In the first part of this series, we explored the “divide and conquer” strategy using sub-agents to create a flexible, modular team of AI specialists. This is perfect for situations where the user is in the driver’s seat, directing the flow of conversation. But what about when the process itself needs to be in charge?

Some tasks are inherently linear. You have to do Step A before Step B, and Step B before Step C. Think about a CI/CD pipeline: you build, then you test, then you deploy. You can’t do it out of order… or if you do, be prepared for havoc!

Let me come back to the Agent Development Kit (ADK) for Java! We recently discussed the many ways to expand ADK agents with tools. But today, I want to explore the multi-agentic capabilities of ADK, by talking about sub-agent workflows.

In upcoming articles in this series, we’ll also talk about sequential, parallel, and loop flows.

The “divide and conquer” strategy

Think of building a complex application. You wouldn’t put all your logic in a single, monolithic class, would you? You’d break it down into smaller, specialized components. The sub-agent workflow applies this same “divide and conquer” principle to AI agents.

Like many developers, I’ve been exploring the creative potential of Large Language Models (LLMs). At the beginning of the year, I crafted a project to build an AI agent that could generate short science-fiction stories. I used LangChain4j to create a deterministic workflow to drive Gemini for the story generation, and Imagen for the illustrations. The initial results were fascinating. The model could weave narratives, describe futuristic worlds, and create characters with seemingly little effort. But as I generated more stories, a strange and familiar pattern began to emerge…

A very common question I get when presenting and talking about advanced RAG (Retrieval Augmented Generation) techniques, is how to best index and search rich documents like PDF (or web pages), that contain both text and rich elements, like pictures or diagrams.

Another very frequent question that people ask me is about RAG versus long context windows. Indeed, models with long context windows usually have a more global understanding of a document, and each excerpt in its overall context. But of course, you can’t feed all the documents of your users or customers in one single augmented prompt. Also, RAG has other advantages like offering a much lower latency, and is generally cheaper.

In the first article of this Advanced RAG series, I talked about an approach I called sentence window retrieval, where we calculate vector embeddings per sentence, but the chunk of text returned (and added in the context of the LLM) actually contains also surrounding sentences to add more context to that embedded sentence. This tends to give a better vector similarity than the whole surrounding context. It is one of the techniques I’m covering in my talk on advanced RAG techniques.

In a nutshell, the AI agent equation is the following:

AI Agent = LLM + Memory + Planning + Tool Use

AI agents are nothing without tools! And they are actually more than mere Large Language Model calls. They require some memory management to handle the context of the interactions (short term, long term, or contextual information like in the Retrieval Augmented Generation approach. Planning is important (with variations around the Chain-of-Thought prompting approach, and LLM with reasoning or thinking capabilities) for an agent to realize its tasks.

As I’m contributing to ADK (Agent Development Kit) for Java, and LangChain4j (the LLM orchestration framework) I interact with MCP (Model Context Protocol) servers and tools to further expand the capabilities of my LLMs.

Recently, I showed how to vibe-code an MCP server using Micronaut. You know I usually talk about Micronaut, but this time, I wanted to experiment with Quarkus, and in particular with its built-in support for implementing MCP servers.

Getting started with Quarkus’ MCP support

I created a brand new Quarkus project from IntelliJ IDEA, with its Quarkus template, and I added a couple key dependencies for JSON marshalling, but even more important, for the MCP support:

Recently on these pages, I’ve covered ADK (Agent Development Kit) for Java, launched at Google I/O 2025. I showed how to get started writing your first Java agent, and I shared a Github template that you can use to kick start your development.

But you also know that I’m a big fan of, and a contributor to the LangChain4j project, where I’ve worked on the Gemini support, embedding models, GCS document loaders, Imagen generation, etc.

With the unveiling of the Java version of Agent Development Kit (ADK) which lets you build AI agents in Java, I recently covered how to get started developing your first agent.

The installation and quickstart documentation also helps for the first steps, but I realized that it would be handy to provide a template project, to further accelarate your time-to-first-conversation with your Java agents! This led me to play with GitHub’s template project feature, which allows you to create a copy of the template project on your own account or organization. It comes with a ready-made project structure, a configured pom.xml file, and a first Java agent you can customize at will, and run from both the command-line or the ADK Dev UI.

When I talk about Generative AI, whether it’s with developers at conferences or with customers, I often find myself saying the same thing: chatbots are just one way to use Large Language Models (LLMs).

Unfortunately, I see many articles or presentations that just focus on demonstrating LLMs at work within the context of chatbots. I feel guilty of showing the traditional chat interfaces too. But there’s so much more to it!