Replay: Modernizing Legacy: COBOL to Cloud with GitHub Copilot

Key Concepts:

• AI-Powered Modernization: Using AI tools like GitHub Copilot and autonomous agents to modernize legacy code, specifically COBOL.
• COBOL: A 65-year-old programming language still used in many financial and government systems.
• GitHub Copilot: An AI-powered code completion and generation tool.
• Autonomous Agents: AI agents that can independently perform tasks, such as code transformation.
• Semantic Kernel SDK: A Microsoft SDK for orchestrating autonomous AI agents.
• Code Preparation: Steps to make legacy code more understandable for AI, including reverse engineering, removing irrelevant information, and translating comments.
• Code Enrichment: Adding comments and structure to code to improve understanding.
• Automation Aids: Tools and techniques to assist in the modernization process, such as call chain analysis and test generation.
• Call Chain Structure: The hierarchy of file calls within a COBOL program.
• TDD (Test-Driven Development): An approach where tests are written before the code.
• Utility Classes: Reusable code components for common tasks.
• Multi-Agent Orchestration: Coordinating multiple AI agents to work together on a task.
• Token Limits: The maximum amount of text that an AI model can process at once.
• LLMs (Large Language Models): AI models trained on vast amounts of text data.

1. Introduction and Context

• Andrea Griffiths from GitHub introduces Julia Cordic, a software global black belt at Microsoft, to discuss AI-powered modernization of COBOL systems.
• COBOL is a 65-year-old language still processing a significant amount of financial transactions and other critical processes.
• A major challenge is the scarcity of COBOL developers, as many are retiring or nearing retirement.
• Julia will demonstrate how GitHub Copilot and autonomous agents can help with COBOL code transformation.

2. Julia Cordic's Background and Role

• Julia is a software global black belt at Microsoft, specializing in software engineering, DevOps, and AI-assisted coding.
• She helps large organizations accelerate developer workflows through platform engineering.
• Julia has experience in software engineering, dev lead, and cloud solutions architecture.
• She did not learn COBOL but has worked on projects involving its modernization.

3. The Problem with Legacy Code

• Organizations often lack a clear understanding of what their legacy code is actually doing.
• Reverse engineering is necessary to understand the code's functionality.
• Legacy code may contain irrelevant information, such as outdated comments and logs.
• Code may be written and commented in languages other than English, making it harder for AI models to process.

4. Three Steps to Legacy Code Modernization

• Code Preparation:
  • Reverse engineering to understand the code's functionality.
  • Removing irrelevant information, such as outdated comments and logs.
  • Translating comments to English for better AI processing.
• Code Enrichment:
  • Adding comments to code to improve understanding.
  • Identifying and leveraging deterministic structures in the code.
• Documentation:
  • Documenting everything generated by Copilot in markdown files to serve as a source of truth.

5. Demonstration of GitHub Copilot Agent Mode

• Julia demonstrates using GitHub Copilot agent mode to analyze COBOL code.
• Copilot generates a markdown file with a file inventory, complexity analysis, and business purpose.
• It also creates a mermaid diagram to visualize the code structure.
• This analysis can be used as a starting point for generating Java code.

6. Automation Aids

• Call Chain Analysis:
  • Generating a call chain hierarchy to understand which files call which other files.
  • Creating a mermaid file to visualize the call chain.
• Test Generation:
  • Using a TDD approach to generate Java tests based on the extracted business logic.
• Dependency Analysis:
  • Identifying utility classes and suggesting potential dependencies in Java or .NET.

7. Microsoft Semantic Kernel and Autonomous Agents

• Julia and her colleague Gustaf created a repository using Microsoft Semantic Kernel for autonomous agent orchestration.
• The framework allows prompting agents from any source to any target.
• It requires an Azure Open AI endpoint or an AI endpoint of choice.
• The framework was tested with GPT-4, GPT-4-turbo, and GPT-3.5-turbo models.

8. Doctor.sh Script

• The doctor.sh script checks the configuration and dependencies of the project.
• It verifies that the necessary variables are set and that the dependencies are installed.
• The doctor.sh run command executes the code transformation process.

9. Agent Prompting and System Prompts

• The agents are initialized with system prompts that define their roles and responsibilities.
• The system prompts provide step-by-step instructions for the agents to follow.
• The prompts can be customized to suit specific needs.

10. Sanitizing COBOL Code

• The framework includes a function to sanitize COBOL code by removing irrelevant information.
• This can also be done manually with GitHub Copilot.

11. Security and Data Privacy

• When using a cloud-hosted LLM, data is sent to the model in the cloud.
• For sensitive data, organizations can host their own LLMs on their own infrastructure.
• Semantic Kernel is just an SDK and does not handle any data itself.

12. Cloud Platforms

• Many financial institutions are working with Azure.
• AWS also has an offering for mainframe modernization, but it is proprietary.
• The choice of cloud platform depends on the organization's needs and preferences.

13. Modernization vs Code Extensions

• There are modernization vs code extensions for Java and .NET.
• These extensions are based on modernization rules that are already known.
• The framework presented in the video is more flexible and can handle more complex scenarios.

14. Conclusion

• AI-powered modernization of COBOL systems is a complex problem that requires a multi-faceted approach.
• GitHub Copilot and autonomous agents can help automate the process, but human expertise is still essential.
• The Microsoft Semantic Kernel SDK provides a powerful framework for orchestrating autonomous agents.
• Organizations can contribute to the open-source project and customize it to suit their specific needs.
• The key is to break down the problem into smaller chunks and tackle it iteratively.

15. Call to Action

• Organizations interested in collaborating on the project can reach out to Julia and Gustaf on LinkedIn or open an issue in the GitHub repository.
• The open-source repo can be found under aka.ms/ccobalt.
• A blog article about the project can be found under aka.ms/cobaltmosblog.
• Julia will be speaking at Porto Techub in October.