Sponsored by Broadcom
Epics like Tolstoy’s “War and Peace” have hundreds of thousands of words and even outliers like the 6-volume Decline and Fall of the Roman Empire top out at around 1.5 million. Large COBOL applications, on the other hand, can easily exceed 10 million words, so understanding them well enough to make changes can be a formidable challenge.
Changing complex applications can feel like maintaining an aircraft engine at 30,000 feet — not for the faint of heart — and the fear of causing a Jenga-like chain reaction is real. With larger applications, teams are often only confident enough to make more narrowly scoped changes than they would otherwise, which can be a drag on business agility.
Documentation, the traditional first port of call for understanding, is problematic as the information is often inaccurate or outdated if it exists at all.
As a result, teams have long relied on the institutional knowledge of senior developers — those who’ve spent years or even decades maintaining the applications and may have even written them — but the most senior ones are retiring en masse.
“Knowledge is knowing that a tomato is a fruit; wisdom is not putting it in a fruit salad.”
Similar to technical debt, retirements are creating cognitive debt within teams — a hole that leaves organizations exposed at a time when executives are demanding even faster time to market. As the pool of tribal knowledge shrinks, the risk of change increases, so replacing the wisdom accumulated through decades of real-life engineering experience has become an urgent need.
Scale of the Challenge
“How much time, on average, does it take you to understand where to make code changes?"
Third-party research has yielded a wide range of results, but when Broadcom posed the same question directly to engineering leaders during our application study tour, the response was stark (for example, one comment: “I wish it was only a matter of days”).
Application understanding and impact analysis is the single biggest friction point in mainframe development and it occurs before the code is ever touched.
Contemporary Technologies and the Path Forward
The key to successful and more expansive change management is beginning with a clear semantic understanding of the application’s business logic, rules, dependencies, and data flows. Know the context and implications before making the changes. AI and static analysis offer the solution, but there’s a practical prerequisite to address first.
Developers using ISPF know the pain of scrolling through code, page by page by page. It’s an obstacle right out of the gate, so the first step is to equip developers with tools like Code4z, which is built on Visual Studio Code. Searching entire folders and using regular expressions are important steps to unlocking the value of AI and static analysis and once developers are familiar with VS Code’s navigation, incorporating the new tools into workflows becomes natural.
GenAI enables developers to interrogate the code intuitively using natural language. It’s the colleague that’s always available, never gets tired of questions, and is capable of providing the broader business context (e.g., intent, why business rules exist). It also enables developers to document their code more easily, which pays additional “understanding” dividends over time.
Static analysis, on the other hand, delivers deterministic precision without hallucination, essential for mission-critical applications and governance in highly regulated industries. The parsing mechanisms use sophisticated mathematical models designed for context mapping (i.e., abstract syntax trees) and visibility can extend beyond source code to schedulers, JCL, databases, and environments.
Together, these two technologies make a powerful one-two punch.
Broadcom’s Approach
For the foreseeable future, we see developers using both technologies seamlessly from their Integrated Development Environment (IDE).
Our solution takes a holistic approach. It offers core static analysis capabilities as a seamless component of the developer workflow, removing the friction common with traditional solutions. It maps exact data flows, enforces compliance, and easily scales across millions of lines of code without hallucinating, and does so with optimized, easy-to-use visual navigation.
At the same time, it enables the use of Endevor-based source code and metadata with whichever AI tools the organization approves or endorses. Rather than proprietary options, we recommend a bring-your-own approach for both LLMs and code assistants (e.g., Claude Code, GitHub Copilot, Gemini Code Assist, Codex, Cursor, Devin, Kiro).
Broadcom’s solution is included in the Broadcom Mainframe DevOps Suite, so many customers already have the necessary licenses.
Which capability should you start with? Scanning source code for static analysis is often a more straightforward exercise than navigating sometimes fluid AI policies, toolsets, and cost controls, especially considering the cost of tokens when accessing million-line applications. Static analysis creates a reliable metadata base without costly AI calls. The goal ultimately is to give developers the freedom to find the right balance, the best of both approaches.
Contact us if you’d like to arrange a Code4z Workshop with access to these tools in a sandbox environment with sample applications, epic ones of course.
Kate Tumanova is the product lead for AI and application understanding in the mainframe DevX and DevOps space at Broadcom. Drawing on nearly 20 years of technical and leadership experience and deeply rooted in her early career as a Mainframe developer, Kate is focused on transforming the developer experience. She is passionate about empowering developer teams by abstracting the complexity of enterprise applications through modern, AI-driven development tools and practices.