You’ve seen the demos. You’ve probably read the whitepapers. The promise of using AI to modernize, maintain, and generate mainframe code is intoxicating.
So, you sign the check for IBM Watsonx Code Assistant for Z, OpenAI Codex, Claude Code, Google Antigravity, or similar. Or in the near future, you may plan to bring in IBM Bob to revolutionize your developer experience.
You wait for the massive productivity gains to roll in.
And then… Nothing happens.
If this sounds familiar, it’s time for a blunt reality check: You are wasting your AI budget. If you are still running on ChangeMan, Endevor, ISPW, or any other legacy SCM tool, you need to modernize your SCM or stop pretending you’re doing AI.
This is the uncomfortable truth that your technical debt is hiding from you:
AI does not work in a vacuum, and it certainly does not work properly in an environment designed in the 1980s.The Fundamental Disconnect: AI vs. The Legacy Vault
Modern AI coding assistants like the IBM WatsonX Code Assistant for Z and IBM Bob are brilliant, but they are not magic. They are context engines. To suggest accurate code, refactor complex legacy monoliths, or understand business logic, these tools need immediate access to your entire codebase. They thrive on deep repository context, modern IDEs, such as VS Code, and fast, iterative feedback loops.
Legacy SCMs are the antithesis of this. They were built for a world of isolation, rigid control, and step-by-step human intervention. When you try to bolt modern AI onto tools like ChangeMan, you hit a brick wall of bottlenecks:
- Context is Locked in a Dungeon: Mainframe code is often locked away in proprietary PDS libraries. AI cannot effectively scan, index, or comprehend relationships across your codebase when the files are fragmented across siloed, proprietary structures. If the AI can’t see the context, it fails (or hallucinates).(More on accessing these PDSs using tools like Zowe below)
- Agents Don’t Do “Check-Outs”: Modern AI relies on agentic workflows. Meaning the AI can plan, write, test, and iterate on code autonomously. Legacy SCMs rely on rigid, manual check-in/check-out processes with complex locking mechanisms. An AI agent cannot navigate a green-screen approval gate to test a simple COBOL refactor(Maybe they can if we build a skill for this, but do we really want this?)
- The IDE Gap: AI assistants are designed to live inside modern editors. Forcing developers to jump between a modern VS Code interface for AI suggestions and a 3270 to actually commit the code creates unbearable friction.
The “Zowe” Band-Aid Is Not a Strategy
At this point, someone on your engineering team might say, “Wait, we can just use Zowe! Copying code from a PDS to a local repository is easy.”
They aren’t wrong. Zowe is a fantastic tool, and bridging the gap by pulling PDS members into a local Git repo for the AI to analyze is a totally viable tactical workaround. But let’s be clear: manual syncing is a band-aid, not an enterprise strategy.
Relying on developers to manually pull from ChangeMan, ask the AI for help in VS Code, and then manually push the changes back through a legacy pipeline is the definition of shadow IT. It introduces version control nightmares, breaks the single source of truth, and completely negates the velocity you just paid millions to achieve.
Adapt or Watch Your Investment Burn
If you want the ROI from AI in your Software Delivery Lifecycle, you have to feed them what these tools need to survive: Git.
The era of proprietary mainframe source control is over. To unlock AI, your mainframe development lifecycle must look exactly like your cloud-native development lifecycle. That means migrating your mainframe source code to Git and implementing modern, automated CI/CD pipelines. Period.
Without Git, your AI tools are blind. Without CI/CD, your AI workflows are paralyzed.
“Moving away from legacy library managers is not a new trend — the first migrations began as early as 2012, and the capabilities to support these transitions have significantly improved over the past 14 years. Organizations that have already made the move are now reaping the benefits of modern AI-driven development tools.”
Rosalind Redcliff, IBM
Speaking at Mighty Mainframe Conference 2026
The Path Forward
Transforming a decades-old mainframe culture and migrating millions of lines of mission-critical code from ChangeMan to Git is intimidating. But the alternative is paying for Ferraris and driving them in traffic jams.
You don’t have to navigate this transition blindly. At CROZ, we specialize in the hard truths and the heavy lifting of enterprise modernization. We know exactly how to untangle legacy SCMs, securely migrate your critical assets to Git, and build the modern CI/CD pipelines required to actually unleash your AI investments.
Stop pretending. Stop wasting your budget. Let’s build an environment where your AI can actually do its job.
Reach out to us today, and let’s get your mainframe really into the age of AI.
Falls Sie Fragen haben, sind wir nur einen Klick entfernt.
