8 Generative AI use cases for the mainframe

The modern field of artificial intelligence (AI) was formally founded in 1956 at a conference in New Hampshire, United States, where the term “artificial intelligence” was coined. That puts modern AI only a few years younger than the mainframe, first used in 1943.

Generative AI

Generative AI refers explicitly to a category of AI algorithms that generate new outputs based on data they trained on. Unlike traditional AI systems designed to recognize patterns and make predictions, generative AI creates new content. More technically, according to MIT, it learns to generate more objects that look like the data it was trained on.

You’ve probably heard about—or dabbled in—ChatGPT, Midjourney, Google Bard/Gemini, IBM® watsonx™, or Adobe Firefly, to name a few on a growing list. Generative AI can create images, code, text, data, translations, diagrams, videos, and music. 

AI and the Mainframe

Approximately 25% of the 2024 Arcati Mainframe Survey participants reported developing or deploying an AI/machine learning model on their mainframes. Additionally, 39% indicated that AI/machine learning is currently a subject of discussion within their organizations.

Why are nearly 64% of mainframe organizations undertaking AI? Among other use cases, companies seek solutions to three mainframe challenges:

• The Ever-Increasing Skills Gap—COBOL-trained employees are aging out, and COBOL and Assembler languages are less frequently taught. The question is, ‘How can the next generation of developers be onboarded to this platform to maintain and enhance mainframe applications?’ Thanks to public and private investment, including IBM’s newly minted Mainframe Skills Council, opportunities to gain mainframe skills continue growing. 
• Aging, Monolithic Applications—The nature of mainframe applications and programming languages means no modularity exists. Applications are monolithic. Yet organizations must keep pace to match digital economy expectations.
• The Rising Costs of IT—It’s no secret that increasing costs accompany rising mainframe transaction volume. Companies can reduce or shift workloads, but mainframe costs aren’t trending down.
  
  

8 Generative AI use cases for the mainframe 

Given the realities of mainframe skills, business agility, and cost, how can generative AI help? At least eight use cases are available now.

1. Generative AI for mentorship. Generative AI can help with skills and education to onboard a new generation to the VSAM platform. AI is arguably superior to quizzing Google or Stack Overflow because users don’t need to sift through results to find the best. AI already did that.

For example, a new developer could enter the prompt: What is VSAM? Generative AI and the programmer can start an informative ‘conversation’ similar to that of a teacher and a curious student.

Or, consider this prompt: What is the best way to optimize my COBOL code for performance? Programmers don’t need runtime tools based on static analysis. Generative AI provides guidelines that a programmer can validate by looking through the code. This helps job efficiency. It’s also a win for customers as generative AI helps them solve problems without exposing or risking any proprietary data. 

2. Generative AI for translating COBOL. A good use case for this is IBM® watsonx™. It uses generative AI to translate COBOL, an older transaction-focused programming language, into modern languages like Java. This translation maintains the correct sequence of operations, which is crucial for system performance. Watsonx aims to modernize mainframe systems, aligning older COBOL code with newer technologies, including AI, for enhanced integration and efficiency. 

3. Generative AI for business agility automation. Historically, there were few mainframe automations because there were no tools. Things have changed. During an online event, Venkat Balabhadrapatruni, a Distinguished Engineer at Broadcom Software, discussed leveraging CLI commands. 

Balabhadrapatruni’s team tested automated scripts to create datasets with dummy data using Zowe CLI commands. Zowe CLI is a command-line interface that lets you interact with the mainframe in a familiar format. It helps increase overall productivity, reduce the learning curve for developing mainframe applications, and exploit the ease of use of off-platform tools.

An example prompt: Generate a Typescript that uses Zowe CLI commands to create test datasets with dummy data to test a COBOL batch application that updates VSAM data.

The result? “Surprisingly accurate type script,” noted Balabhadrapatruni. “AI understood the flow of how commands would work and what a user would need to do to achieve that activity.” The script wasn’t perfect, but it “eliminated a huge chunk of grunt work and research.” For a business, reducing the time to deliver scripts improves agility.

4. Generative AI for generating synthetic data. We know data for testing is critical, but applications for insurance, credit cards, and banks have PII-sensitive information. No one wants unmasked data in an open-source project. But what if you could generate enough volume of synthetic data to leverage it in your testing? Generative AI can.

5. Generative AI for COBOL code considerations. Balabhadrapatruni and his team looked for existing open-source COBOL. Initial results with GPT-3 “were abysmal.” But now, with the evolution of AI, ChatGPT can differentiate between mainframe assembler and standard assembler as a starting point to generate code.

A generative AI response might include: “Here’s an example of a high-level assembler program for copying all members in a dataset to another dataset on a mainframe system…”

6. Generative AI for creating REXX scripts. REXX is considered the first general-purpose scripting language. Much of the automation in the mainframe is written in Rexx scripts. New mainframe employees need to understand the language. With generative AI, they gain a partner. 

7. Generative AI for code documentation. “Your mainframe code has been evolving since the 1960’s, so you better have some really good documentation!” recommends Balabhadrapatruni. With generative AI, users can pull a paragraph of mainframe COBOL, copy it into generative AI, and prompt it to “Explain what this [code] is doing.” Programmers want to understand it, and you want them to as well. Everyone needs well-documented code.

8. Generative AI extensions in VSCode for writing code. ChatGPT extensions allow programmers to gain AI assistance within the programming development environment, reducing the distractions of jumping in and out of VSCode. Features vary by extension but include inline code autocompletion, single-click built-in prompts, unit testing, debugging, documentation, code explanation, and code generation. 

The Future 

Balabhadrapatruni predicts that future generative AI will involve company-specific on-prem models, different from existing data warehouses and lakes. Future AI models will train with the company’s own data set to provide deeper insights and sharper business efficiency. Generative AI will eliminate design documents and building codes in various places and forms. 

“Mainframes are an untapped frontier that gets overlooked when it comes to AI,” notes Tom Taulli, author of Generative AI: How ChatGPT and Other AI Tools Will Revolutionize Business. “It’s a little dreamland for data scientists and an intriguing platform for AI.” 

A closed system would enable companies to use sensitive data without risks of exposure: credit card history, financial transactions, insurance models, health data, IMS. With proper data governance, the possibilities for business insights and efficiency are nearly limitless.

 Is there a downside to generative AI and mainframes? Find out.