I Am the AI Evangelist at a 50-Year-Old Plaintiff Firm

I did not come up with this analogy, but it is the best one I have heard. Tony Stark is a brilliant guy on his own. But when he puts on the Iron Man suit, everything good about him gets turned up to 11. That is what AI should do for lawyers. Not replace them. Amplify them.

I am the person at Cofman Townsley whose job it is to hand out the suits. Our firm has been around since 1973. Over a billion dollars recovered for clients. Just north of 50 employees. We have people who use ChatGPT daily and people who still think AI is a fancier version of Google. Getting all of them to the same place is my problem to solve.

The Tool Stack

When I came to Cofman Townsley in 2024, the firm was using an AI product for medical chronologies and demand drafting. Fairly rudimentary. I pushed it past its boundaries, found the ceiling fast, and started looking for something with more flexibility.

We moved to Claude. The Commercial Teams safeguards cover data usage, with no training on our data and no human review. I built workflows for chronologies, case reviews, and demands. I trained it to write in our voice, the Cofman Townsley tone, the way we actually want to come across to adjusters and opposing counsel. The output was strong. But Claude does not integrate with our case management system. Every demand required sourcing materials, managing documents, and handling encoding and size constraints. It worked. It was also a pain.

That friction led us to Eve. What Eve eliminates is the manual layer between case data and the AI. Everything is already connected. You ask natural language questions of your entire case file without uploading a single document.

I had a situation recently where someone asked me about a potential umbrella policy on a case. I knew I had made a note about it somewhere but could not remember the source. I asked Eve. Within seconds: the acknowledgment letter from State Farm referenced a personal liability umbrella. That kind of instant recall across thousands of pages is not a convenience. It is a competitive advantage.

Having the right tools is one problem. Getting 50 people to use them is a different one. My approach has been to recruit for curiosity, not technical skill. The people who say "I do not fully get this yet, but I want to test it" are your best advocates. They break things, they give real feedback, and they get excited in a way that spreads. No single evangelist can move a whole firm. But you can create the conditions where adoption becomes self-sustaining.

Game Theory in the Demand Chair

For most of my career, negotiation was gut feel. They offer this, I counter with that, and we go back and forth based on experience and instinct. At some point I started thinking there has to be a more analytical way to approach it.

So I started studying game theory. Formal negotiation frameworks. Concepts like BATNA: best alternative to a negotiated agreement. I built prompts in Claude where I feed in the variables — policy limits, maximum potential recovery if we file suit, anticipated litigation expenses on both sides — and let it model where the real leverage sits.

Then instead of just trading numbers, I can say to an adjuster: you are going to spend $50,000 on defense counsel to litigate this case. Take a haircut now, save that money, and I put more in my client's pocket. Everyone walks away.

Beyond the opening math, I track movement patterns across a negotiation. If their offers jumped 10% after a particular argument when prior movement was only 3%, that tells me something. If each subsequent offer is shrinking, I can anticipate where they will land and decide whether now is the time to threaten litigation.

The human judgment still drives the final call. But now the gut has data behind it.

What I Miss

I work pre-litigation at Cofman Townsley, which means I am building the infrastructure that makes cases stronger before they ever get near a courtroom.

The most underrated thing AI does is catch what I would not. Page 23 of a physical therapy record mentions a referral to pain management, but there are no pain management records in the file. A gap like that can mean missing treatment that strengthens a claim. AI surfaces it. I would not have caught it on my own, at least not without six hours of reading.

One of our litigation attorneys recently used Eve to build a quick Excel table showing a client's medical treatment chronology before a hearing. He printed it, handed it to the judge, and the judge told him that having the information laid out that clearly was incredibly helpful. When one side is handing the court organized, digestible information and the other side is not, that changes how a judge looks at a motion. Small tools. Real leverage.

What I Tell the Skeptics

I hear two objections most often. The first is fear of sanctions. Fair enough. The most critical principle is human in the loop. Never send anything to a court, a client, or an insurance company that AI created without thorough review. Mistakes are not unique to artificial intelligence. That is why we already have people proofread each other's work.

The second is that AI will dull your skills. I think the opposite is true. If AI handles the six hours you used to spend reading 3,000 pages of medical records and gets it done in 20 minutes, you now have that time on the phone with clients, thinking through strategy, doing the work that actually sharpens you. The busywork was never what made you good. The thinking was.

The practice of law puts enormous stock in precedent. How things have always been done. That mindset is useful in a courtroom. It is dangerous when applied to how you run a practice.

Steve Duke is an attorney at Cofman Townsley in St. Louis, Missouri, and is a member of the Missouri Association of Trial Attorneys and the St. Louis Metropolitan Bar Association.