Roadmap
14 stages, grouped into 4 phases. Follow them in order. This roadmap aligns with the roadmap.sh AI Agents roadmap.
Phase 1 Foundations Prerequisites, LLMs, prompts.
Phase 2 Agent Core Loops, tools, MCP, memory.
Phase 3 Systems Multi-agent, eval, security.
Phase 4 Production Deploy, monitor, improve.
Phase 1 — Foundations
Get the basics right before building agents.
- Stage 0 - Orientation — what AI agents are and how to use this site
- Stage 1 - Prerequisites — backend, APIs, Git, terminal, JSON, streaming
- Stage 2 - LLM Fundamentals — tokens, context, model choice, cost, controls
- Stage 3 - Prompt Engineering — clear instructions, examples, and prompt testing
Phase 2 — Agent Core
Build agents that reason and use tools.
- Stage 4 - Agent Fundamentals — the agent loop, ReAct, planning, stop rules
- Stage 5 - Tools and Actions — tool design, schemas, function calling
- Stage 6 - MCP — connect tools with the Model Context Protocol
- Stage 7 - RAG and Memory — retrieval, embeddings, short- and long-term memory
- Stage 8 - Agent Architectures — ReAct, RAG, planner-executor, routing
- Stage 9 - Building Agents — direct APIs first, then frameworks
Phase 3 — Systems
Coordinate, measure, and secure agents.
- Stage 10 - Multi-Agent Systems — supervisor-worker, handoffs, A2A
- Stage 11 - Evaluation and Observability — tests, metrics, tracing
- Stage 12 - Security and Ethics — prompt injection, permissions, red teaming
Phase 4 — Production
Ship and operate real systems.
- Stage 13 - Production Deployment — APIs, Docker, CI/CD, monitoring, cost
Completion Standard
Not enough
I read articles about AI agents.
Done well
I built a tool-using agent, tested it with 20 examples, measured cost and latency, fixed two failures, and wrote down the lessons.
See the full topic mapping in Roadmap.sh Alignment.