Most people don't associate law firms with custom software development. Joel Karr does.
As CTO at Keller Postman, a mass torts and personal injury firm that's grown to over 1,000 people, Joel oversees infrastructure that looks more like consumer tech than traditional legal practice.
"When you work with 100,000 clients in one case, it feels more like e-commerce than legal," he explains. "We need to contact 100,000 people to get a form filled out or a document signed. That involves emails, texts, phone calls, predictive dialers."
In his conversation with Lucas Hendrich on CTO2CTO, Joel shares what it takes to build custom software in an industry that traditionally buys tools, how AI fits into his team's workflow (and where it doesn't), and why his background as a collegiate wrestler shapes how he approaches engineering leadership.
When Joel joined Keller Postman a couple of years ago, the firm was drowning in third-party tools.
"There were a lot of different tools, and they didn't know how to integrate them. Every six months they bought a new tool because they didn't know how to use the first one."
His first six months focused on consolidation: identify the platform, figure out what stays, what goes.
That led to custom software development. Today, about 60% of what the team manages is built in-house, particularly anything viewed as a competitive advantage.
"The core framework like user permissions and the CRM are still custom. But we have tools for journeys: ;emails, calls, predictive dialers. Custom forms. And everything has an AI component now."
The team even builds commercial products licensed to other law firms, a shift from traditional CIO-style vendor management to product development.
Building at scale in legal requires infrastructure designed for massive transactions.
"We build on Azure primarily. Cosmos for things like forms where we view those as snapshots. Azure SQL and Postgres for anything we're doing at massive scale."
The CRM is where data lands after all those transactions complete. But here's where acquisitions complicate things: some acquired firms use Salesforce-based CRMs, others use FileVine or Clio.
"Attorneys have worked in these systems for five to ten years. They're not quite ready to give them up. So we build tools that can store data in all these different systems."
It's a pragmatic approach: rather than forcing migration immediately, build bridges.
Joel's team uses GitHub Copilot extensively. The stats look impressive: 80% of code lines written by AI.
But he pushes back on the hype.
"When I asked the team, they said there's no way. The reality is, if you view it as a rough draft, 80% of lines might be written by AI, but not 80% of effort."
The real work—understanding requirements, making architectural decisions, finding synergies—still requires human judgment.
"The team uses it as a first draft. We started talking about what you do before you write the code."
About six months ago, the team had an all-hands focused on breaking down user stories into tasks.
"Fifteen years ago, I started talking about task templates. If the task is 'build a data repository,' everyone knows what that looks like. You don't get random new ideas every time."
Now they feed their library of task templates into AI models.
"AI already knows our strengths, our design patterns. Sometimes it will introduce a task we've never done before, but it's using our foundation."
This transforms something developers hate (writing out tasks in Jira or Azure DevOps) into something more enjoyable: iterating with an AI model, then copying and pasting the output.
The key: AI learns their patterns rather than introducing chaos.
One of Joel's best examples of architectural thinking: recognizing platform synergies across seemingly different problems.
Keller Postman originally built custom forms for client intake. Before that, they used Formstack and seven other tools. The new system now handles 85% of new clients without a phone call (versus 15% before).
Then they tackled medical records.
Mass torts cases involve thousands-page medical records. The goal: find specific pages with citations that legally prove injuries.
"We get questions from attorneys: what date was the diagnosis? What injury did they have? We realized it was very similar to what we built for forms."
The team built over 100 medical review questionnaires on the same platform.
"The users never asked for that. They asked for specific data extraction. But we found the synergy."
That's the value of understanding the business deeply enough to see patterns others miss.
Keller Postman has acquired multiple firms. Each brings its own CRM, its own tools, varying levels of modernization.
"People want tech to come in during due diligence and say there aren't any problems. They don't want tech to be the reason the deal stops."
Joel hasn't personally seen a scenario where tech killed a deal. "You're going to need to figure out how to integrate regardless."
But the real challenge isn't technical: it's human.
"It's very emotional to be acquired. To have new leadership tell you these are the new tools you'll use. A lot of what we do is talk about a plan, then realize the pace will change drastically based on how people react to each change."
He focuses on building credibility first: start with admin-based tools that impact small groups of power users, not the most critical business workflows.
"We spent a year building things ancillary to the core product before tackling the main critical workflow."
Joel's background as a collegiate wrestler at University of Illinois, where his team won Big Ten championships, shows up constantly in how he leads.
"When you're around people operating at that level (teammates who were world team members, a coach who was an Olympian) things literally change your life in the way you view things."
The biggest lesson: operating under pressure.
"You find ways to limit unknowns. Whether you're in a boardroom and someone asks 'how long will that take?' or you're mocking an endpoint for a test case; being able to limit unknowns so you can make snap decisions is critical."
The second lesson: deliberate practice.
"I took the same training mentality from athletics into software engineering. Breaking things down into strengths and weaknesses. When I first got a job and they said 'we're using dependency injection,' I said 'I have no idea what you're talking about.'"
He built a practice plan. Spent a couple days learning, then thirty minutes daily for weeks testing whether he still understood it.
"Everyone in software engineering is an optimist. You would have quit if you weren't. You can't be comfortable being uncomfortable in software engineering and then try to live differently in other parts of your life. When you need to push through solving a problem, you're not doing it just then—you're doing it all the time."
Joel learned to code at age ten from a QBasic textbook. His nine-year-old daughter wanted to learn, so he showed her the same book.
"She opens it and says, 'None of these screens are here.'"
So she asked ChatGPT: "Teach me how to code."
"Having that at your fingertips—where you don't have to follow step-by-step instructions, you can interact and say what you don't understand and focus on that—is a game changer."
Joel sees parallels to how code itself has evolved. IntelliSense was revolutionary. Azure Search turned three-month projects into two days. AI code generation is another step in that progression. "The expectations are going to rise so much. We're going to be doing a lot more. We haven't run out of hard challenges yet."
Despite all the advancements, some skills remain differentiators.
"How do you take a complex system and simplify it in a way that you can talk with a team about it, talk with the business side about it? Whether AI is writing the code or someone spent three weeks typing it twenty years ago—that hasn't changed."
Naming APIs, for example. "We joke about it being one of the hardest things we do. But really we're talking about how you logically group things together. Who's responsible for supporting it? Who decides if something belongs in that API or a different one?"
These architectural decisions still require human judgment.
And when someone says "add this specific button to this page," the skill is understanding what they're really trying to solve—because it might overlap with something already built.
"The role might shift a little bit, but the expectations of being able to ask better questions to understand what people are doing is going to be critical."
Joel Karr's career offers a unique lens on building technology in industries that aren't traditionally "tech companies."
From consolidating tools to building custom platforms to creating commercial products—his work at Keller Postman demonstrates that legal tech is evolving rapidly, and the firms that win will be those that build competitive advantages rather than just buy tools.
His wrestling background reinforces a truth that applies across domains: deliberate practice, operating under pressure, and maintaining an optimistic mindset aren't just athletic principles. They're how you build teams that solve hard problems.
Listen to the full episode of CTO2CTO to hear more about AI as a rough draft, finding synergies across systems, and why everyone in software engineering is secretly an optimist.
Let's talk about where you are today and where you want to go - our experts are ready to help you move forward.
