I’m a massive fan of what Filevine is building so it was super fun to chat with Ryan Anderson, founder and CEO, to talk about Filevine’s approach to building AI, their first AI product, Demands.ai, and the future of law.
The Value of a Well-Tailored Platform in an AI world
The practice of law revolves around cases and contracts. Prior to Filevine, legal software mostly revolved around the business of managing a law firm - timesheets, client intake, and more. What this really meant was that there wasn’t a tight link between the legal platform and the work-product: the cases and contracts.
Ryan and Filevine broke through that paradigm and built their entire architecture around the case itself
This means that when a law firm adopts Filevine, they aren’t simply using it as a supplement to their existing workflows. Instead, it transforms the workflows altogether. Law firms live inside of Filevine and the graphs showing user activity that Ryan looks at everyday prove that out.
Because Filevine becomes crucial to the case, Filevine becomes crucial to the firm. And that has enabled Ryan and the team to build all sorts of products that sit on top of this new architecture, including practice management tooling. All of this becomes increasingly valuable because of how tightly integrated everything is to the core work-product itself.
And it’s this platform approach that makes AI tools - literally work-product tools- so valuable.
AI Construction and Utility
There are really two questions that matter for AI tools: construction and context. In both cases, I think Filevine’s approach is a winner.
Construction
Not only does building a platform around the work-product give you increased ability to affect productivity, but it also gives you the right framework to construct useful AIs. AIs out of the box - straight from Stability, Google, or OpenAI - are data hungry. When they don’t have the right data, they’re prone to hallucinations, often imagining facts out of thin air. That’s a non-starter for any business relying upon AIs for important work.
Precision is crucial. Lawyers spend countless hours aggregating facts and interpreting the law. If an AI can’t tap into that - or worse, hallucinates facts - it’s a lemon.
The way to rectify this is by finetuning AIs with the right data, and Filevine has a wealth of legal data.
With that data, Filevine is able to create accurate and valuable AI products for their clients. And that’s what they’ve done: finetuning base models to make them actually useful to legal practice.
Context
“AI has to come to the work, not work to the AI.”
- Ryan Anderson
Data doesn’t only matter in the fine-tuning stage, it matters in the act of generation as well. In order to get useful work-product, I have to give AIs the right set of facts. If I use an external AI, I’m going to be spending time compiling and uploading documents, correcting errors, and more.
But if I am already on a platform containing all my case documents, client information, and more, I get to start using the AI instantaneously. With Filevine, AI comes to the work with with their first AI product, Demands.ai.
Demands.ai
Law is full of bottlenecks. Some of these bottlenecks are intentional; due process takes time. Others unnecessarily delay the resolution of a legal matter. Ryan thinks that Filevine should exist to remove as many of these bottlenecks as possible.
Demand letters happen to be the perfect bottleneck for AI to remove.
Every case starts with a demand letter, detailing damages, the facts of the case, and the legal precedent for the action.
But while important, demand letters aren’t where a lot of a lawyer’s value lies. As Ryan puts it, that’s instead in the strategizing, evidence gathering, thinking, and subsequent negotiations.
But it’s still important as it’s typically the first document a client, a legal counterparty, and potential third parties will review. The higher quality the demand, the more likely a case is settled.
The actual drafting is time-intensive and often not for the right reasons. Demands are really the compilation stage of all the previous evidence and fact gathering a lawyer has done. This compilation looks like spending additional hours combing through all of these materials to develop medical summaries, damage claims, and a narrative around the evidence. Sure it has independent value as a presentation of legal claims to counterparties, but it’s mostly a bottleneck. It limits the throughput of the firm and a lawyer’s time spent in other high value matters.
But what if you could outsource the drafting entirely? And what if you also knew that the demand would contain a robust presentation of all the facts, damages, and legal precedent you have established with little to no drafting work done by your firm?
That’s the future Filevine is building with Demands.ai.
Demands.ai works by ingesting medical records, case research, really any document that would feature in a demand letter.
Instead of having a lawyer in the loop acting as the compiler, AI comes in and does it faster and with the same or better accuracy. Demands.ai then creates medical summaries and synthesizes data to estimate damage claims.
In short, it really is a legal compiler. Except the beauty here, is that it’s a compiler not from a programming language to a machine, but a compiler from unstructured legal research to a structured document.
With that said, in the same manner that modern programming languages unlocked greater developer productivity due to the strength of the compilers undergirding them, there’s a similar dynamic going on here.1
When you don’t have to worry about a lot of the compilation stage, engineers spend a lot more time thinking about how to design the system to achieve the necessary result.
That’s the same dynamic we are going to witness here: as lawyers spend less pure hours on document compilation, they can spend more time thinking about how to design the case strategy to achieve the best result. They’re still intimately involved in the actual arguments and theories of the case spelled out in actual documents, but their role isn’t to build out citations, generate medical summaries from the evidence gathered, and more - it’s far more expansive.
From there, once Demands.ai is initiated, a Filevine legal partner will assist in putting the finishing touches on the letter and ensure it’s up to par.
Lawyers end up with a high-quality demand letter in record time and almost instantly reap the benefits of a new AI paradigm.
Word, Work-Product, and .vine
Lawyers have been in Microsoft Word forever. Filevine wants to deprecate it for legal use. In many ways, the same line of logic that applies to AI also applies here.
The ideal legal editor should bidirectionally sync up with my case management solution. I should be able to draft documents that save into a case file, but I should also be able to directly parse information from a case into my text editor.
So far that’s what Filevine has built: a bi-directional system of record. But now with AI, this system of record gets supercharged and taken to the next level.
Nat Friedman had a great interview with Dwarkesh Patel where he talked about Github CoPilot and the development process. In hindsight, Copilot looks obvious - of course auto-complete is the right approach for engineers. But Copilot didn’t start there, instead looking at other form factors for the tool. But auto-complete turned out to be the perfect form factor.
Ryan thinks that this sort of autocomplete is the form factor that lawyers want as well. And they’re empowering their text editor with some of this predictive power. It’s already built with templating and bidirectional sync. With AI, Filevine ends up with what I’m deeming “tridirectional sync”, where work can be initiated anywhere and update everywhere and AI has increased scope to dramatically impact productivity.
I don’t mean to suggest that this vision is fully realized yet, but you’d be surprised by how much progress Filevine has made, and how rapidly they can accelerate because of their existing infrastructure.
In the Demands.ai example and in the fullness of time, imagine that you obtain an AI-generated demand letter, and get some new medical records that you’d like to add in. You can initiate this in many different ways. You could prompt an AI directly to look at the case data and update any documents. But more likely, you could simply prompt a Copilot-esque tool inside the text editor to update the medical summary table directly.
In my view, this sort of tridirectional syncing is the future of platforms and I think Filevine is the par exemple of what’s to come.
The Future of Law
Every lawyer and every legal technologist is attempting to figure out what AI means for the future of law. There’s been no shortage of prognostications on the death of the legal industry, but Ryan sees it a bit differently.
He starts from the premise that there isn’t a set amount of legal work in the world. Legal access is a big problem and plenty of cases, contracts, advisory, and more doesn’t happen because of how expensive legal work can get. There’s latent demand due to a pricing problem.
And at the same time, plenty of lawyers get burnt out as they are unable to scale their work without inputting raw hours.
But if both sides of this problem change - more simple legal work gets outsourced to AI and lawyers are finally able to focus far more energy on high-value strategy and advisory rather than rote work - you could imagine both clients and lawyers reaching far higher levels of fulfillment in their interactions.
That in turn will create new pricing models, new firm structures, and potentially a cottage industry of micro-firms running out of Filevine to serve increasingly more niche areas of legal practice.
There are plenty of other downstream implications from AI, first the ABA is probably not going away - data security and privacy are going to become even more pressing and regulating who can practice law is going to take on increased priority when bots are running amuck.
But the future Filevine is building is one where for the first time, legal work can scale, and the legal industry starts to move a lot faster.
There’s a popular notion in the crypto world that “law is code.” I don’t think you need to buy into the formalism implied to note the similarities.