Back office ops · Production

Keeping Java as the Core, Python to Lead Agentic Systems

The problem

Attempting to build agentic AI systems in Java generated excessive boilerplate, glue code, and DIY re-implementation of Python examples, making it hard to iterate quickly on prompts, tools, and reasoning flows in an ecosystem that defaults to Python.

First attempt

Early Java-first attempts at MCP servers using Spring Framework and local AI agents using Java with Google ADK produced working code but felt like swimming upstream — too much ceremony for a fast-changing experimentation context.

Workflow diagram · grounded in source

Customer request triggers agent

trigger

“Given a customer's request, decide whether to approve an order, check credit, apply discounts, and notify.”

LLM reasons and selects tools

ai_action

“Take user intent → figure out what tools to call → call them in the right order → reason about the responses → decide the next step.”

Java services called via MCP

integration

“Calls those Java services via MCP tools or HTTP APIs”

High-risk credit escalation

human_review

“Check credit → if high-risk, ask for manual review”

Pricing, approval, and notification

output

“fetch pricing → apply rules → approve/deny → send notification”

Reported outcome

The author adopted a layered architecture: Java services as the stable core for business logic and transactions, thin Python MCP server adapters, and Python agents with LLMs for orchestration and fast-changing experimentation.

Reported stack

Spring Frameworkgoogle ADKMCP

Source

https://medium.com/walmartglobaltech/keeping-java-as-the-core-python-to-lead-agentic-systems-e2960693e1cd

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The author adopted a layered architecture: Java services as the stable core for business logic and transactions, thin Python MCP server adapters, and Python agents with LLMs for orchestration and fast-changing experim…

What tools did this team use?

Spring Framework, google ADK, MCP.

What failed first in this deployment?

How is this back office ops AI workflow structured?

Customer request triggers agent → LLM reasons and selects tools → Java services called via MCP → High-risk credit escalation → Pricing, approval, and notification.