LitigusAI Blog · The Origin Story
The AI associate I was sold didn't exist, so I built LitigusAI.
Casetext’s CoCounsel was a great platform for its time, and I was a happy customer. Thomson Reuters bought it for $650 million and, in my opinion, ruined it with guardrails. I told my Westlaw rep exactly how I felt about that. The answer was a pitch for their new agentic platform, at a premium price, as if agentic were a magic word. My wallet opened, and in September 2025 I signed a contract that committed me to a one-year term at $1,450 a month.
What I had was a chatbot bolted to a document vault. Half my files wouldn't upload; they were too large or in the wrong format. The "workflows" were pre-canned prompts. It couldn't read a client file from my document system, couldn't produce a paper in my firm's format or any court's, and couldn't do the one thing an associate actually does: take an assignment in plain English, find the documents where they live, come back with questions and a plan, and then build the papers. My renewal and my first line of LitigusAI code are dated the same month.
Maybe you've been burned by one of those promises too. Or maybe you've never trusted AI with anything more than smarter searching, which, given what's been on offer, is fair. Either way, here's what happened next: I was told a real AI associate couldn't be built yet. I'm a litigator who refused to accept that answer.
November 2022
I didn't realize it at the time, but I started building LitigusAI in November 2022, when I sent “Hi there!” to ChatGPT-3.5.
I saw the future, and it was bright. I wasn't new to legal AI, either; I'd spent years using machine learning in discovery: technology-assisted review, the "expert systems" generation. Good tools, nothing like what was coming. You couldn't build a real AI associate on GPT-3.5, but the future was crystal clear. I needed that future more than most.
I'd spent my career in large firms before building my own practice, and solo law is eat-what-you-kill: land a big matter and you've won the work, all of it, on top of all the work for every other client. I love what I do. I needed help doing it. I tried hiring; finding an associate who'd put in the hours, do excellent work, stay for the long haul, and want a salary a solo can pay is nearly impossible. A colleague who's brilliant, always available, and can be in five cases at once? Science fiction, with a path I could see.
I did what a technologist does with a few years of runway before the technology arrives: I prepared. I read everything I could find on capabilities, on failure modes, on how these models go wrong and how builders were learning to keep them right. I built small things to learn: toy games, experimental tools, multi-model debate systems where AI agents argued against each other. I spent hundreds of hours interrogating the frontier models themselves about how to build with them, theory first, then practice. By late 2025 I was building in earnest: a discovery platform for my own cases, then the agent framework, then LitigusAI itself.
The broken promise stung because I knew exactly what was possible; I'd spent three years watching it become possible. It didn't stop me. LitigusAI isn't a pivot into a hot market. It's the thing I was waiting to build.
Instructions don't govern AI; architecture does.
Here's the honest version of how a lawyer builds a platform: in stages, with a lot of failure. I tried coding with GPT-4, then o1, then each new Claude as it shipped: fun to play with, but you couldn't build anything genuinely useful. Opus 4.1 helped. Sonnet 4.5 dropped in the fall of 2025, and that was the oh-shit moment: the associate could be brought to life, now.
What I brought wasn't code; it was something older. I'd litigated a few software cases over the years, patent, copyright, and trade secret disputes that each ran long enough to teach me how systems come together on paper; learning it for a case, it turns out, is not the same as building one. I've been taking things apart my whole life, cars and electronics alike, and a workflow comes apart the same way. I sat down with every workflow I ran in my practice, broke each one into its smallest parts, and tried to get a model to perform each part in sequence. I failed, changed the approach, and tried again. As the models got stronger and I learned how the pieces fit, the experiments started working. One piece at a time.
Out of the failures came the two ideas at LitigusAI's foundation. Instructions don't govern models; a prompt is a request, and the models treat it that way. I stopped asking and started building gates: checkpoints the work cannot pass until it meets the requirement. I also found a way for two live AI sessions, from different labs, to talk with each other; suddenly a drafter could summon its own reviewer without me relaying messages in the middle. Cross-model review stopped being a step I ran by hand and became a step LitigusAI runs on every draft, before the work ever reaches the attorney. My own review never moved; it sits where it always has, at the end, on everything.
The last piece arrived right on schedule, when the industry converged on an open standard that plugs agents into files, tools, and courts. When it snapped into place, I was holding the associate I'd imagined in 2022. I'll admit it: I was surprised.
The commercial realization came next: nothing like it existed for lawyers. The industry sells cloud portals where you upload your documents into someone else's database and fight file-size limits; the tools built the right way, working on your machine and in your files, were made for programmers. My homemade version changed my practice, but it is a build-it-yourself project, not a product.
The gap is the company. LitigusAI is the same idea with the hard parts automated: no command line, no configuration files, no learning curve. You get a concierge conversation, your existing folders, and an associate that goes to work.
Security wasn't a feature; it was the starting point.
Every design decision started from two questions: where do the client's files live, and how do we make this platform fast, comfortable, and, most importantly, secure? The first answer never changed. Your files stay on your machine, in your folders, in your custody. They are never saved or stored on LitigusAI's hardware or any provider's; we keep no database of your work product or client documents. Your prompts and outputs are never used to train any model and are never sold or shared. Every AI provider that touches client content is based in the United States or the European Union and operates under strict zero-data-retention terms: nothing stored, nothing surviving the request.
The architecture that works best for litigation happens to be the most secure one: a file that never leaves your machine cannot be taken from someone else's. My clients trust me with their secrets. I built the platform I was willing to point at my own files, and yours gets the same standard. A full security deep dive is coming; this is the principle underneath it.
The promise
LitigusAI does the preparation; the judgment stays yours. It hands you work that reads like a strong mid-level associate's, attorney review is a stage of the pipeline rather than a footnote, and what you get back is the other 80–90 percent of the hours a set of papers used to take. It changed my nights, my deadlines, and my business.
Every day my team and I make LitigusAI better, and heading into beta we went a little over the top: the platform learns from its own work, drafts in your voice, and sets itself up through a conversation about how you practice. Beneath it all sits the Litigus Library, centuries deep and built for AI first: LitigusAI can run more searches in a minute than you could in a week, nothing rate-limits them, and context compression keeps every turn lean.
None of it is there for show. Every feature exists so the friction disappears: no uploads, no configuration files, no learning curve, no waiting on a vendor's meter. You practice law; LitigusAI handles the rest.
We are just getting started, but the hard part is done: LitigusAI exists. Three years ago I typed “Hi there!” into a chatbot and saw the future; today it drafts complete sets of pleadings, motions, and discovery, and then hands me back time to build my practice and spend time with my family. Let LitigusAI do the same for you.
The private beta is open for applications. If you are interested in beta testing LitigusAI at no cost to you, please submit an application. Seats are limited, and we review applications as they open.