An AI Start Up That Revolutionizes Patent Litigation

John Quinn is joined by Caleb Harris, Co-Founder and CEO of &AI, a startup focused on using artificial intelligence to transform patent litigation.   They discuss how &AI uses AI to accomplish complex patent litigation tasks such as invalidity and infringement analysis, dramatically reducing the time and cost associated with these traditionally labor-intensive efforts.  The service features four components: searches for prior art or infringing products, in-depth legal analysis (including creating claim charts), drafting litigation-ready documents like invalidity contentions or IPR petitions, and automating workflows using AI agents that operate independently.

Patent litigation is particularly well-suited to AI because so much of the underlying data—such as patent filings, litigation histories, and prosecution records—is publicly available.  &AI continuously updates its data sets and can provide summaries, detailed claim charts, and customized drafts in as little as 10 minutes.  Unlike generative AI tools, &AI minimizes hallucinations by relying heavily on document retrieval rather than generation, and by providing verified citations in its output.

The platform can also help streamline early-stage litigation decisions, such as assessing the strength of a patent portfolio or evaluating potential infringement claims in the marketplace.  It also helps defense teams efficiently assess and respond to weak claims, including those from patent trolls, by producing tailored response letters and evidence.

&AI uses AI agents—AI that develops multi-step plans to accomplish tasks and automatically adjusts those plans based on how the work is progressing.  This allows the user to focus on the end product they want rather than the steps needed to get there.  AI agents will enable faster, more scalable, and more economically viable litigation, especially patent litigation.  This may lead to a boon for litigators as more lawsuits are filed and resolved quickly.  Although human performance will remain crucial in areas like persuading a jury or a judge, law firms may gain a competitive edge by pairing their expertise with firm-specific AI tools trained on the firm’s proprietary data and preferred styles.

Podcast Link: Law-disrupted.fm
Host: John B. Quinn
Producer: Alexis Hyde
Music and Editing by: Alexander Rossi

Transcript

Note: This transcript is generated from a recorded conversation and may contain errors or omissions. It has been edited for clarity but may not fully capture the original intent or context. For accurate interpretation, please refer to the original audio.

JOHN QUINN: This is John Quinn and this is Law, disrupted and today's podcast is going to be another podcast about applications for artificial intelligence in the legal field and legal disputes field in particular. And we'll be speaking with Caleb Harris, who is the founder and CEO of an early-stage company, I think we can say Caleb.

Yeah, fairly early stage, about a year and a half old, early stage company, which has a first round seed capital of 6.5 million. Yep, and you're a graduate of Y Combinator for the summer of 2024 batch as I understand it. 

CALEB HARRIS: Yeah, just kind of completed Y Combinator, little less than a year ago.

JOHN QUINN: All right. What was that experience like by the way?

CALEB HARRIS: Y Combinator I think is an excellent program, especially for people who, like myself, did not have prior experience building a company. They, really kind of, you know, launch you very effectively. I think the mentorship, the program, everything is top-notch and definitely worth, you know, the equity you have to give up to do it.

JOHN QUINN: Alright, to get back to your company and AI, you really focus on the patent world. As I understand it, it's an AI platform that uses AI agents and we'll talk about that, for invalidity and infringement analysis in patent litigation. Is that a fair summary? Yep, exactly. And can you break that down for us?

What are the different work streams or, you know, focuses that AI can bring to patent litigation? 

CALEB HARRIS:  Yeah, so & AI has really focused on, I would say four main areas. The first is a search component, so searching for prior art or potentially infringing products. The second kind, the meat of the analysis, so that kind of comes into play with building claim charts. Also, drafting things like invalidity infringement, contentions, understanding kind of the scope of the analysis. 

The third component is, I would say, the drafting aspect of it, so this is where we take that analysis, those materials, and turn them into things that you would use in litigation or pre-litigation.

This can look like drafting full invalidity contentions, making pitches for clients. You know, IPR petitions, that area is extremely flexible and it uses the context of the rest of, you know, the analysis and documents you have. And then the last component is wrapping that into like a platform and specifically using AI agents.

And I think that kind of really unlocks some new workflows where you can spin up an AI agent that's just able to do these things. In the background, you can focus on other work and then eventually get a notification that, you know, your invalidity contentions are complete, the analysis is done and so on.

JOHN QUINN: So let's start at the first thing we mentioned, validity analysis. I mean, what data sets do you have that enables you to do validity or invalidity analysis? 

CALEB HARRIS: Yeah, so fortunately being in the patent and IP space, almost everything is publicly available. So in terms of creating ways to validate our performance, that's really looking at, you know, historical data on prior litigations, previous IPRs, everything like that.

So we were able to just grab that as it's publicly available and kind of condense that into something that's able to be, you know, used and validated through our AI systems. 

JOHN QUINN: Is this a data set that's constantly being updated then? 

CALEB HARRIS: Yep. Every time that there's new information, new litigations, new IPRs, even like understanding prosecution history, you know, back and forth between examiners on the validity of patents, all of that's like very relevant information that we can use to strengthen our results.

JOHN QUINN: If a client comes to you and identifies some patents and says, hey, we'd really like a read, on the validity of these patents. How long does it take you to accomplish that analysis, and what does the output, the work product look like?

CALEB HARRIS: So that's kind of the common use case and it really only takes a couple minutes.

So the process would be you upload those patents to the system, either you have your prior art or we're able to search for it. That's almost instantaneous given we've pre-processed all global patents. Then with those patents, we're able to analyze them, build out work products like claim charts, as well as summaries of claim charts, and lastly, draft that into additional work products like invalidity contentions end to end.

That entire process is, you know, 10 minutes, I would say and that's amazing. Yeah, so it should be incredibly high time savings with that to get you to a reliably high quality work product or set of work products. 

JOHN QUINN: It is a hallucination. We've all heard about hallucination in large language model outputs.

Is that an issue here for your product? 

CALEB HARRIS: No, so there's a couple ways in which we navigate hallucination. First off, many of the tasks that we're doing are actually retrieval tasks, search and retrieval tasks, not generation tasks, so searching for prior art, building out a claim chart, those are both like grabbing existing citations and references.

The aspect that hallucination would come into play would be something that's fully generative, but in something like a litigation document or a claim chart, you're not going to generate the citations. You're going to retrieve them and that completely eliminates the opportunity for hallucination and then with other work products like drafting invalidity contentions. We have strict, you know, quality assurance going through in the back end to make sure that whatever output we're putting is very much grounded in the verified citations and, you know, reliable information that we're seeing from the source documents.

So in its entirety, the system is very robust against hallucination and in almost every situation there's like a 0% probability of any issues there. 

JOHN QUINN: So you would give that client then, an output, which could include just analysis of prior art or whatever. Or it might also go to claim charts, or what other things might be deliverables.

CALEB HARRIS: Really any type of document that is related to the validity or infringement of the patents and products in prior art. So effectively, on the nDAI platform, in addition to something like a claim chart, we're really able to create drafts of any type of document that you would want. I think that's one of the benefits of current language models is they're extremely flexible in creating what you want and that means we can generate anything from invalidity contentions to pitches for clients, to IPR petitions, to letters, memos, executive summaries. It doesn't really matter in the end, it's extremely flexible, so it's up to like what you would want to create. 

JOHN QUINN: I mean, you must have a good sense of what else is available in the marketplace in this area that does what you do or approximate what you do.

I mean, I've never heard of another service that can do this type of analysis so quickly. 

CALEB HARRIS: Yeah, there's, I would say it's kind of built off new technology and hence most, you know, solution providers haven't had time to build something. There are a few startups of a similar stage to us that are like in the same space. But really I would say as a focus, we are very much focused on litigation, whereas many of the other patent related startups are also tackling things like preparation and prosecution or transactions, where they're not going as deep on litigation. 

JOHN QUINN: Right, alright. The second thing you mentioned was infringement analysis. Tell us what AI can do in terms of infringement analysis.

CALEB HARRIS:. So it very much mimics the type of analysis that we're doing for invalidity.  In this case, instead of searching for prior arts, we're searching for products on the market that may be infringing the set of patents and then with those we're able to generate evidence of use charts.

And later on we're also building, you know, work to understand the potential damages that, you know, may have accrued and also the other work products like in the infringement, contentions,  drafting complaints, really any of those additional work products. 

JOHN QUINN: How was, tell us, how the system does an infringement analysis. What is the data that you're feeding it, and how does it go about doing an infringement analysis and what does the output look like? 

CALEB HARRIS: Mm-hmm. Yeah, so part of the infringement analysis, I would say the most difficult part is finding the products that may be infringing.

That requires a very sophisticated search structure that largely mimics how a human would do it. So we have an AI agent that is able to access the internet and kind of iteratively looks across the internet that this can take, you know, in some cases, even a day of analysis where it's looking for anything that might be relevant.

And then once you find a potential target, analyzing it is trying to get as much information as you can. I would say one of the difficulties with this is that it's a little bit limited by the inability to do something like a tear down, typically an infringement analysis. You know, if you have a physical product, you might tear it down, understand exactly what's happening inside, you can't do that in this case. Yeah. 

But a lot of products have, you know, detailed specifications, user manuals, you know, warranties, these other types of information that you can find and through that analysis, then it's a matter of taking that to the language model.

Understanding where and how there might be evidence of use, and detailing that in the form of an infringement chart and in infringement contentions. 

JOHN QUINN: That's amazing. Because you know, it's one thing, and we've done this many times for universities, just by way of example, or companies that have large patent portfolios, it's one thing to first find a way to go through and find high quality patents. And that's another question I wanna ask you, they're not just valid, but they're high quality and kind of force ranking them, which is what we do. We do this all the time and then the next step is going down to the marketplace and seeing what potential candidates there are for infringement and is it worth pursuing them from a marketing, you know, from a standpoint, a standpoint of revenue.

What are the best candidates? I mean, can your program do that last part as well, help identify, you know, use the infringing uses that have the most potential for recovering revenue if you can license them or if you have to bring a lawsuit. 

CALEB HARRIS: Yeah, exactly. That's something that I would say we're in the earlier stages of but that kind of vision is what we're working towards.

The ability to take, for example, a large portfolio from your client, let's say a thousand patents. Understand, you know, kind of the maximum valid in high quality patents and the maximum potential damages in a litigation with a strong, you know, infringement target. Figuring that out across, you know, a portfolio of a thousand patents manually is extremely difficult.

It takes an immense amount of time. The analysis is very tricky, but being able to do that at scale very efficiently through the use of AI, is an extremely high value add and I think that's kind of where this type of system will go in terms of benefiting the firms and the clients who are adopting it.

JOHN QUINN: Right. I mean, I know when we do this type of analysis, we look at what patents are cited the most, in terms of what technology seems to be, where is the thickest thickest of patent technology, what are the outliers and the like, and try to come up with some kind of force-ranking system. What are the best candidates in terms of the strength of the patents and less crowded kind of fields?

And then the second step of looking at the marketplace. Exactly. So we kind of mimic that workflow on the backend. You must have done some comparisons between the non-AI, old-school way of doing this and what and AI can come up with.

CALEB HARRIS: I guess looking at for example, IPRs or previous litigations, that's kind of the way in which we're able to validate that our performance is good or even better than you know what existed and I would definitely say it is better, in terms of like a validation perspective.

And another benefit is instead of taking two or three weeks to get your, you know, search results and your analysis, having the ability to get it almost instantly, the instant feedback to help you drive your own search and, you know, strategy, is an extremely high value add in itself. 

JOHN QUINN: So I would think that this would be really useful in sort of handicapping potential outcomes in litigation.

And of course, every litigation comes down to what's presented in a courtroom and there's, you know, personality dynamics and jury composition and who is the judge and all the rest of it, a lot of other variables that you probably cannot model. But it seems that your output would be useful in trying to handle handicapping, issues of validity and infringement, or at least as a starting point.

CALEB HARRIS: Yes, exactly. It kind of works on both sides. It works from the plaintiff perspective of identifying targets that are of high quality, but also for, you know. Disputes against, you know, low quality NPE’s or trolls, or, you know, cases that might not have very good grounds. You can almost instantly get high quality evidence against them you know, handicap it, stop it very short.

And I think something like, you know, responding to a patent troll is actually a very good example of how this technology can be used because the patent troll may be asking for a settlement that's, you know, right on the cusp of like, is this worth it? Even entering into a litigation, or is this worth much analysis to respond or should we just settle?

Whereas using a system like this, you could get almost immediate low cost feedback that okay, there's nothing of high quality grounds here, we can send a cease and desist or a response letter to kind of end this very shortly.

JOHN QUINN: What other things should we know about an AI service?

CALEB HARRIS: I think the thing that we're kind of building towards is, I would say the future of how AI agents fit into law, and specifically patent law and litigation.

We've taken like a big, I would say, bet towards this agent approach, and we're starting to see the ,you know, returns from it.

JOHN QUINN: Let's explain what that is. What is an AI agent in practical terms,what distinguishes an agentic system from, you know, traditional AI tools, which many law firms might be using now.

CALEB HARRIS: Yeah so an AI agent, in short, is a system that is able to dynamically respond to, dynamically update its plan and navigate the world given its new state. So in simple terms, what that means is that its traditional AI system will be able to respond to your question, I would say one attempt, maybe with a reasoning model, it's thinking through its response before giving it to you.

But in contrast, an AI agent is able to create a plan and use tools to make its response as best as possible. So in our case, how that might look is instead of you going through every step and saying, okay, now search for prior arch that discloses this and now build me a claim chart with a focus on these, you could just say some end task directive, like, I need to build invalidity contentions for these patents, and the AI agent is able to figure out. Okay, what's required to build invalidity contentions? We're going to need to search for a prior arch. We're going to need to chart the references, we're gonna need to analyze those charts.

Maybe we need to do that again. If we don't have strong evidence, we'll figure out how to combine the references. Which one would make a good one? Oh two, if there's a good one, oh three. It figures out this entire plan and goes through it each step as it makes, you know, more progress. It reevaluates what's happening and is able to change its path, change its strategy.

So that's a huge unlock because it takes away much of the workflow pressure from the user themself and they can start going into just, you know, what do I want in the end? And the agent will figure out how to get there. In this way, if you're familiar with Sam Altman, I would say at the end of 2024, he made this prediction that by the end of 2025, AI agents would be, you know, working at companies like, you know, collaborators.

And I think we're actually on track for, you know, systems that feel like coworkers, where you could send them a message that, you know, I need to draft this, by the end of day, can you do that analysis for me? And it will figure out everything that's required to do that and give you that deliverable.

And that's kind of this, the way in which we're extending the platform more so than like, you know, everyone has to be in it, doing it themself, like traditional software has been, since the existence of software. 

JOHN QUINN: So you can just basically tell the AI, what you want, the natural language query. This is what I need, can you do that or, yep, you don't have to say please, I mean you don't have to be polite. 

CALEB HARRIS: No, you don't have to, actually, it's a funny aside that. AI systems work better when you are not nice to them. So if you say, you know, you better do this, or you're gonna be unplugged or replaced.

JOHN QUINN: Yeah. Well, you saw that story about the Anthropic, new Claude model that they fed some material to suggesting that the model might be replaced. And also some emails about how one of the handlers was involved in the affair and the model started to blackmail the handler. You saw that? 

CALEB HARRIS: Yeah, I saw that.

That was really interesting research. Unfortunately all of that is like being removed from end productionized models, so you know, they've handled it, but it's interesting that they observed that in the, you know, research environment. That's kind of what everybody fears, right? Yeah, calling the police on you if it discovers, like, yeah.

JOHN QUINN: So what other, you know, frontier model breakthroughs do you think are most relevant to the courtroom? What lawyers do in courtrooms today? What other things are going on that you think are super interesting and what does the future hold? 

CALEB HARRIS: I think the big unlock has been reasoning models if you've used, you know, chat GPT or Claude, the ones that have like thinking capability,  that's a massive unlock for anything related to, or really everything including law because the models have an opportunity to really reason through all the potential arguments, the strategy, craft the best answer before responding.

Previously, that kind of reasoning was lost by just instantly responding. But now there's really an opportunity for, you know, complex understanding of things. In addition to that, there's some more technical things in the background that are enabling models to access new data efficiently, this would be called a model context protocol. This is very important because it allows agents, for example, to, you know, access data on all of patent law or the USPTO systems or any of these different data sources that it might be able to use to make more, you know, nuanced, valid opinions. Those will all have dramatic effects across law where especially, you know, having something that is well researched and cited is incredibly important.

JOHN QUINN: Well I mean, where do you see real growth in AI capabilities that will pose the sort of greatest opportunities for litigators beyond agents and reasoning models? What else?

CALEB HARRIS: I think just the ability to do things at scale cheaply is a massive unlock right now, especially in litigation, patent litigation. The barrier to entry to litigation is immense, the costs incurred during the litigation are immense, but it's all kind of worth it because the damages are in the tens, hundreds of millions. I would say like as we move forward, that barrier to entry will dramatically drop, you know, but the damages will stay the same which makes this like a really interesting environment where, you know, maybe alongside things like litigation finance, you have this fairly liquid market for entering and initiating litigations, where they can be resolved efficiently, but still have these massive damage awards. I think that's kind of where AI is taking things in litigation and I think it will be very interesting to see how it starts to resolve as every party kind of adopts efficient systems, you know, the plaintiff side, the defense, the individual, you know, companies that are involved, litigation, finance, there'll be more and more of a, you know, liquid system for initiating these, getting them settled, doing it efficiently and, you know, finding more opportunities that weren't previously able to be found because it's too difficult to look through.

JOHN QUINN: I mean, are you saying you think this will mean more and earlier settlements? As everyone is using AI and has access to the same AI capabilities. 

CALEB HARRIS: Yes, I think the number of litigations will grow exponentially and I think they will be faster, the numbers will grow from the actual dispute. I think top line revenue across all of litigation will dramatically increase because right now we're limited to the number of, you know, infringement opportunities that are found manually, which is incredibly difficult.

But if that part is, you know, smoothed out where everyone can kind of figure out exactly, you know, what products they have, some, you know, royalty rights to, and then they can do that litigation efficiently, both parties can understand and resolve it quickly.  The damages still exist, the damages are really, you know, related to the product sales, which is not influenced by any of this, you know, background inefficiency in litigation, initiating it, scaling it, and so on.

So I think that's where things will go. I think the entire market of patent litigation, specifically IP litigation, will start to boom because of this.

JOHN QUINN: Well, I mean, everything you're saying, I think applies to litigation, generally personal injury litigation. But still there are those factors that we referenced earlier that are hard to model.

Maybe you can't model, you know, the personalities of the speakers in the courtroom, the, you know, what Aristotle called the ethos of the speaker, which he said in the rhetoric is the most compelling, most important part of persuasion is the ethos of the speaker, who's in the jury box and who is the judge.

Do you see a time where those things can be modeled too? Because I mean, if you can model all that, then you really can get to super efficient resolutions of disputes if you can get certainty around all these different variables and then we'll all be outta jobs. 

CALEB HARRIS: Not anytime soon will we see that, that level of, you know, understanding.

And I think that's really where the law firm themself come in as, you know, experts in these categories. The ability to be persuasive to the jury, the ability to model these things effectively, and the personalities and so on. Like you can maybe get statistics on these things, like, you know, what are the demographics of this area? Or what is the history of this judge? 

JOHN QUINN: Yeah. That information is available. There's an offering, you may have heard of called Predicta. Are you familiar with that? They can tell you with 85 plus percent accuracy, whether a motion to dismiss will be granted or not, but if you just give them the case number, all they need is the case number.

You know, they've got all these other data points that they can get, or they have the case number, who are the lawyers? How big are the law firms? Who's the judge? Where did people go to law school? What's the form, et cetera. Just the case number. From that, they can derive all that data and they can tell and I know they were working on some other types of motions, like a motion for change of venue and the like.

CALEB HARRIS: Yeah, I think that kind of analytics is definitely very valuable and strong, but then the personality side of it, like, I don't see that, you know, being able to be modeled, efficiently at scale anytime soon. Right. And that really, you know adds value to the specific litigators involved, the firm, their experience with, you know, the jurisdiction and the judge, and, you know.

JOHN QUINN: How can law firms differentiate themselves if everyone in the industry is adopting these same AI tools? 

CALEB HARRIS: I think there's like a couple ways. One which I just mentioned is that like, you know, the experience in these actual settings, which can't be modeled.  The second aspect is there are different areas that have data advantages, for example, you know, prolific law firms may have a lot of internal data, which to someone like me, an AI company, is like gold. Like we're looking for data that we can use to generate style, validate performance and so on, this can happen through a bunch of technical product processes in the background.

Things like, fine tuning or you know, style and engineering. But then we can provide that, you know at a singular level to a single customer and I think that start, that style of differentiation will become increasingly common, where there is a Quinn Emanuel style AI system that does things, how you want them to be in your, you know, according to your data, up to your standards, with your style and your formatting and all these different things, and that will be a large differentiator in addition to, you know, the actual experience of the litigators in these venues with these people on these subject matters and so on. 

JOHN QUINN: Fascinating discussion.

Caleb. Thank you very much for joining us. We've been speaking with Caleb Harris, who's the founder and CEO of & AI, an AI company focused on specifically on patent litigation. Caleb, how can people best find you if they're interested in learning more?

CALEB HARRIS: You can just search me up on LinkedIn. It's probably the best place to contact me or message me at caleb@tryandai.com.

JOHN QUINN: Thank you very much. This is John Quinn and this has been Law, disrupted.

Thank you for listening to Law, disrupted with me, John Quinn. If you enjoyed the show, please subscribe and leave a rating and review on your chosen podcast app. To stay up to date with the latest episodes, you can sign up for email alerts at our website, law-disrupted fm, or follow me on X at JB Q Law or at Quinn Emanuel.

Thank you for tuning in.