Everyone is talking about AI code generators. Nobody is talking about what happens when the code needs to run on a mainframe.
Open any tech publication this month and you will find the same headline in a dozen variations: AI is transforming software development. GitHub Copilot, ChatGPT, Claude, Gemini – developers are using AI assistants to generate code faster than ever before. The consensus is that the developer's job is changing fundamentally.
They are right. And almost all of them are missing something important.
IBM z/OS mainframes run the core of the global financial system. Every major bank, most insurance companies, the majority of airlines, and a significant portion of government systems run on mainframe. This is not legacy infrastructure waiting to be replaced – it is the deliberate choice of organisations that cannot afford downtime, data loss, or security breaches.
This infrastructure is not going away. It is being extended, modernised, and increasingly integrated with cloud services. The people who maintain it still need to connect to it. And the way you connect to a mainframe has not changed in fifty years: you use a 3270 terminal emulator.
Here is what the AI narrative misses: the rise of AI code generation tools is actually increasing the need for people who can connect to and operate mainframe systems – not decreasing it.
Consider what happens when a developer uses an AI assistant to generate COBOL for the first time. The AI produces the code. Now what? The developer needs to connect to z/OS, submit the job, read the output, debug the abend, and iterate. They need a terminal. But they may never have used one before.
The AI wrote the code. But the S0C7 abend – a data exception, probably a numeric field receiving alphabetic data – still needs a human to diagnose it. That human needs to browse the dump, check the dataset, re-submit the job. All of that requires a working, capable terminal connection to z/OS.
The developer working with AI tools is overwhelmingly working on Linux or macOS. They use VSCode or Neovim. They live in a terminal. When they need to connect to a mainframe, they need an emulator that fits into their world – not a Windows-only legacy tool from 2001.
This is exactly the gap IM3270 was built to fill. A native 3270 terminal emulator for Linux that behaves like a modern application: tabbed sessions, split screen, macro recording, IND$FILE file transfer, and a retro CRT mode for the developers who appreciate the aesthetic of the machines they are learning to operate.
The conversation about AI in software development is important and exciting. But the infrastructure that runs the world's critical financial systems is not going to be replaced by a language model. It is going to be operated by a new generation of developers – developers who grew up with modern tools and expect the mainframe tooling to meet them where they are.
The terminal emulator is not a relic. It is the interface between the developer and the machine. Making it modern, capable, and native to Linux is not a small thing. For the organisations that run mainframe infrastructure, it is the difference between developers who can work effectively and developers who give up and go elsewhere.
AI is changing what developers build. It is not changing what the mainframe runs, or the fact that someone needs to connect to it.