Job description

AI agent chains** to extract, analyse, and understand this estate at depth — then drive forward engineering onto a modern Spring Boot / Java 21 / Angular / MongoDB cloud-native stack, with AI agents accelerating design, code generation, test authoring, and migration validation at every step. • Must Have Technical/Functional Skills • ## The Modernisation Mission • Legacy estates in scope typically include: • Mainframe COBOL/CICS/IMS batch and online transaction processing • Hierarchical and relational databases (IMS, DB2) with deeply embedded business logic • Proprietary messaging middleware (IBM MQ) and brittle point-to-point integrations • Legacy OO platforms (VisualAge Smalltalk, Tonel format) with no test coverage or documentation • JCL/Assembler job streams woven into business-critical workflows • Your mission: deploy AI agent chains to extract, analyse, and understand this estate at depth — then drive forward engineering onto a modern Spring Boot / Java 21 / Angular / MongoDB cloud-native stack, with AI agents accelerating design, code generation, test authoring, and migration validation at every step. • ## Key Responsibilities • ### AI-Augmented Reverse Engineering • Design and deploy custom AI agent pipelines that ingest legacy artefacts — COBOL programs, IMS DBDs/PSBs, DB2 schemas, JCL, Smalltalk Tonel sources — and produce structured outputs: business rule inventories, data-flow maps, domain entity models, and dependency graphs • Build multi-agent chains that cross-reference extracted business logic against live transaction traces, test outputs, and production data patterns to validate completeness and surface hidden edge cases • Use agents to auto-generate legacy comprehension artefacts: annotated COBOL walkthroughs, IMS segment relationship diagrams, CICS program call trees, and DB2-to-document data-model mappings • Orchestrate agent workflows that identify dead code, duplicated logic, and tightly coupled components — producing prioritised decomposition candidates for the modernisation backlog • Validate agent-extracted business rules against domain SMEs; build feedback loops that improve agent accuracy over successive extraction cycles • ### AI-Augmented Forward Engineering • Design forward engineering agent chains that consume reverse-engineered domain models and produce: Spring Boot service skeletons, OpenAPI 3.1 contracts, MongoDB schema designs, Angular component scaffolds, and JUnit 5 test suites — all aligned to team coding standards • Build agents that enforce architectural patterns during code generation: no business logic in adapters, domain models free of persistence concerns, API contracts decoupled from internal representations • Deploy agents for migration validation — automatically comparing migrated service behaviour against legacy outputs across a curated test corpus, flagging behavioural divergence before human review • Use AI to accelerate CI/CD pipeline authoring, infrastructure-as-code generation (Terraform, Helm), and runbook drafting — with engineers reviewing and owning the outputs, not rubber-stamping them • Chain agents to continuously scan modernised code for legacy anti-patterns bleeding into new services, enforce non-functional requirements (observability hooks, circuit breakers, health endpoints), and flag design drift from approved blueprints • ### Custom Agent Design & Engineering • Architect multi-agent systems using one or more agentic AI platforms and frameworks: • Claude Code CLI (Anthropic) — agentic coding, slash commands, MCP tool integration, custom agent loops • Cursor — AI-native IDE agent workflows, codebase-wide context, rule-based agent behaviour • Gemini CLI (Google) — Gemini-powered agent pipelines with tool use and long-context reasoning • LangChain / LangGraph — chain and graph-based agent orchestration, tool registries, state machines • AutoGen / CrewAI — multi-agent conversation frameworks, role-based agent specialisation • Anthropic Agent SDK / OpenAI Assistants API — programmatic agent construction with tool use, memory, and structured output • Select the right orchestration pattern for each workstream: sequential chains, parallel fan-out, supervisor/worker, reflection loops, human-in-the-loop checkpoints • Build domain-specific agent tools: legacy code readers, schema extractors, API contract validators, test harness runners, cloud cost estimators, IaC generators • Design human-in-the-loop checkpoints : define what agents decide autonomously, what they flag for engineer review, and what requires architect sign-off • Evaluate, benchmark, and improve agent chain quality: extraction completeness, forward-engineering accuracy, false-positive rates, and time-to-output • ### Solution Design & Technical Authority • Own end-to-end solution design for modernisation workstreams — producing LLD documents, sequence diagrams, PlantUML/Mermaid data-model mappings, strangler-fig migration maps, and API surface designs • Evaluate architectural trade-offs: lift-and-shift vs. re-platform vs. re-architect, agent-generated vs. hand-crafted, monolith decomposition sequencing — all documented as ADRs with explicit rationale • Define integration patterns for hybrid-state environments: mainframe co-existence, MQ-to-event-streaming migration, dual-write data consistency, feature-flag-controlled cutovers • Lead design reviews; drive alignment between AI workstream leads, legacy SMEs, domain engineers, and cloud platform teams • ### Technical Leadership & Team Development • Lead a cross-functional team spanning backend, frontend, data migration, and AI/agent engineering • Conduct structured code reviews across both hand-authored and agent-generated code — human review of AI output is non-negotiable; agents accelerate, engineers own • Establish standards for agent-assisted development: what must be reviewed, what must be tested, how agent outputs are versioned and audited • Mentor engineers on agentic AI patterns, prompt engineering for code tasks, and responsible use of AI-generated artefacts in production systems • Coach engineers unfamiliar with legacy systems to read COBOL/IMS structures via agent-assisted comprehension tools you have built • ### Delivery Execution • Break modernization epics into sprint-deliverable stories with measurable progress indicators: % business logic migrated, legacy endpoints retired, agent pipeline accuracy metrics • Track and communicate migration coverage — human-readable progress dashboards built partly by agents, owned by you • Identify and mitigate transition risks: agent hallucination in business rule extraction, data consistency during dual-write phases, performance parity of migrated services • Own sprint-level commitments; surface blockers with proposed mitigations, not status updates

Requirements

Must-haves

Must Have Technical/Functional Skills
## The Modernisation Mission
Mainframe COBOL/CICS/IMS batch and online transaction processing
Hierarchical and relational databases (IMS, DB2) with deeply embedded business logic
Proprietary messaging middleware (IBM MQ) and brittle point-to-point integrations
Legacy OO platforms (VisualAge Smalltalk, Tonel format) with no test coverage or documentation
JCL/Assembler job streams woven into business-critical workflows
## Key Responsibilities
### AI-Augmented Reverse Engineering
Build multi-agent chains that cross-reference extracted business logic against live transaction traces, test outputs, and production data patterns to validate completeness and surface hidden edge cases
Use agents to auto-generate legacy comprehension artefacts: annotated COBOL walkthroughs, IMS segment relationship diagrams, CICS program call trees, and DB2-to-document data-model mappings
Orchestrate agent workflows that identify dead code, duplicated logic, and tightly coupled components — producing prioritised decomposition candidates for the modernisation backlog
Validate agent-extracted business rules against domain SMEs; build feedback loops that improve agent accuracy over successive extraction cycles
### AI-Augmented Forward Engineering
Build agents that enforce architectural patterns during code generation: no business logic in adapters, domain models free of persistence concerns, API contracts decoupled from internal representations
Deploy agents for migration validation — automatically comparing migrated service behaviour against legacy outputs across a curated test corpus, flagging behavioural divergence before human review
Use AI to accelerate CI/CD pipeline authoring, infrastructure-as-code generation (Terraform, Helm), and runbook drafting — with engineers reviewing and owning the outputs, not rubber-stamping them
### Custom Agent Design & Engineering
Architect multi-agent systems using one or more agentic AI platforms and frameworks:
Claude Code CLI (Anthropic) — agentic coding, slash commands, MCP tool integration, custom agent loops
Cursor — AI-native IDE agent workflows, codebase-wide context, rule-based agent behaviour
Gemini CLI (Google) — Gemini-powered agent pipelines with tool use and long-context reasoning
LangChain / LangGraph — chain and graph-based agent orchestration, tool registries, state machines
AutoGen / CrewAI — multi-agent conversation frameworks, role-based agent specialisation

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aijavaangularapici/cd

Platform: jobdiva
Effort: medium