Java-first · Open source · One concept at a time
Agentic Developer Handbook
Understand agentic systems by building them.
From a single model call to a bounded agent runtime, one concept at a time, with an honest answer to "When should I not use this?"
Not an agent framework. Not a library catalog. Not a skill marketplace.
The learning path
Eight runnable Java labs. Each adds one capability — and says when not to use it.
Planned Evaluation · Observability · Security · Deployment — on the roadmap, not implemented yet.
Start here — three ways to use it
01 Learn
8 runnable Java labs
Start from a single model call and move step by step toward a bounded agent runtime. The labs, the Java code, and the documentation live in the repository.
02 Compose
Capstone 01
Bring together only the capabilities a realistic scenario actually needs, build it, and compare with one reference design that deliberately leaves capabilities out.
03 Decide
Readiness Assessment
Work through your own use case and decide whether it needs an LLM,
RAG, tools, memory, or an agent at all. Not a score — it exports an
ASSESSMENT.md.
You probably don't need all of these.
The path is for learning, not a mandatory architecture: stop when the problem is solved. A single LLM call is not an agent. A database lookup does not become RAG because the application also calls a model. A local Java method does not need MCP.
Open source · Apache License 2.0
The repository is the source of truth.
- The Java labs, the docs, and the roadmap live on GitHub — this site is an interactive layer over them.
- Issues, corrections, learning feedback, and proposals happen there.
- Your answers on this site stay in your browser: no backend, no accounts, no analytics, no cookies.
# clone and verify — no API key needed
git clone \
https://github.com/cagridursun/agentic-developer-handbook.git
cd agentic-developer-handbook
./mvnw verify