Case Study | Untangling a Law Firm's Codebase with Claude Code

A law firm serving local government collections had built its operations on a single .NET and SQL system that grew for more than a decade. Hundreds of clients submit data overnight in formats that differ by state law, court system, and software vendor, and every one of those variations had been handled with client-specific logic added to the same codebase. Even though the system worked, it had grown past its documentation and several of the developers who wrote its most intricate sections had moved on. Worthwhile worked alongside the firm's developers to use AI assisted coding to untangle the layered system.

‍

The Challenge

A prominent law firm in collection services was founded in the 1970’s on the premise of assisting historically underserved local governments with revenue-driving collection services. More than 50 years later, the firm occupies 15 primary offices and multiple satellite locations with more than 400 employees and 60 attorneys. Their work is characterized by high-volume legal and administrative processing across many jurisdictions simultaneously, with the strict compliance and accuracy requirements that come with serving public-sector clients. The expansion beyond the firm's original Texas footprint was a genuine feat but it also introduced significant back-end complexity.

Hundreds of clients submit data overnight in wildly inconsistent formats, with no standardization possible across differing state laws, court systems, and software vendors. To handle that variation, the system had grown over more than a decade into a large, monolithic codebase in which client-specific logic accumulated steadily over time.

“It’s like a house that gets added a new addition each time a unique issue comes up. Making the whole thing a bit topsy-turvy.” – Chief Information Officer

‍

The Solution

The law firm's developers were highly capable at writing and troubleshooting code, but making sense of a system this layered called for a scalpel rather than a jackhammer, and doing it by hand would have meant countless hours of manual investigation before any new work could begin. With AI-assisted development, the team wanted something more nuanced than "vibe coding" or AI simply writing the code.

“I really need it to identify the issues in the code itself, not write the program. Carefully untangle the web and point us in the direction of the problems.” – Senior Software Developer

‍

Worthwhile brought both the strategy and the hands-on engineering to make that possible. Rather than treating the problem as something to be solved in a single pass, they laid out a deliberate sequence: establish automated tests as guardrails first, use Claude Code to understand what the existing code was doing, and only then begin refactoring toward something more maintainable. Worthwhile's senior engineers worked directly in the codebase alongside the firm's developers, while also functioning as a strategic partner.

‍

Worthwhile guided the law firm to Claude Code due to its reputation and experienced capabilities. Anthropic's models were regarded as best-in-class for coding, which mattered for a job that was less about generating new features and more about comprehending dense, existing logic across a .NET and SQL codebase. Just as valuable was the flexibility to match the model to the task: higher-reasoning models could be used for the thinking-intensive work of planning a change and reviewing it before it was committed, while lighter, faster models handled the implementation in between, an approach that kept both quality and cost under control.

‍

The Outcome

The clearest early signal came in developer productivity. Across the team, individual developer throughput improved by 20–100% depending on the mix of work. Rather than replacing the team's expertise, Claude Code amplified it, allowing skilled developers to accomplish considerably more in the same amount of time.

“As we’ve continued to work in Claude Code it has opened the door to other development projects that were sitting on the backburner due to their complexity. AI has made big refractors feasible.” – Senior Software Developer

‍