Use this kit to run the Agent Engineering Roadmap as a study group, internal training, or community workshop.
| Format | Best for | Commitment |
|---|
| 4-week sprint | Experienced engineers | 2 sessions per week |
| 8-week cohort | Mixed skill levels | 1 session per week |
| 1-day workshop | Team kickoff | 4 focused blocks |
| Week | Focus | Run | Deliverable |
|---|
| 1 | Foundations, tools, MCP | Examples 01-03 | Agent spec and tool contract |
| 2 | Memory, RAG, workflow | Examples 04, 05, 08 | RAG eval and workflow sketch |
| 3 | Graphs, multi-agent, approval | Examples 06, 09 | Approval graph and routing tests |
| 4 | Production, observability, security | Examples 10, 11, benchmark | Trace review and injection defense |
| Week | Topic | Discussion Prompt |
|---|
| 1 | What is an agent? | Where does a chatbot stop being enough? |
| 2 | Tool use and MCP | Which tools should never be callable without approval? |
| 3 | Memory | What should an agent forget? |
| 4 | RAG | How do we know an answer is grounded? |
| 5 | Workflow graphs | Which states should be explicit? |
| 6 | Multi-agent systems | When does adding agents make quality worse? |
| 7 | Observability | Which trace would explain a bad answer? |
| 8 | Security and capstone | What attack would break your design? |
- Open with one failure story.
- Draw the mental model.
- Run the smallest example.
- Change one thing and predict the result.
- Run the benchmark or eval.
- Record one improvement as an issue.