from these. there are examples where this is now becoming much easier too. So it’s not as if those are like 100 % safe and everything else is 100 % not safe. Because now when I
use Claude, you know, personally, it has access to all of my emails. I cannot write emails, but it can search across my emails. There’s some limitations where, you know, it doesn’t have access to everything. But, you know, same for the calendar. It’s really effective if it can help you find who was in this meeting from three months ago. I forgot, you know, it can look in your calendar and find that. And so if you have a new interface that is connected to your old systems, maybe it’s connected to your iManage or to your SharePoint or to your emails.
then suddenly the system of record is actually very vulnerable to being taken over by a different interface. And sometimes the data belongs to the law firms or belongs to the big corporates. And sometimes the system of record has built up these high walls. And I think if you look at the big publishers like TR and Lexis, they are actually rarely acting as a system of record at all. And their value comes more from having…
customer trust and relationships but also having been the only ones that had the data in the past.
Greg Lambert (06:50)
Yeah,
well, let me point out a couple of things. One, I wrote that question and I said that Claude’s offering this for free. Well, that’s not quite true in the fact that a lot of this you have to use co-work, which right now only works in a Mac environment. And I can’t remember if it was in the Macs plan, which is the $200 a month plan. I think they may have.
drop that down somewhat, but so there’s that. want to clarify that. But also, I want to push back, because I think you understand this as well. It’s not just about having the data. It’s about how well a lot of these companies have structured the data. And so that it’s not just flat data. You can have all the case law, right?
there’s a lot of issues with just searching all the case law and not understanding there’s hierarchy to that, there’s precedence to that. And these companies have been around for decades or even like a century and a half to help kind of structure that data. Is there still value in how they’re structuring the data that they have?
Marlene Gebauer (08:05)
structuring and connecting ⁓ to other relevant sources that other people might need.
Greg Lambert (08:07)
Yeah.
Otto (08:12)
Yeah, that’s a great question. I’m going to answer in two parts. First about the data, does it matter if iManage has everything structured and can anybody now just take over without having all the structure? And then second, just know more about Claude and whether it’s now free and people used to pay for stuff like that. think…
Greg Lambert (08:28)
Okay.
Otto (08:29)
know, we, Midpage is a data company, right? We collect data from all the courts, ⁓ meaning we have hundreds of web scrapers and we download new cases from the courts multiple times per day. So as soon as it comes up, and it’s not just a court case data. And what we got really good at is ingesting these huge amounts of data and organizing everything so that it feels like the published content would feel like on a platform like TR or so on. And there’s not just, you know,
AI is not even that heavily involved in this. It’s a lot of code, but these are things that used to be done sort of by humans. And there’s like definitely parts where AI is very involved. And I think the interesting part is that it’s very clear that taking very messy and unstructured data and moving into whatever perfect schema we would like to have is actually very easy and, you know, a hundred times easier than ever before, as long as the information is somewhere in there.
I’m sure there are some business processes that are just not being captured digitally anywhere, then it’s going to be difficult. And of course, just having the ability to transform the data into an organized repository does not mean that the users and the law firms are going to want you to do that or that they’re going to adapt to a new process that works with a different data structure. I think the job for…
many of the big law firms, but also the legal tech vendors is going to be to go to all those unstructured piles of data and help the law firms make it way more organized than it ever has been before. And then suddenly the law firms can say, it’s now so easy to look at every single conversation related to a very specific niche legal question that they had and have this knowledge be instantly available to their agents, but also to their lawyers. And I think
This is just what is going to happen to any professional field that is a white collar job, right? That everything will be recorded and everything will be available and it will be easy for agents to search. And so yeah, think if you’re a legal tech vendor and you’re going to a company who’s used iManage, nobody really organizes, like no law firm uses iManage in the correct way in the sense that it’s so much work to label everything as to where it should be going that they only do
you know, half of the labels most of the time. And AI is actually gonna be much, much better at keeping all these things super accurate, super consistent and avoiding some critical errors where data get put into the wrong box. And so I think it’s gonna be a lot more, you know, fun and easy to work with. And I think that some of the vendors are going out to the law firms and using forward deployed legal engineers to now help them do these things, right? Help them make messy things organized.
And now I’m gonna go on to the second part of the question with Claude unless you wanna interrupt there.
Marlene Gebauer (10:56)
Yeah, no, go ahead.
Greg Lambert (10:56)
Well,
now let me me interrupt for just a second because that’s that makes sense with our internal data. But if I’m a you know if I’m a litigator and I’m relying upon you know the information that I’m getting that that you know I’m using in my case and citing to case law. I want to make sure that that that data is accurate that I know if it’s been cited.
I want to know if my AI is hallucinating or not. And I want to make sure that I’m citing good law and I’m accurate in that. Is there still value in the big legal publishers in having that information highly structured? Because I think, yeah, I can take my internal data and make it really good and clean, but it’s still kind of flat.
with legal information, there’s this hierarchy between local, state, federal information, whether it’s been overruled or not over time. a case might have been cited a thousand times, but if it’s been overruled in the last six years, then it may not have a good balance. I’m trying to think of the arguments that the West laws and Lexuses will have on
Why you still need them and you still need to work directly with them? Do you see that holding true?
Otto (12:22)
Well, I think we’ll definitely need those companies in the future, but for slightly different reasons. Like I think, you know, they will definitely stay around mostly because they own, you know, proprietary data, like secondary data. That too, that too. But,
Greg Lambert (12:32)
Was it because they own all the stuff?
Otto (12:35)
they have a lot of secondary publications that are really important to niche fields and those are gonna continue to relying on it, whether it’s humans or agents. It is true that even there is a lot of humans doing work integrating these publications, that human work is now also feels a lot like just somebody asking Claude across a database to write a memo. And it’s gonna take some while until the data gets to that point. But definitely data collected by humans, if it can also be collected
by agents, eventually agents are gonna be able to do that better without needing to be more intelligent than the humans. They’re just gonna be more thorough and more broad. And if it comes to the cases that you mentioned when TR has the best citation and jeopardization system and so on, and they have been famous for key sites for decades or centuries now.
However, the way that this works today is you have a huge campus of thousands of people manually reading every new case, going over them and marking them as, this case is overruling another case over here. And then they look at the typos, they fix those, they make sure everything is edited into a nice clean format. All these things are…
type of labor that humans are not that great at compared to even today’s AI. If you tell someone you have to read 10,000 cases every day, maybe you’re a team of thousands of people, this is going to cause many mistakes. And that’s why, of course, today they have teams of people correcting and they probably do it twice so that they see a certain difference, that they look at the difference.
But it is clearly something that involves huge amounts of reading and annotating of texts with very low, you know, low intelligence tasks that AI can probably already do better today. And these are the things that we’re doing. You do have to be careful about.
you while you build up the data, you build up your capacity to evaluate and have a process that is going to lead to very accurate results. So whether this is a human organization or a software and agents organization, you still have to build the processes to make everything accurate and put it into the structured format. But yes, it is now doable for a small company and it used to require a company that had billions in revenue.
Greg Lambert (14:44)
Yeah, that makes sense.
Okay, sorry. Enough of my interrupting.
Marlene Gebauer (14:47)
⁓ So KM
isn’t dead yet.
Greg Lambert (14:50)
Careful.
Otto (14:51)
I think KMs will become more important, or less important.
Greg Lambert (14:53)
Yeah, I think so too.
Marlene Gebauer (14:54)
Mm-hmm.
Greg Lambert (14:55)
But we are biased on that front. So all right, sorry. We’ll let you get into the second part, which.
Otto (15:02)
Claude.
Yeah, so I think that
most law firms in the future will have both a general purpose chat tool like Claude or ChatyBT rolled out enterprise-wide and they will have several legal tech companies. Maybe it’s a Harvey Ligora. Maybe there’s a few other tools that they have as well. And I think it is a bit, is it going to be a bit like Microsoft Word where it’s becoming so ubiquitous. It’s just, you everybody has it for some reasons. And then the big, you know, Harvey’s going to compete
with chat EPT on whether a lawyer spends their time but they will both have a purpose they will be somewhat overlapping but also sometimes not overlapping and probably eventually
both of them will have access to all of the law firm’s documents and emails and so on. But I wouldn’t be surprised if at the beginning, Harvey is going to be much better at implementing the right processes and controls from making sure access is handled in exactly the right way, because it is really understand how law firms operate. We’re doing all the same things, too. However, of course, the number of lawyers in the world is very large and not all of them work in the very big enterprises. I think that the number of
of professional legal users is gonna be way bigger in Claude and Chat2PT and perplexity and others by many orders of magnitude. And what MintPage is doing now is we’re building integrations to all of these general purpose chat platforms and providing all of the case law data and the statutes and regulations data and all the things that you need to have your Claude or your Chat2PT become better than any,
Co-counsel or Lexis plus AI would have been before because now you have the benefits of Claude plus all the caseload data and ⁓ And now these tools are much more widespread. So whether you’re in a corporate or maybe you’re not a litigator But you’re someone who is working in transactional. This has like a one-off questions about whether this particular contract clause is enforceable You’re not gonna open up Lexis in the future because it’s so overwhelming to look at all of these thousands of cases You’re gonna use some kind of chat interface and it’s probably gonna be the one
that you use for all of your other tasks as well. And if it has access to case law, that’s sort of the best of both. And these ecosystems are just very powerful. You want fewer tools, not more tools. And if you can fit in a lot of integrations into your Claude, that is gonna be a place where most people do their legal work. And I think that Claude…
with the SharePoint integration, many smaller firms are already gonna say, actually, this is a very, very powerful, very universal tool for their use cases.
Marlene Gebauer (17:28)
And we’ll get into the integrations in a minute, but I don’t know. sounds like it sounds like. It sounds like you’d have to buy more of these, these tools. sounds like you would have a numerous ones that you’d have to use. And if there’s integration, ⁓ normally what we have found is, is that you have to have a subscription to, know, whatever’s underlying. so, there’s, there’s sort of more things that firms and organizations are going to have to buy and, and, ⁓
You know, I’m sure, you know, people are listening and saying, was like, how are we going to pay? How are we going to pay for that? So so I’m curious, like any thoughts on that?
Otto (18:04)
I think it’s a little bit like the app store on smartphones where at the beginning there were only a few apps and over the time it was thousands of millions. I do think that big enterprises will take a long time until they have a large number of apps that they use within Claude. But I think it will get there because it’s just the most convenient solution for everyone. And…
I think there’s less value in having your own website interface now because it’s going to be agents doing the work there anyway. So you don’t have to have a website. You just have to have something that you’re selling that nobody else is selling. ⁓
Greg Lambert (18:36)
So
it’s kind of like to bring up your Microsoft Word example. It’s kind of like our plugins that we have in Word. Yeah, Word’s the interface, but to make it really work in the way that we want it to work, we…
put these plugins and connect to Latera or Harvey or Legora or all these other tools and so the interface stays the same, chat GPT, Claude, whatever, but the backend we have more tools that it can access and more data. Am I interpreting that correctly or am missing something?
Otto (19:11)
I mean, yes and no, I do think that it’s going to be so much more impactful than Word plugins ever were. You know, the plugins in Claude are going to be bigger than the plugins in Word. So I’m not sure if it’s going to feel similar. That’s why I made the comparison with the phones and the apps for there, because at the time, there was a time where that was new and where people thought of apps of more like a nice website, ⁓ but not more. I think that…
The main reason is that, of course, agents are now doing a lot of the work. And if the agents are operating in a world that is getting bigger and bigger, they can do more more share of the work. For example, as a software company, our own usage of coding agents has grown a lot. We’re now spending probably, I think, around $8,000 per month just on the developers that are using coding tools to help them and are totally at cost, of course. It’s much, much higher than that.
And this number, well, yes, but with the new models coming up, we’re always switching to the most expensive models right away. So as soon as…
Greg Lambert (20:02)
But that’s still cheaper than one engineer, right?
Marlene Gebauer (20:04)
Hahaha
Otto (20:13)
If the growth with developer costs continues, then in a year we’d be at much more AI developer cost than human developer cost. The growth has been very rapid. We started at 100 a month just a year ago. So that’s 80x in one year. And I think for much…
for many white collar workers, but by a bit it’ll feel similar and the software coding is usually the first one because coders build the tools right for themselves so they build it there first. And for lawyers, I think that the money that they spend on agents doing work for them or work with them is also going to rise in a similar fashion.
We spoke to one of our users on Tuesday where they were talking about writing briefs, writing 20 page briefs and doing research on the page and doing this, taking over the work that would have been done by three people. He actually, in this case, let go of two other people. think often they just do more work. They also, that same person was just doing a lot more business than before on a fixed fee rather than an hourly. ⁓
I
think this means that your agents are taking more of your work and you’ll be paying more to have the AI thinking.
Marlene Gebauer (21:21)
We could chat about this all day, but I am going to switch topics and talk about your specific news. auto you’ve launched an integration using what is called model context protocol or MCP. And that sort of lit up everything, know, in January. so for listeners who aren’t developers, this has been described and I read this as the, the USB C port for, for AI. So maybe you can.
explain a little bit about how it works and why you chose to integrate into Claude rather than trying to keep lawyers on the mid-page platform. And yeah, so we’re curious about that.
Otto (21:59)
Yeah.
I mean, the choice was easy. Like we have a platform that feels a lot like Claude, but it also feels a lot like a pretty version of your Lexis or Westlaw. And so you can go that a search for cases and you have an agent that can do research with you and that can draft and you can read the cases and it all feels very familiar, maybe just with a more modern and fast touch to it. But of course, if lawyers are now more and more spending time on Claude and if Claude has, I don’t know,
how many right now, like 500 million, maybe they have 200 million users right now and out of those, 20 million have legal questions once in a while. And maybe at this point, 10 million or so are actual legal professionals. It means that we can get a huge reach by going via Claude. And we actually have heard many of our users that they’re using Claude at the same time as they’re using the page. And I think the…
Marlene Gebauer (22:51)
How does that work? curious, like,
are they, just have like a couple windows open and they’re asking different things or?
Greg Lambert (22:55)
copy
and paste results out of one into the into quad and then work with that we have we have lawyers that do that with with Harvey they’ll go pull stuff from Lexis Westlaw or from the I manage drop it all into Harvey and then work with it with it there so I imagine something similar to that
Marlene Gebauer (22:59)
See you later.
Otto (23:14)
Exactly, just like that. The thing is that these general purpose platforms are adding many features very fast that you’re always playing catch up. And so if you can’t beat them, you’ve got to join them. And I think Harvey has email integrations, but I’m not so sure whether they have the same level of…
know, Calendar, SharePoint, and these 10 other apps as well, and then the web search, and that are all becoming very relevant. And I think with the chat platforms, like we can in a way coexist very nicely where there’s some things that…
is very legal specific that Claude is not doing. And the legal specific parts are also growing, right? Because you can build software so easily. We’re building the largest legal data set of the world and building this is now.
You know, this will in the future also include things like court rules and things like having an API to submit things from the court or request things from the court that aren’t publicly available. And so the scope there is expanding too. And there’s a chance that some of that will be easy for any AI agent to do in the future. We’ll see about where the difficult parts are, where the lawyers get stuck today and just do those parts.
Greg Lambert (24:19)
I mean, we were talking before we got on because it’s not just Claude that you’re integrating with. There’s also, you have a relationship with OpenAI and ChetGPT model as well. And it’s interesting because on February 5th, Claude announced that Opus 4.6 is out. And then like an hour later, ChetGPT said that
5.3, four codexes out. So I mean, like you were saying, there’s just this constant movement in the AI market and improvement in the large language models. Can you talk to us about how you’re positioning yourself with both of those models? And maybe if you have a chance, maybe give us like a little demo of what you got going on.
Otto (25:13)
Yeah, happy to. So the differences between those are not that big. Of course, they’re in a race and they’re even though the anthropic management says that they have a slightly different focus because they have a lot more enterprise revenue and coding revenue and OpenAIR has a lot more consumer revenue. I still think they are extremely overlapping and ads now. Yes. Yeah, I agree. I am.
Greg Lambert (25:33)
And ads now. I’m looking forward to the Super Bowl ads.
Otto (25:42)
And so right now there’s currently a stark difference in one thing that affects us is that Claude has announced this Claude Cowork. And Claude Cowork is very much like tools that developers have had for the last six months. And I’m sure that ChatsBD is going to be right around the corner with their equivalent version. But what Cowork does for those that might not know yet is you can open Claude on your desktop and you can give it access to a folder of anything.
be your entire, you know, can be all the projects or matters on your desktop if you want to. You might have to ask your IT for security first, but the advantage is now. Yeah.
Greg Lambert (26:14)
⁓ you will have to ask IT
Marlene Gebauer (26:16)
You better. You better.
Greg Lambert (26:17)
Marlene?
Otto (26:20)
Exactly. But so, you know, if you have that allowed, Claude can now actually edit Word documents, you know, create new ones. It can read hundreds of PDFs and it can do that all over tasks that can span, you know, 30 minutes, 60 minutes. And so a lot of work where we would have previously had to like upload a file or two, see what it says about that, and then gone back to, you know, your local Word and then continue there. A lot of that process is much more interesting now.
Now, Cloud Book Co-work is very new, was released a couple days ago, and still has a couple of bugs and glitches, but it is, you know, those are obviously just very temporary pain points. And in the context of a litigator, this could mean that maybe you have a huge amount of e-discovery files, or a huge amount of transcripts and just a very large docket, maybe for various reasons. It is now going to have access across all of these. If you then also have the mid-page
then of course you can ask how the things that it finds in your folder are related to case law. And I think that’s sort of a really exciting big shift. And because it can now also edit and display Word documents very well, people are gonna really jump at that, especially the smaller law firms, because they’ll get permission first.
Greg Lambert (27:25)
One of the things that I’ve been curious about because of the way that Claude and OpenAI and Gemini are kind of positioning themselves and just the reaction of the market just by releasing ⁓ a small plug-in for legal, what kind of tremors that sent through the market.
⁓ What’s to stop a, you know, ⁓ Claude from doing ⁓ what a lot of people saw Amazon doing with small vendors that were using them as their platform? And that is they see people using this data or buying these products and then all of a sudden they start making these products and selling them directly.
What’s to stop them from taking on the Amazon model and kind of squeezing out the small vendors who were using their platform to sell their product?
Otto (28:28)
Yeah, think that for Amazon, there are probably a lot of companies that are very happily and successfully continuing to sell products through Amazon. And so the ones that got squeezed out…
you’re always dependent on a platform if you sell through that. And we are gonna be dependent on the chat platforms. We’re gonna be using multiple different ones. There’s clearly not a monopoly. And a monopoly would be really bad if we had an AI monopoly for some reason. But at this point, because it’s still so easy to switch right now, it doesn’t feel at all like that. And…
And yes, we’re very, you know, we always look at is this feature something that Claude is going to have in two months or do we build it ourselves? And sometimes we’ve even built things that we know they’re going to have eventually, but just for the next year, it’s going to be a real pain point for our users. And by then we’ll have like other things too. And so I think Claude is, you know, they’re going to run, try and run after, you know, 20 different things at once, but not after a thousand different things. And the economy is just a very big place with many, many, many different problems.
problems.
Greg Lambert (29:23)
alright uh… did you want to and did you want to show
what all you’re doing.
Otto (29:28)
Yeah, totally. So let me quickly show how this looks. I’m going to share my screen.
Now, again, this is, whether this is ChatyBT or Claude, it doesn’t actually make that much of a difference anymore. ⁓ I talked a lot about Claude, but let’s show the version in ChatyBT. ⁓ And…
Here I’m just asking a question about, can you find a non-compete case where the duration of a non-compete was over five years? And this is a good example because it’s relevant for both corporate and litigators sometimes. And five-year non-compete is actually very rare. Most of them have much shorter non-competes. And you see here, mid-page return back some search results. And then it starts sort of in the chat answering the question. And because it’s a real integration,
not just going to pull results from Google, it’s going to search our entire database of case law ⁓ and come back with links, with citations. You can click to open the case ⁓ and then you know this is not ever going to be an AI generated case ⁓ because you can open it and verify it.
⁓ And the great thing is because it’s chat-a-bitee or because it’s GLaD, you can then upload maybe a complaint. You could ask it to go through each of the arguments and citations and create, you know, find examples for each one that show the opposite side. And you can sort of go really in depth with these things. And the agent is not just doing one search. It’s often doing, you know, five, 10 searches. ⁓ It’s then finding search results that are not quite on point, but that are interesting. then opens these, finds more cases that are being mentioned.
and then there’s more searches and follow-ups based on that. So it really starts feeling like litigator who has this long multi-step process all through this simple interface. ⁓
Yeah, and one of the cool things is that ⁓ you can see these platforms are evolving quickly. You now can sort of start to show a little bit of UI inside of the chat. ⁓ And over time, ⁓ the chat experience will feel a bit more as if it can, when relevant or when useful, maybe display a document preview or display sort of more interactive things.
Greg Lambert (31:40)
Interesting, and as a, I’m purchasing the mid-page MCP add-in to Claude or ChatGPT, it recognizes me, and I saw you did the at sign to get to your product. Is that kind of what the process would be for the end user as well?
Otto (32:05)
So you actually don’t even have to type in AdmitPage. That’s just because the chat.tpt version is just not fully released yet. But in Claude, you just ask a legal question. if it’s the type of question that you can answer with case law, then it will automatically switch and use it. You do have to.
Greg Lambert (32:20)
So it just knows you’re a
subscriber to MidPage. I imagine as long you’re using the correct login and it will access that.
Otto (32:31)
Exactly, like of course if you’re in a enterprise, your IT admin will have to like enable the plugin and then and from then onwards, you you just ask a question and then it answers with the page.
Greg Lambert (32:44)
Yeah, looks pretty convenient.
Otto (32:47)
Yeah, and it’s going to be the same experience
Marlene Gebauer (32:48)
It’s name of the game.
Otto (32:49)
very soon, whether it’s from the Word plugins or any place that has AI in it that has a chat. will add our legal data.
Greg Lambert (32:56)
Okay. I want to jump over into some vibe coding. We talked about this briefly earlier and I saw some comments on ex Twitter that you made about vibe coding. said, and I imagine this is mostly for the legal industry that you think there’s a real kind of dramatic underestimating of what
the vibe coding trend is going to mean for this industry. And do you mind just kind of talking about, and I think you mentioned vibe litigation. Does that mean that lawyers stop writing sentences and they start acting more like architects in building these tools as they go?
Otto (33:49)
There’s certainly going to be some of it, right? There will be some of that and that portion will increase every year over the next couple of years. I think that now with coding, we’ve already reached a point where in most young companies, most code is probably 90 % of the code is entirely written by AI and usually the humans still review it. Sometimes they just run a bunch of tests without reading the code specifically. So we really sort of have engineers are now becoming
product managers that are managing just teams of agents that are implementing various features. And I think there will be moments like that where law feels similar. But of course, there’s also a lot more relationship management. There’s a lot more nuance in the law where you can’t just test things before releasing them in the law, right? You have to be sure that this brief is bulletproof. You can’t just find out later and fix it. And so I think the timing of those things will be very different.
Marlene Gebauer (34:36)
It’s interesting with the vibe coming. I’ve been in some conversations where, you know, if people are doing this on their own and creating their own sort of internal solutions and if, you know, they get a wrong result or something doesn’t work and, you know, what’s the, you know, what’s the firm liability for that? You know, since they’ve kind of made it on their own, it’s not something that’s necessarily, you know, firm approved. Any thoughts on that?
Otto (35:00)
Sorry, can you rephrase that?
Marlene Gebauer (35:02)
Sure, ⁓
if people are kind of putting together these internal solutions and they’re for them and it’s not something that’s sort of been reviewed by the IT department. the underlying technology has, but what they’ve developed has not. Where does, if something goes wrong, who’s responsible?
Otto (35:20)
Yeah, yeah, I got it. Sorry. I think that’s a great question too, because everybody’s now thinking, well, I can build stuff myself for the first time and I can build all these cool things that I’ve always dreamed about that I’m only, you know, that only I want to work or so. And I think in many cases, those people are underestimating that the people that are, you know, software companies developing software, they also have the same tools and they can now build a hundred times better. So the bar is rising really quick on what, good your software should work, how many things you should
be able to do. And so I think that in many cases it will be professionals that build and deploy software. And sometimes it will be maybe in a company internal where there’s a business process where people now develop a piece of Gen.AI coded software instead of buying this from a vendor.
But then there’ll be like change management, there’ll be internal evaluation. And because all of those things are important and all of these things are happening today, maybe some of the evaluation will be done by agents and we do that too sometimes. But of course there will be, it’ll be better if you put more focus into it rather than everybody just by coding everything.
Marlene Gebauer (36:26)
It’s like they heard my question. The police are coming. yeah.
Greg Lambert (36:30)
I got one out my window now.
Marlene Gebauer (36:32)
so let’s talk about the, citator. you know, you’ve called the citator, the, the Holy grail of legal research, and you’ve even engaged with Vanderbilt law school, ⁓ to independently audit, audit yours. why is the citator like the ultimate moat? You know, is that because I love the sound effects, you know, is it because the generative, ⁓
Otto (36:48)
you
Marlene Gebauer (36:55)
Part of AI is easy, you know, negative, the negative treatment part is, is harder.
Otto (37:00)
It’s a good question. It’s definitely not the ultimate mode. It’s one of the more hairy problems that we had to deal with. And it’s not just the but also you have to have this big graph of how is each case related to each other case. And there are many things, you there’s a treatment of one judge to the other judge, but there are also other things that are going on there. And the cytotiler is something that you have to be able to trust and it’s hard to do all the verification yourself. And that’s why it has to be really good.
Marlene Gebauer (37:12)
Knowledge Graph.
Otto (37:26)
him.
I think that we will be collaborating more with academia to help them publish reports about how our site data compares to the one of Bloomberg and Vlex and Lexis and Westlaw. As far as we know so far, it is much, much better than the one of Bloomberg and Vlex and is not always as good as the one from Lexis and Westlaw. And this means the Nexus and Westlaw are not at 100%. They are at 90%. Maybe we’re at
sort of, know.
89 or 86 percent or something like that and the other ones are Quite far apart, but I think everybody’s improving their software at the same time I think this is really a moving target and in the future we might have cited as that can tell you not just all this case was Criticized by judge, but it will tell you given that I know exactly what case you’re working on You know these 20 of a thousand cases are very specifically ruled out because ⁓ you know they were overruled for the
the
very specific things that you’re trying to do. And so the AI set is over the future will be much better than the manual side that is that we have today.
Greg Lambert (38:32)
I want to go back to some of the income that the Thompson Reuters and Lexuses out there. And I know that you’ve seen them as well trying to build their own agents. think co-council has been talking about this now well over a year to have these agents inside their own little walled gardens that they have.
You on the other hand are trying to build these pipes that feed into these open agents. So you’ve got walled gardens, pipes. Everyone’s got that in their head now. So how do you see this progressing for you over time? Where do you think the…
Marlene Gebauer (39:10)
It goes out to the open meadow. How’s that?
Greg Lambert (39:19)
while garden approach is gonna go and where do you think the open pipe structure is gonna go?
Otto (39:24)
I th-
I’m not sure if this is the best way to separate the two, because I think that probably Lexus and Rogers will continue as they have done before to do a lot of acquisitions. And they might just switch out their flagship product with a different flagship product by built by someone else. They do have a huge customer base. And I think that today in the times where software can be built so easily, but customer relations and contracts are so much more long lasting.
think that there might even be like, you know, private equity or &A action just to buy other companies for their customers, delete all their software, you know, sadly, maybe also get rid of many of the people who built it and have the customers now use a much better software built by someone else.
And I think there’s a chance of that happening. And I also believe that they will find ways to make their content more valuable because they can maybe now bring it into more places. And maybe they’ll also get more efficient because many things that they were doing manually can also be started to automate. They just might not be the quickest ones to do it. Like I think our data set, you know, still not as big as the one from Rogers because they have many, many countries in all of the world and we only have the U.S. even though there we’re very happy with our coverage.
and we’ll be expanding because we have to do everything software-based. Maybe we’ll be expanding our coverage faster than these alt players will. But I think that they have such a strong position with customers all over the world that they will find ways to leverage that.
Greg Lambert (40:46)
So before we wrap this up, I want to give you a chance. I saw this on another podcast that I listened to. ⁓ Is there anything that we did not cover that you think is really important for the audience to hear right now?
Marlene Gebauer (40:54)
shamelessly stole it. ⁓
Otto (41:02)
I think that they should play around with using ⁓ creating and editing Word documents in Cloud. And you have to have not the legal plugin enabled, but you have to have the code execution enabled. I think for those who are curious, that’s actually really impressive to see.
Greg Lambert (41:17)
Interesting.
So how about you on a personal level? What kind of things are you personally using AI for in your daily life?
Otto (41:28)
you
I love learning about new things and I spend so much time reading about science, about biochemistry, often about the law too, because as a computer scientist in a legal tech field, they have many tasks to do. And it’s just so much fun to learn new things now. And I think whether it’s work or productivity, sorry, work or entertainment, it’s just so much more fun to do many things and to go in more depth.
than be more curious than before.
Greg Lambert (41:54)
Well, we talked about this before, when you were on before. There’s been a lot of advancements in AI and robotics. Do you wish you’d kind of stayed in the robotics field?
Otto (42:03)
I’m
not sure if I should answer that question.
Greg Lambert (42:08)
It just seems exciting, but I mean, what you’re obviously doing is exciting too.
Marlene Gebauer (42:12)
is also exciting.
Otto (42:14)
Yeah, yeah. No, the main reason why I went into this field was, first I had many friends that were lawyers and that were having all these problems. And then also…
It was at the time it felt like software is going so fast. and especially, you know, AI with text is going so fast that this is the place to be right now. And robotics is moving in development cycles that are, know, decades long. And now robotics is starting to accelerate. It’s still much, much slower. but I think if you want to, you know, do things and have fun in life, it’s, nice when things happen quickly. And that’s why I like being a legal tech because, you know, even though everybody thought that lawyers were kind of slow adapters, because it’s such a good fit.
Everything is moving faster.
Marlene Gebauer (42:50)
So, um, we’re, we’re at the crystal ball question now. looking into your crystal ball, you know, the next, I don’t know, three to five years, you know, we’re seeing the rise of a genetic workflows, you know, AI, um, doesn’t just answer the question goes and does the whole job. And, know, in, in, know, in this, in this world, we’re talking about does the, legal research platform, you know, as a standalone website disappear.
do lawyers, you know, stop searching and more like assigning type of work.
Otto (43:22)
I would say it becomes a much smaller part of where you spend your time because agents are going to be doing the groundwork, which is often the legal research, at least the first wave of it. They’re going to be doing all of that. And so you’ll be talking to the results that AI has collected and doing some follow-ups. And so yes, you might not be seeing, you’re not going to walk past somebody’s screens in the office and see them having like the legal research website open all the time, but they will be, you know, they will be looking at the results.
Greg Lambert (43:50)
Otto von Zastrow from MidPage. Thank you very much for jumping on today and navigating the wreckage of the clog crash on the market and ⁓ explaining vibe coding and having a great conversation with us. We appreciate it.
Marlene Gebauer (44:06)
Thank you.
Otto (44:08)
Thanks, you too.
Marlene Gebauer (44:10)
And thanks to all of you, our listeners, for taking the time to listen to the Geek in Review podcast. If you enjoyed the show, please share it with a colleague, especially those who are panicking about their legal tech stocks.
Greg Lambert (44:20)
My 401Ks.
Marlene Gebauer (44:23)
That’s the Easter
egg. Let’s see who listens to the end and like posts that they heard this. You know, we’d love to hear from you on LinkedIn, Blue Sky, or our new Substack channel.
Greg Lambert (44:32)
⁓ Otto, where can people learn more about what you’re doing both with Claude and other things?
Otto (44:39)
Yeah, we have a sub stack as well. We have a website. People can…
Greg Lambert (44:41)
Yay, twins.
Marlene Gebauer (44:43)
All the cool kids
Otto (44:45)
We have a website at midpage.ai and of course we encourage everyone to just like go to the website. There you have instructions on how to install this in Claude and soon in ChatGPT and just give it a spin.
Marlene Gebauer (44:56)
Great. And as always, the music you hear is from Jerry David DeCicca Thank you, Jerry, and have a good day, everybody.