Workshops: Agentic Coding by Tech Stack

COBOL & Mainframe Modernisation with AI Agents: Workshop

Many banks, insurers, public authorities and industrial companies still run their core business logic in COBOL, while the people who wrote it are retiring. AI agents can read large COBOL programs, explain them in business terms and support migration, but only with careful preparation and consistent verification. In this two-day workshop your team learns to extract, document and safely modernise mainframe business logic with agents.

Who is this workshop for?

  • COBOL and PL/I developers maintaining mainframe applications
  • Java or C# developers taking over or replacing mainframe functionality
  • Modernisation architects and programme leads
  • Business analysts who need to understand grown business rules
  • Operations and test teams for batch and online processing

What you will take away

  • Prepare COBOL sources, copybooks and JCL so agents can analyse programs that exceed their context window.
  • Have agents turn business rules, data flows and call relationships into documentation that domain experts can verify.
  • Build characterisation tests from recorded inputs and outputs before changing any logic.
  • Evaluate migration paths and translate selected modules with agent support, verified by parallel runs.
  • Recognise typical translation pitfalls, from packed decimals and rounding to EBCDIC sort orders.

Agenda

  1. Module 1

    Day 1: Making the mainframe readable

    • Getting sources into Git: Zowe, vendor tooling or export routines
    • Claude Code, OpenAI Codex and specialised offerings from IBM and cloud providers
    • Strategies for very large programs: slicing by section, expanding copybooks
  2. Module 2

    Day 1: Understanding and documenting

    • Exercise: extracting business rules from a real batch program
    • Data dictionaries from copybooks, job flows from JCL
    • Validating documentation with domain experts
  3. Module 3

    Day 1: Securing behaviour

    • Characterisation tests based on test files and recorded transactions
    • Fast feedback with GnuCOBOL for programs without CICS or DB2 dependencies
    • Test data without production records
  4. Module 4

    Day 2: Modernisation paths

    • Refactor in place, rehost, re-platform or rewrite: criteria and effort
    • Strangler approach with APIs in front of CICS transactions
    • Choosing the first module
  5. Module 5

    Day 2: Agent-assisted migration

    • Exercise: translating a module to Java or C# with the agent
    • Decimal arithmetic, rounding, EBCDIC and date handling
    • Parallel runs and automated output comparison
  6. Module 6

    Day 2: Programme and governance

    • Review and sign-off with domain experts
    • Documentation as a lasting asset, even without migration
    • A roadmap for your application portfolio

Why COBOL differs from other legacy code

COBOL programs often run to thousands of lines, share data structures through copybooks and only make sense together with JCL, CICS definitions and database schemas. Business logic is spread across batch chains that run in a fixed order. An agent that sees one program in isolation will explain it plausibly and still miss the point.

We show how to assemble context deliberately: expanded copybooks, job flows, call graphs and glossaries of business terms. With that, agents become remarkably good at turning old code into verifiable documentation.

No big bang

Fully automated translation of a whole portfolio sounds attractive but rarely produces maintainable code. We favour incremental modernisation: understand and document first, protect behaviour with tests, then migrate module by module with parallel runs. Even if the mainframe stays, documented business logic is a lasting gain.

Confidentiality and regulated environments

Mainframe applications often process financial, insurance or citizen data and are subject to strict rules on data protection and outsourcing. We discuss options from enterprise AI plans to self-hosted models and work with anonymised extracts if needed. Led by neooutput AI Code Engineers and AI Solutions Architects, remote or on-site, in English or German.

Frequently asked questions

Can agents really understand COBOL?

COBOL is verbose and well represented in public code, so current models read it surprisingly well. The bigger challenge is context from copybooks, JCL and system dependencies. With good preparation, explanations become reliable enough for domain experts to review efficiently.

Do we need mainframe access during the workshop?

No. We work with exported sources and test data. If you have a mainframe or emulated development environment, we can include it for compilation and tests.

Will translated code look like Java written by people?

Only if you specify it and review accordingly. Line-by-line translation produces COBOL in Java syntax. We show how to set target architecture and coding standards so the result stays maintainable.

Workshop

Get in touch

Want to secure the knowledge in your COBOL code before it retires?

Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.

Prefer to talk first?

15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.

Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)

Request a training / workshop

Workshop: COBOL & Mainframe Modernisation with AI Agents: Workshop · 2 days

Dates that work for you (optional)

Suggestions about 2 weeks ahead – pick as many as you like

Reply usually within 24 hours · jocapps is ISO 27001 certified · Privacy